id	sid	tid	token	lemma	pos
ejpam-6638	1	1	european	european	PROPN
ejpam-6638	1	2	journal	journal	PROPN
ejpam-6638	1	3	of	of	ADP
ejpam-6638	1	4	pure	pure	ADJ
ejpam-6638	1	5	and	and	CCONJ
ejpam-6638	1	6	applied	applied	ADJ
ejpam-6638	1	7	mathematics	mathematic	NOUN
ejpam-6638	1	8	2025	2025	NUM
ejpam-6638	1	9	,	,	PUNCT
ejpam-6638	1	10	vol	vol	NOUN
ejpam-6638	1	11	.	.	PROPN
ejpam-6638	1	12	18	18	NUM
ejpam-6638	1	13	,	,	PUNCT
ejpam-6638	1	14	issue	issue	NOUN
ejpam-6638	1	15	4	4	NUM
ejpam-6638	1	16	,	,	PUNCT
ejpam-6638	1	17	article	article	NOUN
ejpam-6638	1	18	number	number	NOUN
ejpam-6638	1	19	6638	6638	NUM
ejpam-6638	1	20	issn	issn	VERB
ejpam-6638	1	21	1307	1307	NUM
ejpam-6638	1	22	-	-	SYM
ejpam-6638	1	23	5543	5543	NUM
ejpam-6638	1	24	–	–	PUNCT
ejpam-6638	1	25	ejpam.com	ejpam.com	X
ejpam-6638	1	26	published	publish	VERB
ejpam-6638	1	27	by	by	ADP
ejpam-6638	1	28	new	new	PROPN
ejpam-6638	1	29	york	york	PROPN
ejpam-6638	1	30	business	business	PROPN
ejpam-6638	1	31	global	global	PROPN
ejpam-6638	1	32	enhancing	enhance	VERB
ejpam-6638	1	33	deep	deep	ADJ
ejpam-6638	1	34	image	image	NOUN
ejpam-6638	1	35	prior	prior	ADV
ejpam-6638	1	36	with	with	ADP
ejpam-6638	1	37	multiscale	multiscale	ADJ
ejpam-6638	1	38	attention	attention	NOUN
ejpam-6638	1	39	,	,	PUNCT
ejpam-6638	1	40	svd	svd	PROPN
ejpam-6638	1	41	pooling	pool	VERB
ejpam-6638	1	42	,	,	PUNCT
ejpam-6638	1	43	and	and	CCONJ
ejpam-6638	1	44	preconditioned	precondition	VERB
ejpam-6638	1	45	optimizers	optimizer	NOUN
ejpam-6638	1	46	for	for	ADP
ejpam-6638	1	47	image	image	NOUN
ejpam-6638	1	48	processing	processing	NOUN
ejpam-6638	1	49	muhammad	muhammad	PROPN
ejpam-6638	1	50	israr1	israr1	PROPN
ejpam-6638	1	51	,	,	PUNCT
ejpam-6638	1	52	waqar	waqar	PROPN
ejpam-6638	1	53	afzal1	afzal1	PROPN
ejpam-6638	1	54	,	,	PUNCT
ejpam-6638	1	55	shahbaz	shahbaz	PROPN
ejpam-6638	1	56	ahmad1	ahmad1	PROPN
ejpam-6638	1	57	,	,	PUNCT
ejpam-6638	1	58	daniel	daniel	PROPN
ejpam-6638	1	59	breaz2,∗	breaz2,∗	PROPN
ejpam-6638	1	60	,	,	PUNCT
ejpam-6638	1	61	luminiţa	luminiţa	PROPN
ejpam-6638	1	62	-	-	PUNCT
ejpam-6638	1	63	ioana	ioana	PROPN
ejpam-6638	1	64	cotîrlă3	cotîrlă3	NOUN
ejpam-6638	1	65	1	1	NUM
ejpam-6638	1	66	abdus	abdus	NOUN
ejpam-6638	1	67	salam	salam	PROPN
ejpam-6638	1	68	school	school	PROPN
ejpam-6638	1	69	of	of	ADP
ejpam-6638	1	70	mathematical	mathematical	ADJ
ejpam-6638	1	71	sciences	science	NOUN
ejpam-6638	1	72	,	,	PUNCT
ejpam-6638	1	73	government	government	NOUN
ejpam-6638	1	74	college	college	NOUN
ejpam-6638	1	75	university	university	NOUN
ejpam-6638	1	76	,	,	PUNCT
ejpam-6638	1	77	68	68	NUM
ejpam-6638	1	78	-	-	SYM
ejpam-6638	1	79	b	b	NOUN
ejpam-6638	1	80	,	,	PUNCT
ejpam-6638	1	81	new	new	ADJ
ejpam-6638	1	82	muslim	muslim	ADJ
ejpam-6638	1	83	town	town	NOUN
ejpam-6638	1	84	,	,	PUNCT
ejpam-6638	1	85	lahore	lahore	PROPN
ejpam-6638	1	86	54600	54600	NUM
ejpam-6638	1	87	,	,	PUNCT
ejpam-6638	1	88	pakistan	pakistan	PROPN
ejpam-6638	1	89	2	2	NUM
ejpam-6638	1	90	department	department	NOUN
ejpam-6638	1	91	of	of	ADP
ejpam-6638	1	92	mathematics	mathematic	NOUN
ejpam-6638	1	93	,	,	PUNCT
ejpam-6638	1	94	“	"	PUNCT
ejpam-6638	1	95	1	1	NUM
ejpam-6638	1	96	decembrie	decembrie	NOUN
ejpam-6638	1	97	1918	1918	NUM
ejpam-6638	1	98	”	"	PUNCT
ejpam-6638	1	99	university	university	PROPN
ejpam-6638	1	100	of	of	ADP
ejpam-6638	1	101	alba	alba	PROPN
ejpam-6638	1	102	iulia	iulia	PROPN
ejpam-6638	1	103	,	,	PUNCT
ejpam-6638	1	104	510009	510009	NUM
ejpam-6638	1	105	alba	alba	PROPN
ejpam-6638	1	106	iulia	iulia	PROPN
ejpam-6638	1	107	,	,	PUNCT
ejpam-6638	1	108	romania	romania	PROPN
ejpam-6638	1	109	3	3	NUM
ejpam-6638	1	110	department	department	NOUN
ejpam-6638	1	111	of	of	ADP
ejpam-6638	1	112	mathematics	mathematic	NOUN
ejpam-6638	1	113	,	,	PUNCT
ejpam-6638	1	114	technical	technical	ADJ
ejpam-6638	1	115	university	university	PROPN
ejpam-6638	1	116	of	of	ADP
ejpam-6638	1	117	cluj	cluj	PROPN
ejpam-6638	1	118	-	-	PUNCT
ejpam-6638	1	119	napoca	napoca	NOUN
ejpam-6638	1	120	,	,	PUNCT
ejpam-6638	1	121	400114	400114	NUM
ejpam-6638	1	122	cluj	cluj	PROPN
ejpam-6638	1	123	-	-	PUNCT
ejpam-6638	1	124	napoca	napoca	PROPN
ejpam-6638	1	125	,	,	PUNCT
ejpam-6638	1	126	romania	romania	PROPN
ejpam-6638	1	127	abstract	abstract	NOUN
ejpam-6638	1	128	.	.	PUNCT
ejpam-6638	2	1	deep	deep	ADJ
ejpam-6638	2	2	image	image	NOUN
ejpam-6638	2	3	prior	prior	ADV
ejpam-6638	2	4	(	(	PUNCT
ejpam-6638	2	5	dip	dip	NOUN
ejpam-6638	2	6	)	)	PUNCT
ejpam-6638	2	7	is	be	AUX
ejpam-6638	2	8	effective	effective	ADJ
ejpam-6638	2	9	in	in	ADP
ejpam-6638	2	10	recovering	recover	VERB
ejpam-6638	2	11	high	high	ADJ
ejpam-6638	2	12	-	-	PUNCT
ejpam-6638	2	13	quality	quality	NOUN
ejpam-6638	2	14	images	image	NOUN
ejpam-6638	2	15	in	in	ADP
ejpam-6638	2	16	inverse	inverse	NOUN
ejpam-6638	2	17	problems	problem	NOUN
ejpam-6638	2	18	such	such	ADJ
ejpam-6638	2	19	as	as	ADP
ejpam-6638	2	20	restoration	restoration	NOUN
ejpam-6638	2	21	and	and	CCONJ
ejpam-6638	2	22	mri	mri	NOUN
ejpam-6638	2	23	,	,	PUNCT
ejpam-6638	2	24	but	but	CCONJ
ejpam-6638	2	25	it	it	PRON
ejpam-6638	2	26	often	often	ADV
ejpam-6638	2	27	suffers	suffer	VERB
ejpam-6638	2	28	from	from	ADP
ejpam-6638	2	29	overfitting	overfitte	VERB
ejpam-6638	2	30	and	and	CCONJ
ejpam-6638	2	31	spectral	spectral	ADJ
ejpam-6638	2	32	bias	bias	NOUN
ejpam-6638	2	33	.	.	PUNCT
ejpam-6638	3	1	to	to	PART
ejpam-6638	3	2	address	address	VERB
ejpam-6638	3	3	these	these	DET
ejpam-6638	3	4	challenges	challenge	NOUN
ejpam-6638	3	5	,	,	PUNCT
ejpam-6638	3	6	we	we	PRON
ejpam-6638	3	7	introduce	introduce	VERB
ejpam-6638	3	8	a	a	DET
ejpam-6638	3	9	novel	novel	ADJ
ejpam-6638	3	10	dip	dip	NOUN
ejpam-6638	3	11	-	-	PUNCT
ejpam-6638	3	12	based	base	VERB
ejpam-6638	3	13	architecture	architecture	NOUN
ejpam-6638	3	14	that	that	PRON
ejpam-6638	3	15	integrates	integrate	NOUN
ejpam-6638	3	16	enhanced	enhance	VERB
ejpam-6638	3	17	svd	svd	PROPN
ejpam-6638	3	18	pooling	pool	VERB
ejpam-6638	3	19	to	to	PART
ejpam-6638	3	20	preserve	preserve	VERB
ejpam-6638	3	21	global	global	ADJ
ejpam-6638	3	22	contextual	contextual	ADJ
ejpam-6638	3	23	features	feature	NOUN
ejpam-6638	3	24	,	,	PUNCT
ejpam-6638	3	25	softplus	softplus	ADJ
ejpam-6638	3	26	activation	activation	NOUN
ejpam-6638	3	27	to	to	PART
ejpam-6638	3	28	ensure	ensure	VERB
ejpam-6638	3	29	smooth	smooth	ADJ
ejpam-6638	3	30	and	and	CCONJ
ejpam-6638	3	31	stable	stable	ADJ
ejpam-6638	3	32	gradient	gradient	NOUN
ejpam-6638	3	33	propagation	propagation	NOUN
ejpam-6638	3	34	,	,	PUNCT
ejpam-6638	3	35	and	and	CCONJ
ejpam-6638	3	36	multiscale	multiscale	ADJ
ejpam-6638	3	37	attention	attention	NOUN
ejpam-6638	3	38	mechanisms	mechanism	NOUN
ejpam-6638	3	39	with	with	ADP
ejpam-6638	3	40	dilated	dilated	ADJ
ejpam-6638	3	41	convolutions	convolution	NOUN
ejpam-6638	3	42	to	to	PART
ejpam-6638	3	43	effectively	effectively	ADV
ejpam-6638	3	44	recover	recover	VERB
ejpam-6638	3	45	fine	fine	ADJ
ejpam-6638	3	46	image	image	NOUN
ejpam-6638	3	47	details	detail	NOUN
ejpam-6638	3	48	.	.	PUNCT
ejpam-6638	4	1	to	to	PART
ejpam-6638	4	2	further	far	ADV
ejpam-6638	4	3	accelerate	accelerate	VERB
ejpam-6638	4	4	training	training	NOUN
ejpam-6638	4	5	and	and	CCONJ
ejpam-6638	4	6	enhance	enhance	VERB
ejpam-6638	4	7	generalization	generalization	NOUN
ejpam-6638	4	8	,	,	PUNCT
ejpam-6638	4	9	we	we	PRON
ejpam-6638	4	10	adopt	adopt	VERB
ejpam-6638	4	11	advanced	advanced	ADJ
ejpam-6638	4	12	preconditioned	precondition	VERB
ejpam-6638	4	13	optimizers	optimizer	NOUN
ejpam-6638	4	14	such	such	ADJ
ejpam-6638	4	15	as	as	ADP
ejpam-6638	4	16	diffgrad	diffgrad	PROPN
ejpam-6638	4	17	and	and	CCONJ
ejpam-6638	4	18	apollo	apollo	PROPN
ejpam-6638	4	19	.	.	PUNCT
ejpam-6638	5	1	our	our	PRON
ejpam-6638	5	2	data	data	NOUN
ejpam-6638	5	3	-	-	PUNCT
ejpam-6638	5	4	independent	independent	ADJ
ejpam-6638	5	5	restoration	restoration	NOUN
ejpam-6638	5	6	strategy	strategy	NOUN
ejpam-6638	5	7	is	be	AUX
ejpam-6638	5	8	thoroughly	thoroughly	ADV
ejpam-6638	5	9	evaluated	evaluate	VERB
ejpam-6638	5	10	on	on	ADP
ejpam-6638	5	11	image	image	NOUN
ejpam-6638	5	12	denoising	denoise	VERB
ejpam-6638	5	13	and	and	CCONJ
ejpam-6638	5	14	inpainting	inpainte	VERB
ejpam-6638	5	15	benchmarks	benchmark	NOUN
ejpam-6638	5	16	,	,	PUNCT
ejpam-6638	5	17	yielding	yield	VERB
ejpam-6638	5	18	notable	notable	ADJ
ejpam-6638	5	19	improvements	improvement	NOUN
ejpam-6638	5	20	in	in	ADP
ejpam-6638	5	21	mse	mse	PROPN
ejpam-6638	5	22	,	,	PUNCT
ejpam-6638	5	23	psnr	psnr	NOUN
ejpam-6638	5	24	,	,	PUNCT
ejpam-6638	5	25	and	and	CCONJ
ejpam-6638	5	26	ssim	ssim	NOUN
ejpam-6638	5	27	.	.	PUNCT
ejpam-6638	6	1	we	we	PRON
ejpam-6638	6	2	supplement	supplement	VERB
ejpam-6638	6	3	our	our	PRON
ejpam-6638	6	4	quantitative	quantitative	ADJ
ejpam-6638	6	5	results	result	NOUN
ejpam-6638	6	6	with	with	ADP
ejpam-6638	6	7	rich	rich	ADJ
ejpam-6638	6	8	visualizations	visualization	NOUN
ejpam-6638	6	9	,	,	PUNCT
ejpam-6638	6	10	including	include	VERB
ejpam-6638	6	11	side	side	NOUN
ejpam-6638	6	12	-	-	PUNCT
ejpam-6638	6	13	by	by	ADP
ejpam-6638	6	14	-	-	PUNCT
ejpam-6638	6	15	side	side	NOUN
ejpam-6638	6	16	restoration	restoration	NOUN
ejpam-6638	6	17	outputs	output	NOUN
ejpam-6638	6	18	,	,	PUNCT
ejpam-6638	6	19	ssim	ssim	NOUN
ejpam-6638	6	20	/	/	SYM
ejpam-6638	6	21	psnr	psnr	NOUN
ejpam-6638	6	22	trends	trend	NOUN
ejpam-6638	6	23	over	over	ADP
ejpam-6638	6	24	epochs	epoch	NOUN
ejpam-6638	6	25	,	,	PUNCT
ejpam-6638	6	26	3d	3d	NUM
ejpam-6638	6	27	visualizations	visualization	NOUN
ejpam-6638	6	28	of	of	ADP
ejpam-6638	6	29	dominant	dominant	ADJ
ejpam-6638	6	30	neuron	neuron	NOUN
ejpam-6638	6	31	activations	activation	NOUN
ejpam-6638	6	32	,	,	PUNCT
ejpam-6638	6	33	and	and	CCONJ
ejpam-6638	6	34	comparative	comparative	ADJ
ejpam-6638	6	35	3d	3d	NUM
ejpam-6638	6	36	bar	bar	NOUN
ejpam-6638	6	37	plots	plot	NOUN
ejpam-6638	6	38	illustrating	illustrate	VERB
ejpam-6638	6	39	optimizer	optimizer	NOUN
ejpam-6638	6	40	performance	performance	NOUN
ejpam-6638	6	41	.	.	PUNCT
ejpam-6638	7	1	the	the	DET
ejpam-6638	7	2	source	source	NOUN
ejpam-6638	7	3	code	code	NOUN
ejpam-6638	7	4	for	for	ADP
ejpam-6638	7	5	all	all	DET
ejpam-6638	7	6	the	the	DET
ejpam-6638	7	7	simulations	simulation	NOUN
ejpam-6638	7	8	conducted	conduct	VERB
ejpam-6638	7	9	in	in	ADP
ejpam-6638	7	10	this	this	DET
ejpam-6638	7	11	paper	paper	NOUN
ejpam-6638	7	12	is	be	AUX
ejpam-6638	7	13	available	available	ADJ
ejpam-6638	7	14	at	at	ADP
ejpam-6638	7	15	the	the	DET
ejpam-6638	7	16	following	follow	VERB
ejpam-6638	7	17	link	link	NOUN
ejpam-6638	7	18	:	:	PUNCT
ejpam-6638	7	19	https://github.com/israrjamali/dip-svd-code-for-image-denoising-and-inpainting	https://github.com/israrjamali/dip-svd-code-for-image-denoising-and-inpainte	VERB
ejpam-6638	7	20	.	.	PUNCT
ejpam-6638	8	1	2020	2020	NUM
ejpam-6638	9	1	mathematics	mathematic	NOUN
ejpam-6638	9	2	subject	subject	NOUN
ejpam-6638	9	3	classifications	classification	NOUN
ejpam-6638	9	4	:	:	PUNCT
ejpam-6638	9	5	68u10	68u10	NUM
ejpam-6638	9	6	,	,	PUNCT
ejpam-6638	9	7	65k10	65k10	NUM
ejpam-6638	9	8	,	,	PUNCT
ejpam-6638	9	9	68t07	68t07	NUM
ejpam-6638	9	10	key	key	ADJ
ejpam-6638	9	11	words	word	NOUN
ejpam-6638	9	12	and	and	CCONJ
ejpam-6638	9	13	phrases	phrase	NOUN
ejpam-6638	9	14	:	:	PUNCT
ejpam-6638	9	15	deep	deep	ADJ
ejpam-6638	9	16	image	image	NOUN
ejpam-6638	9	17	prior	prior	ADV
ejpam-6638	9	18	,	,	PUNCT
ejpam-6638	9	19	untrained	untrained	ADJ
ejpam-6638	9	20	neural	neural	ADJ
ejpam-6638	9	21	networks	network	NOUN
ejpam-6638	9	22	,	,	PUNCT
ejpam-6638	9	23	image	image	NOUN
ejpam-6638	9	24	processing	processing	NOUN
ejpam-6638	9	25	,	,	PUNCT
ejpam-6638	9	26	image	image	NOUN
ejpam-6638	9	27	inpainting	inpainting	NOUN
ejpam-6638	9	28	,	,	PUNCT
ejpam-6638	9	29	svd	svd	PROPN
ejpam-6638	9	30	pooling	pool	VERB
ejpam-6638	9	31	,	,	PUNCT
ejpam-6638	9	32	multiscale	multiscale	ADJ
ejpam-6638	9	33	attention	attention	NOUN
ejpam-6638	9	34	mechanisms	mechanism	NOUN
ejpam-6638	9	35	,	,	PUNCT
ejpam-6638	9	36	softplus	softplus	ADJ
ejpam-6638	9	37	activation	activation	NOUN
ejpam-6638	9	38	,	,	PUNCT
ejpam-6638	9	39	preconditioned	preconditioned	ADJ
ejpam-6638	9	40	optimizers	optimizer	NOUN
ejpam-6638	9	41	,	,	PUNCT
ejpam-6638	9	42	diffgrad	diffgrad	PROPN
ejpam-6638	9	43	,	,	PUNCT
ejpam-6638	9	44	apollo	apollo	PROPN
ejpam-6638	9	45	optimizer	optimizer	NOUN
ejpam-6638	9	46	,	,	PUNCT
ejpam-6638	9	47	deep	deep	ADJ
ejpam-6638	9	48	learning	learning	NOUN
ejpam-6638	9	49	for	for	ADP
ejpam-6638	9	50	inverse	inverse	NOUN
ejpam-6638	9	51	problems	problem	NOUN
ejpam-6638	9	52	1	1	NUM
ejpam-6638	9	53	.	.	PUNCT
ejpam-6638	10	1	introduction	introduction	NOUN
ejpam-6638	10	2	image	image	NOUN
ejpam-6638	10	3	restoration	restoration	NOUN
ejpam-6638	10	4	is	be	AUX
ejpam-6638	10	5	a	a	DET
ejpam-6638	10	6	fundamental	fundamental	ADJ
ejpam-6638	10	7	problem	problem	NOUN
ejpam-6638	10	8	in	in	ADP
ejpam-6638	10	9	computer	computer	NOUN
ejpam-6638	10	10	vision	vision	NOUN
ejpam-6638	10	11	and	and	CCONJ
ejpam-6638	10	12	signal	signal	NOUN
ejpam-6638	10	13	processing	processing	NOUN
ejpam-6638	10	14	which	which	PRON
ejpam-6638	10	15	attempts	attempt	VERB
ejpam-6638	10	16	to	to	PART
ejpam-6638	10	17	recover	recover	VERB
ejpam-6638	10	18	some	some	DET
ejpam-6638	10	19	action	action	NOUN
ejpam-6638	10	20	such	such	ADJ
ejpam-6638	10	21	as	as	ADP
ejpam-6638	10	22	denoising	denoise	VERB
ejpam-6638	10	23	,	,	PUNCT
ejpam-6638	10	24	super	super	NOUN
ejpam-6638	10	25	-	-	NOUN
ejpam-6638	10	26	resolution	resolution	NOUN
ejpam-6638	10	27	,	,	PUNCT
ejpam-6638	10	28	inpainting	inpainting	NOUN
ejpam-6638	10	29	,	,	PUNCT
ejpam-6638	10	30	and	and	CCONJ
ejpam-6638	10	31	deblurring	deblurring	NOUN
ejpam-6638	10	32	.	.	PUNCT
ejpam-6638	11	1	its	its	PRON
ejpam-6638	11	2	aim	aim	NOUN
ejpam-6638	11	3	is	be	AUX
ejpam-6638	11	4	to	to	PART
ejpam-6638	11	5	recover	recover	VERB
ejpam-6638	11	6	a	a	DET
ejpam-6638	11	7	high	high	ADJ
ejpam-6638	11	8	-	-	PUNCT
ejpam-6638	11	9	quality	quality	NOUN
ejpam-6638	11	10	image	image	NOUN
ejpam-6638	11	11	based	base	VERB
ejpam-6638	11	12	on	on	ADP
ejpam-6638	11	13	its	its	PRON
ejpam-6638	11	14	degraded	degraded	ADJ
ejpam-6638	11	15	observation	observation	NOUN
ejpam-6638	11	16	;	;	PUNCT
ejpam-6638	11	17	the	the	DET
ejpam-6638	11	18	process	process	NOUN
ejpam-6638	11	19	is	be	AUX
ejpam-6638	11	20	necessary	necessary	ADJ
ejpam-6638	11	21	to	to	ADP
ejpam-6638	11	22	many	many	ADJ
ejpam-6638	11	23	applications	application	NOUN
ejpam-6638	11	24	,	,	PUNCT
ejpam-6638	11	25	such	such	ADJ
ejpam-6638	11	26	as	as	ADP
ejpam-6638	11	27	satellite	satellite	NOUN
ejpam-6638	11	28	photography	photography	NOUN
ejpam-6638	11	29	and	and	CCONJ
ejpam-6638	11	30	medical	medical	ADJ
ejpam-6638	11	31	imaging	imaging	NOUN
ejpam-6638	11	32	.	.	PUNCT
ejpam-6638	12	1	the	the	DET
ejpam-6638	12	2	complexity	complexity	NOUN
ejpam-6638	12	3	of	of	ADP
ejpam-6638	12	4	degradation	degradation	NOUN
ejpam-6638	12	5	process	process	NOUN
ejpam-6638	12	6	in	in	ADP
ejpam-6638	12	7	a	a	DET
ejpam-6638	12	8	real	real	ADJ
ejpam-6638	12	9	-	-	PUNCT
ejpam-6638	12	10	world	world	NOUN
ejpam-6638	12	11	setting	setting	NOUN
ejpam-6638	12	12	tended	tend	VERB
ejpam-6638	12	13	to	to	PART
ejpam-6638	12	14	truncate	truncate	VERB
ejpam-6638	12	15	the	the	DET
ejpam-6638	12	16	effectiveness	effectiveness	NOUN
ejpam-6638	12	17	of	of	ADP
ejpam-6638	12	18	conventional	conventional	ADJ
ejpam-6638	12	19	solutions	solution	NOUN
ejpam-6638	12	20	which	which	PRON
ejpam-6638	12	21	often	often	ADV
ejpam-6638	12	22	required	require	VERB
ejpam-6638	12	23	the	the	DET
ejpam-6638	12	24	construction	construction	NOUN
ejpam-6638	12	25	of	of	ADP
ejpam-6638	12	26	hand	hand	NOUN
ejpam-6638	12	27	-	-	PUNCT
ejpam-6638	12	28	specific	specific	ADJ
ejpam-6638	12	29	priors	prior	NOUN
ejpam-6638	12	30	and	and	CCONJ
ejpam-6638	12	31	multifaceted	multifaceted	ADJ
ejpam-6638	12	32	mathematical	mathematical	ADJ
ejpam-6638	12	33	formulae	formulae	NOUN
ejpam-6638	12	34	.	.	PUNCT
ejpam-6638	13	1	image	image	NOUN
ejpam-6638	13	2	enhancement	enhancement	NOUN
ejpam-6638	13	3	is	be	AUX
ejpam-6638	13	4	essential	essential	ADJ
ejpam-6638	13	5	for	for	ADP
ejpam-6638	13	6	many	many	ADJ
ejpam-6638	13	7	applications	application	NOUN
ejpam-6638	13	8	,	,	PUNCT
ejpam-6638	13	9	including	include	VERB
ejpam-6638	13	10	photography	photography	NOUN
ejpam-6638	13	11	,	,	PUNCT
ejpam-6638	13	12	remote	remote	ADJ
ejpam-6638	13	13	sensing	sensing	NOUN
ejpam-6638	13	14	,	,	PUNCT
ejpam-6638	13	15	and	and	CCONJ
ejpam-6638	13	16	medical	medical	ADJ
ejpam-6638	13	17	imaging	imaging	NOUN
ejpam-6638	13	18	.	.	PUNCT
ejpam-6638	14	1	its	its	PRON
ejpam-6638	14	2	goal	goal	NOUN
ejpam-6638	14	3	is	be	AUX
ejpam-6638	14	4	to	to	PART
ejpam-6638	14	5	recover	recover	VERB
ejpam-6638	14	6	high	high	ADJ
ejpam-6638	14	7	-	-	PUNCT
ejpam-6638	14	8	quality	quality	NOUN
ejpam-6638	14	9	images	image	NOUN
ejpam-6638	14	10	from	from	ADP
ejpam-6638	14	11	damaged	damage	VERB
ejpam-6638	14	12	or	or	CCONJ
ejpam-6638	14	13	corrupted	corrupted	ADJ
ejpam-6638	14	14	∗corresponding	∗corresponde	VERB
ejpam-6638	14	15	author	author	NOUN
ejpam-6638	14	16	.	.	PUNCT
ejpam-6638	15	1	doi	doi	NOUN
ejpam-6638	15	2	:	:	PUNCT
ejpam-6638	15	3	https://doi.org/10.29020/nybg.ejpam.v18i4.6638	https://doi.org/10.29020/nybg.ejpam.v18i4.6638	ADJ
ejpam-6638	15	4	email	email	NOUN
ejpam-6638	15	5	addresses	address	NOUN
ejpam-6638	15	6	:	:	PUNCT
ejpam-6638	16	1	israrjamali@gmail.com	israrjamali@gmail.com	PROPN
ejpam-6638	16	2	(	(	PUNCT
ejpam-6638	16	3	m.	m.	NOUN
ejpam-6638	16	4	israr	israr	PROPN
ejpam-6638	16	5	)	)	PUNCT
ejpam-6638	16	6	,	,	PUNCT
ejpam-6638	16	7	wafzal.math@gmail.com	wafzal.math@gmail.com	X
ejpam-6638	16	8	(	(	PUNCT
ejpam-6638	16	9	w.	w.	PROPN
ejpam-6638	16	10	afzal	afzal	PROPN
ejpam-6638	16	11	)	)	PUNCT
ejpam-6638	16	12	,	,	PUNCT
ejpam-6638	16	13	shahbaz@gcu.edu.pk	shahbaz@gcu.edu.pk	NOUN
ejpam-6638	16	14	(	(	PUNCT
ejpam-6638	16	15	s.	s.	PROPN
ejpam-6638	16	16	ahmad	ahmad	PROPN
ejpam-6638	16	17	)	)	PUNCT
ejpam-6638	16	18	,	,	PUNCT
ejpam-6638	16	19	dbreaz@uab.ro	dbreaz@uab.ro	PROPN
ejpam-6638	16	20	(	(	PUNCT
ejpam-6638	16	21	d.	d.	NOUN
ejpam-6638	16	22	breaz	breaz	PROPN
ejpam-6638	16	23	)	)	PUNCT
ejpam-6638	16	24	,	,	PUNCT
ejpam-6638	16	25	luminita.cotirla@math.utcluj.ro	luminita.cotirla@math.utcluj.ro	NOUN
ejpam-6638	16	26	(	(	PUNCT
ejpam-6638	16	27	l.-i	l.-i	PROPN
ejpam-6638	16	28	.	.	PUNCT
ejpam-6638	16	29	cotîrlă	cotîrlă	PROPN
ejpam-6638	16	30	)	)	PUNCT
ejpam-6638	16	31	https://www.ejpam.com	https://www.ejpam.com	NOUN
ejpam-6638	17	1	1	1	NUM
ejpam-6638	17	2	copyright	copyright	NOUN
ejpam-6638	17	3	:	:	PUNCT
ejpam-6638	17	4	©	©	PROPN
ejpam-6638	17	5	2025	2025	NUM
ejpam-6638	17	6	the	the	DET
ejpam-6638	17	7	author(s	author(s	NOUN
ejpam-6638	17	8	)	)	PUNCT
ejpam-6638	17	9	.	.	PUNCT
ejpam-6638	18	1	(	(	PUNCT
ejpam-6638	18	2	cc	cc	NOUN
ejpam-6638	18	3	by	by	ADP
ejpam-6638	18	4	-	-	PUNCT
ejpam-6638	18	5	nc	nc	PROPN
ejpam-6638	18	6	4.0	4.0	NUM
ejpam-6638	18	7	)	)	PUNCT
ejpam-6638	18	8	m.	m.	NOUN
ejpam-6638	18	9	israr	israr	NOUN
ejpam-6638	18	10	et	et	PROPN
ejpam-6638	18	11	al	al	PROPN
ejpam-6638	18	12	.	.	PUNCT
ejpam-6638	18	13	/	/	SYM
ejpam-6638	18	14	eur	eur	PROPN
ejpam-6638	18	15	.	.	PUNCT
ejpam-6638	19	1	j.	j.	PROPN
ejpam-6638	19	2	pure	pure	PROPN
ejpam-6638	19	3	appl	appl	PROPN
ejpam-6638	19	4	.	.	PROPN
ejpam-6638	19	5	math	math	PROPN
ejpam-6638	19	6	,	,	PUNCT
ejpam-6638	19	7	18	18	NUM
ejpam-6638	19	8	(	(	PUNCT
ejpam-6638	19	9	4	4	NUM
ejpam-6638	19	10	)	)	PUNCT
ejpam-6638	19	11	(	(	PUNCT
ejpam-6638	19	12	2025	2025	NUM
ejpam-6638	19	13	)	)	PUNCT
ejpam-6638	19	14	,	,	PUNCT
ejpam-6638	19	15	6638	6638	NUM
ejpam-6638	19	16	2	2	NUM
ejpam-6638	19	17	of	of	ADP
ejpam-6638	19	18	22	22	NUM
ejpam-6638	19	19	versions	version	NOUN
ejpam-6638	19	20	.	.	PUNCT
ejpam-6638	20	1	in	in	ADP
ejpam-6638	20	2	the	the	DET
ejpam-6638	20	3	past	past	NOUN
ejpam-6638	20	4	,	,	PUNCT
ejpam-6638	20	5	restoration	restoration	NOUN
ejpam-6638	20	6	techniques	technique	NOUN
ejpam-6638	20	7	used	use	VERB
ejpam-6638	20	8	variational	variational	ADJ
ejpam-6638	20	9	models	model	NOUN
ejpam-6638	20	10	and	and	CCONJ
ejpam-6638	20	11	manually	manually	ADV
ejpam-6638	20	12	created	create	VERB
ejpam-6638	20	13	priors	prior	NOUN
ejpam-6638	20	14	.	.	PUNCT
ejpam-6638	21	1	the	the	DET
ejpam-6638	21	2	total	total	ADJ
ejpam-6638	21	3	variation	variation	NOUN
ejpam-6638	21	4	(	(	PUNCT
ejpam-6638	21	5	tv	tv	NOUN
ejpam-6638	21	6	)	)	PUNCT
ejpam-6638	21	7	method	method	NOUN
ejpam-6638	21	8	of	of	ADP
ejpam-6638	21	9	rudin	rudin	PROPN
ejpam-6638	21	10	et	et	PROPN
ejpam-6638	21	11	al	al	PROPN
ejpam-6638	21	12	.	.	PUNCT
ejpam-6638	22	1	[	[	X
ejpam-6638	22	2	1	1	X
ejpam-6638	22	3	]	]	PUNCT
ejpam-6638	22	4	for	for	ADP
ejpam-6638	22	5	instance	instance	NOUN
ejpam-6638	22	6	,	,	PUNCT
ejpam-6638	22	7	enforced	enforce	VERB
ejpam-6638	22	8	piecewise	piecewise	NOUN
ejpam-6638	22	9	smoothness	smoothness	NOUN
ejpam-6638	22	10	in	in	ADP
ejpam-6638	22	11	the	the	DET
ejpam-6638	22	12	recovered	recover	VERB
ejpam-6638	22	13	image	image	NOUN
ejpam-6638	22	14	and	and	CCONJ
ejpam-6638	22	15	worked	work	VERB
ejpam-6638	22	16	well	well	ADV
ejpam-6638	22	17	for	for	ADP
ejpam-6638	22	18	denoising	denoise	VERB
ejpam-6638	22	19	and	and	CCONJ
ejpam-6638	22	20	deblurring	deblurring	ADJ
ejpam-6638	22	21	tasks	task	NOUN
ejpam-6638	22	22	.	.	PUNCT
ejpam-6638	23	1	similarly	similarly	ADV
ejpam-6638	23	2	,	,	PUNCT
ejpam-6638	23	3	for	for	ADP
ejpam-6638	23	4	improved	improved	ADJ
ejpam-6638	23	5	restoration	restoration	NOUN
ejpam-6638	23	6	quality	quality	NOUN
ejpam-6638	23	7	,	,	PUNCT
ejpam-6638	23	8	roth	roth	NOUN
ejpam-6638	23	9	and	and	CCONJ
ejpam-6638	23	10	black	black	NOUN
ejpam-6638	23	11	’s	’s	PART
ejpam-6638	23	12	field	field	NOUN
ejpam-6638	23	13	of	of	ADP
ejpam-6638	23	14	experts	expert	NOUN
ejpam-6638	23	15	model	model	NOUN
ejpam-6638	23	16	captured	capture	VERB
ejpam-6638	23	17	higherorder	higherorder	NOUN
ejpam-6638	23	18	statistics	statistic	NOUN
ejpam-6638	23	19	of	of	ADP
ejpam-6638	23	20	natural	natural	ADJ
ejpam-6638	23	21	images	image	NOUN
ejpam-6638	23	22	[	[	X
ejpam-6638	23	23	2	2	NUM
ejpam-6638	23	24	]	]	PUNCT
ejpam-6638	23	25	.	.	PUNCT
ejpam-6638	24	1	the	the	DET
ejpam-6638	24	2	non	non	ADJ
ejpam-6638	24	3	-	-	ADJ
ejpam-6638	24	4	local	local	ADJ
ejpam-6638	24	5	means	mean	NOUN
ejpam-6638	24	6	(	(	PUNCT
ejpam-6638	24	7	nlm	nlm	NOUN
ejpam-6638	24	8	)	)	PUNCT
ejpam-6638	24	9	method	method	NOUN
ejpam-6638	24	10	by	by	ADP
ejpam-6638	24	11	buades	buade	NOUN
ejpam-6638	24	12	et	et	PROPN
ejpam-6638	24	13	al	al	PROPN
ejpam-6638	24	14	.	.	PROPN
ejpam-6638	24	15	took	take	VERB
ejpam-6638	24	16	advantage	advantage	NOUN
ejpam-6638	24	17	of	of	ADP
ejpam-6638	24	18	the	the	DET
ejpam-6638	24	19	self	self	NOUN
ejpam-6638	24	20	-	-	PUNCT
ejpam-6638	24	21	similarity	similarity	NOUN
ejpam-6638	24	22	in	in	ADP
ejpam-6638	24	23	images	image	NOUN
ejpam-6638	24	24	by	by	ADP
ejpam-6638	24	25	averaging	average	VERB
ejpam-6638	24	26	similar	similar	ADJ
ejpam-6638	24	27	patches	patch	NOUN
ejpam-6638	24	28	across	across	ADP
ejpam-6638	24	29	the	the	DET
ejpam-6638	24	30	whole	whole	ADJ
ejpam-6638	24	31	image	image	NOUN
ejpam-6638	24	32	[	[	X
ejpam-6638	24	33	3	3	NUM
ejpam-6638	24	34	]	]	PUNCT
ejpam-6638	24	35	.	.	PUNCT
ejpam-6638	25	1	in	in	ADP
ejpam-6638	25	2	order	order	NOUN
ejpam-6638	25	3	to	to	PART
ejpam-6638	25	4	successfully	successfully	ADV
ejpam-6638	25	5	preserve	preserve	VERB
ejpam-6638	25	6	texture	texture	ADJ
ejpam-6638	25	7	information	information	NOUN
ejpam-6638	25	8	during	during	ADP
ejpam-6638	25	9	restoration	restoration	NOUN
ejpam-6638	25	10	,	,	PUNCT
ejpam-6638	25	11	portilla	portilla	PROPN
ejpam-6638	25	12	et	et	PROPN
ejpam-6638	25	13	al	al	PROPN
ejpam-6638	25	14	.	.	PUNCT
ejpam-6638	26	1	[	[	X
ejpam-6638	26	2	4	4	X
ejpam-6638	26	3	]	]	PUNCT
ejpam-6638	26	4	developed	develop	VERB
ejpam-6638	26	5	a	a	DET
ejpam-6638	26	6	statistical	statistical	ADJ
ejpam-6638	26	7	model	model	NOUN
ejpam-6638	26	8	in	in	ADP
ejpam-6638	26	9	the	the	DET
ejpam-6638	26	10	wavelet	wavelet	NOUN
ejpam-6638	26	11	domain	domain	NOUN
ejpam-6638	26	12	for	for	ADP
ejpam-6638	26	13	natural	natural	ADJ
ejpam-6638	26	14	images	image	NOUN
ejpam-6638	26	15	.	.	PUNCT
ejpam-6638	27	1	however	however	ADV
ejpam-6638	27	2	,	,	PUNCT
ejpam-6638	27	3	these	these	DET
ejpam-6638	27	4	conventional	conventional	ADJ
ejpam-6638	27	5	models	model	NOUN
ejpam-6638	27	6	struggled	struggle	VERB
ejpam-6638	27	7	to	to	PART
ejpam-6638	27	8	adjust	adjust	VERB
ejpam-6638	27	9	to	to	ADP
ejpam-6638	27	10	a	a	DET
ejpam-6638	27	11	variety	variety	NOUN
ejpam-6638	27	12	of	of	ADP
ejpam-6638	27	13	complex	complex	ADJ
ejpam-6638	27	14	distortions	distortion	NOUN
ejpam-6638	27	15	.	.	PUNCT
ejpam-6638	28	1	deep	deep	ADJ
ejpam-6638	28	2	learning	learning	NOUN
ejpam-6638	28	3	-	-	PUNCT
ejpam-6638	28	4	based	base	VERB
ejpam-6638	28	5	techniques	technique	NOUN
ejpam-6638	28	6	have	have	AUX
ejpam-6638	28	7	recently	recently	ADV
ejpam-6638	28	8	allowed	allow	VERB
ejpam-6638	28	9	for	for	ADP
ejpam-6638	28	10	a	a	DET
ejpam-6638	28	11	new	new	ADJ
ejpam-6638	28	12	level	level	NOUN
ejpam-6638	28	13	of	of	ADP
ejpam-6638	28	14	proficiency	proficiency	NOUN
ejpam-6638	28	15	in	in	ADP
ejpam-6638	28	16	image	image	NOUN
ejpam-6638	28	17	restoration	restoration	NOUN
ejpam-6638	28	18	tasks	task	NOUN
ejpam-6638	28	19	.	.	PUNCT
ejpam-6638	29	1	dong	dong	PROPN
ejpam-6638	29	2	et	et	PROPN
ejpam-6638	29	3	al	al	PROPN
ejpam-6638	29	4	.	.	PUNCT
ejpam-6638	30	1	[	[	X
ejpam-6638	30	2	5	5	NUM
ejpam-6638	30	3	]	]	PUNCT
ejpam-6638	30	4	provided	provide	VERB
ejpam-6638	30	5	the	the	DET
ejpam-6638	30	6	first	first	ADJ
ejpam-6638	30	7	demonstration	demonstration	NOUN
ejpam-6638	30	8	of	of	ADP
ejpam-6638	30	9	the	the	DET
ejpam-6638	30	10	ability	ability	NOUN
ejpam-6638	30	11	of	of	ADP
ejpam-6638	30	12	srcnns	srcnn	NOUN
ejpam-6638	30	13	(	(	PUNCT
ejpam-6638	30	14	super	super	ADJ
ejpam-6638	30	15	-	-	ADJ
ejpam-6638	30	16	resolution	resolution	ADJ
ejpam-6638	30	17	convolutional	convolutional	ADJ
ejpam-6638	30	18	neural	neural	ADJ
ejpam-6638	30	19	networks	network	NOUN
ejpam-6638	30	20	)	)	PUNCT
ejpam-6638	30	21	to	to	PART
ejpam-6638	30	22	learn	learn	VERB
ejpam-6638	30	23	image	image	NOUN
ejpam-6638	30	24	specifics	specific	NOUN
ejpam-6638	30	25	from	from	ADP
ejpam-6638	30	26	systems	system	NOUN
ejpam-6638	30	27	.	.	PUNCT
ejpam-6638	31	1	ledig	ledig	NOUN
ejpam-6638	31	2	et	et	PROPN
ejpam-6638	31	3	al	al	PROPN
ejpam-6638	31	4	.	.	PUNCT
ejpam-6638	32	1	[	[	X
ejpam-6638	32	2	6	6	NUM
ejpam-6638	32	3	]	]	PUNCT
ejpam-6638	32	4	made	make	VERB
ejpam-6638	32	5	significant	significant	ADJ
ejpam-6638	32	6	progress	progress	NOUN
ejpam-6638	32	7	in	in	ADP
ejpam-6638	32	8	the	the	DET
ejpam-6638	32	9	field	field	NOUN
ejpam-6638	32	10	by	by	ADP
ejpam-6638	32	11	combining	combine	VERB
ejpam-6638	32	12	perceptual	perceptual	ADJ
ejpam-6638	32	13	loss	loss	NOUN
ejpam-6638	32	14	and	and	CCONJ
ejpam-6638	32	15	adversarial	adversarial	ADJ
ejpam-6638	32	16	training	training	NOUN
ejpam-6638	32	17	to	to	PART
ejpam-6638	32	18	improve	improve	VERB
ejpam-6638	32	19	detail	detail	NOUN
ejpam-6638	32	20	in	in	ADP
ejpam-6638	32	21	super	super	ADJ
ejpam-6638	32	22	-	-	ADJ
ejpam-6638	32	23	resolved	resolved	ADJ
ejpam-6638	32	24	images	image	NOUN
ejpam-6638	32	25	.	.	PUNCT
ejpam-6638	33	1	the	the	DET
ejpam-6638	33	2	effectiveness	effectiveness	NOUN
ejpam-6638	33	3	of	of	ADP
ejpam-6638	33	4	residual	residual	ADJ
ejpam-6638	33	5	learning	learning	NOUN
ejpam-6638	33	6	in	in	ADP
ejpam-6638	33	7	conjunction	conjunction	NOUN
ejpam-6638	33	8	with	with	ADP
ejpam-6638	33	9	batch	batch	NOUN
ejpam-6638	33	10	normalization	normalization	NOUN
ejpam-6638	33	11	and	and	CCONJ
ejpam-6638	33	12	cross	cross	ADJ
ejpam-6638	33	13	-	-	ADJ
ejpam-6638	33	14	domain	domain	ADJ
ejpam-6638	33	15	generalization	generalization	NOUN
ejpam-6638	33	16	was	be	AUX
ejpam-6638	33	17	confirmed	confirm	VERB
ejpam-6638	33	18	by	by	ADP
ejpam-6638	33	19	zhang	zhang	PROPN
ejpam-6638	33	20	et	et	PROPN
ejpam-6638	33	21	al	al	PROPN
ejpam-6638	33	22	.	.	PUNCT
ejpam-6638	34	1	[	[	X
ejpam-6638	34	2	7	7	NUM
ejpam-6638	34	3	]	]	ADJ
ejpam-6638	34	4	framework	framework	NOUN
ejpam-6638	34	5	for	for	ADP
ejpam-6638	34	6	residual	residual	ADJ
ejpam-6638	34	7	learning	learning	NOUN
ejpam-6638	34	8	of	of	ADP
ejpam-6638	34	9	deep	deep	ADJ
ejpam-6638	34	10	cnn	cnn	PROPN
ejpam-6638	34	11	for	for	ADP
ejpam-6638	34	12	image	image	NOUN
ejpam-6638	34	13	denoising	denoising	NOUN
ejpam-6638	34	14	.	.	PUNCT
ejpam-6638	35	1	the	the	DET
ejpam-6638	35	2	framework	framework	NOUN
ejpam-6638	35	3	turned	turn	VERB
ejpam-6638	35	4	out	out	ADP
ejpam-6638	35	5	to	to	PART
ejpam-6638	35	6	be	be	AUX
ejpam-6638	35	7	more	more	ADV
ejpam-6638	35	8	effective	effective	ADJ
ejpam-6638	35	9	with	with	ADP
ejpam-6638	35	10	gaussian	gaussian	ADJ
ejpam-6638	35	11	noise	noise	NOUN
ejpam-6638	35	12	.	.	PUNCT
ejpam-6638	36	1	ulyanov	ulyanov	PROPN
ejpam-6638	36	2	et	et	PROPN
ejpam-6638	36	3	al	al	PROPN
ejpam-6638	36	4	.	.	PUNCT
ejpam-6638	37	1	[	[	X
ejpam-6638	37	2	8	8	NUM
ejpam-6638	37	3	]	]	PUNCT
ejpam-6638	37	4	took	take	VERB
ejpam-6638	37	5	a	a	DET
ejpam-6638	37	6	more	more	ADV
ejpam-6638	37	7	extreme	extreme	ADJ
ejpam-6638	37	8	turn	turn	NOUN
ejpam-6638	37	9	by	by	ADP
ejpam-6638	37	10	suggesting	suggest	VERB
ejpam-6638	37	11	the	the	DET
ejpam-6638	37	12	deep	deep	ADJ
ejpam-6638	37	13	image	image	NOUN
ejpam-6638	37	14	prior	prior	ADV
ejpam-6638	37	15	(	(	PUNCT
ejpam-6638	37	16	dip	dip	NOUN
ejpam-6638	37	17	)	)	PUNCT
ejpam-6638	37	18	,	,	PUNCT
ejpam-6638	37	19	which	which	PRON
ejpam-6638	37	20	differs	differ	VERB
ejpam-6638	37	21	from	from	ADP
ejpam-6638	37	22	earlier	early	ADJ
ejpam-6638	37	23	approaches	approach	NOUN
ejpam-6638	37	24	because	because	SCONJ
ejpam-6638	37	25	it	it	PRON
ejpam-6638	37	26	does	do	AUX
ejpam-6638	37	27	not	not	PART
ejpam-6638	37	28	require	require	VERB
ejpam-6638	37	29	prior	prior	ADJ
ejpam-6638	37	30	external	external	ADJ
ejpam-6638	37	31	data	datum	NOUN
ejpam-6638	37	32	,	,	PUNCT
ejpam-6638	37	33	relying	rely	VERB
ejpam-6638	37	34	solely	solely	ADV
ejpam-6638	37	35	on	on	ADP
ejpam-6638	37	36	the	the	DET
ejpam-6638	37	37	architecture	architecture	NOUN
ejpam-6638	37	38	of	of	ADP
ejpam-6638	37	39	the	the	DET
ejpam-6638	37	40	network	network	NOUN
ejpam-6638	37	41	itself	itself	PRON
ejpam-6638	37	42	.	.	PUNCT
ejpam-6638	38	1	although	although	SCONJ
ejpam-6638	38	2	it	it	PRON
ejpam-6638	38	3	was	be	AUX
ejpam-6638	38	4	a	a	DET
ejpam-6638	38	5	breakthrough	breakthrough	NOUN
ejpam-6638	38	6	,	,	PUNCT
ejpam-6638	38	7	issues	issue	NOUN
ejpam-6638	38	8	such	such	ADJ
ejpam-6638	38	9	as	as	ADP
ejpam-6638	38	10	slow	slow	ADJ
ejpam-6638	38	11	convergence	convergence	NOUN
ejpam-6638	38	12	and	and	CCONJ
ejpam-6638	38	13	excessive	excessive	ADJ
ejpam-6638	38	14	focus	focus	NOUN
ejpam-6638	38	15	on	on	ADP
ejpam-6638	38	16	noise	noise	NOUN
ejpam-6638	38	17	(	(	PUNCT
ejpam-6638	38	18	overfitting	overfitting	NOUN
ejpam-6638	38	19	)	)	PUNCT
ejpam-6638	38	20	when	when	SCONJ
ejpam-6638	38	21	trained	train	VERB
ejpam-6638	38	22	for	for	ADP
ejpam-6638	38	23	a	a	DET
ejpam-6638	38	24	long	long	ADJ
ejpam-6638	38	25	time	time	NOUN
ejpam-6638	38	26	were	be	AUX
ejpam-6638	38	27	other	other	ADJ
ejpam-6638	38	28	challenges	challenge	NOUN
ejpam-6638	38	29	for	for	ADP
ejpam-6638	38	30	dip	dip	NOUN
ejpam-6638	38	31	.	.	PUNCT
ejpam-6638	39	1	heckel	heckel	PROPN
ejpam-6638	39	2	and	and	CCONJ
ejpam-6638	39	3	hand	hand	NOUN
ejpam-6638	39	4	[	[	X
ejpam-6638	39	5	9	9	NUM
ejpam-6638	39	6	]	]	PUNCT
ejpam-6638	39	7	attempted	attempt	VERB
ejpam-6638	39	8	to	to	PART
ejpam-6638	39	9	improve	improve	VERB
ejpam-6638	39	10	dip	dip	NOUN
ejpam-6638	39	11	by	by	ADP
ejpam-6638	39	12	evaluating	evaluate	VERB
ejpam-6638	39	13	its	its	PRON
ejpam-6638	39	14	early	early	ADJ
ejpam-6638	39	15	stopping	stop	VERB
ejpam-6638	39	16	behavior	behavior	NOUN
ejpam-6638	39	17	and	and	CCONJ
ejpam-6638	39	18	proved	prove	VERB
ejpam-6638	39	19	that	that	SCONJ
ejpam-6638	39	20	some	some	DET
ejpam-6638	39	21	form	form	NOUN
ejpam-6638	39	22	of	of	ADP
ejpam-6638	39	23	regularization	regularization	NOUN
ejpam-6638	39	24	arises	arise	VERB
ejpam-6638	39	25	from	from	ADP
ejpam-6638	39	26	under	under	ADV
ejpam-6638	39	27	-	-	PUNCT
ejpam-6638	39	28	trained	train	VERB
ejpam-6638	39	29	networks	network	NOUN
ejpam-6638	39	30	.	.	PUNCT
ejpam-6638	40	1	xi	xi	X
ejpam-6638	40	2	et	et	PROPN
ejpam-6638	40	3	al	al	PROPN
ejpam-6638	40	4	.	.	PUNCT
ejpam-6638	41	1	[	[	X
ejpam-6638	41	2	10	10	NUM
ejpam-6638	41	3	]	]	PUNCT
ejpam-6638	41	4	applied	apply	VERB
ejpam-6638	41	5	latent	latent	NOUN
ejpam-6638	41	6	regularization	regularization	NOUN
ejpam-6638	41	7	methods	method	NOUN
ejpam-6638	41	8	to	to	PART
ejpam-6638	41	9	stabilize	stabilize	VERB
ejpam-6638	41	10	dip	dip	NOUN
ejpam-6638	41	11	training	training	NOUN
ejpam-6638	41	12	and	and	CCONJ
ejpam-6638	41	13	reduce	reduce	VERB
ejpam-6638	41	14	artifacts	artifact	NOUN
ejpam-6638	41	15	.	.	PUNCT
ejpam-6638	42	1	recent	recent	ADJ
ejpam-6638	42	2	research	research	NOUN
ejpam-6638	42	3	attempts	attempt	VERB
ejpam-6638	42	4	to	to	PART
ejpam-6638	42	5	fill	fill	VERB
ejpam-6638	42	6	the	the	DET
ejpam-6638	42	7	gap	gap	NOUN
ejpam-6638	42	8	of	of	ADP
ejpam-6638	42	9	these	these	DET
ejpam-6638	42	10	weaknesses	weakness	NOUN
ejpam-6638	42	11	by	by	ADP
ejpam-6638	42	12	incorporating	incorporate	VERB
ejpam-6638	42	13	the	the	DET
ejpam-6638	42	14	idea	idea	NOUN
ejpam-6638	42	15	of	of	ADP
ejpam-6638	42	16	attention	attention	NOUN
ejpam-6638	42	17	in	in	ADP
ejpam-6638	42	18	dip	dip	NOUN
ejpam-6638	42	19	,	,	PUNCT
ejpam-6638	42	20	spectral	spectral	ADJ
ejpam-6638	42	21	pooling	pooling	NOUN
ejpam-6638	42	22	and	and	CCONJ
ejpam-6638	42	23	optimization	optimization	NOUN
ejpam-6638	42	24	improvements	improvement	NOUN
ejpam-6638	42	25	.	.	PUNCT
ejpam-6638	43	1	due	due	ADP
ejpam-6638	43	2	to	to	ADP
ejpam-6638	43	3	their	their	PRON
ejpam-6638	43	4	ability	ability	NOUN
ejpam-6638	43	5	to	to	PART
ejpam-6638	43	6	highlight	highlight	VERB
ejpam-6638	43	7	informative	informative	ADJ
ejpam-6638	43	8	features	feature	NOUN
ejpam-6638	43	9	in	in	ADP
ejpam-6638	43	10	various	various	ADJ
ejpam-6638	43	11	receptive	receptive	ADJ
ejpam-6638	43	12	field	field	NOUN
ejpam-6638	43	13	,	,	PUNCT
ejpam-6638	43	14	attention	attention	NOUN
ejpam-6638	43	15	modules	module	NOUN
ejpam-6638	43	16	,	,	PUNCT
ejpam-6638	43	17	in	in	ADP
ejpam-6638	43	18	general	general	ADJ
ejpam-6638	43	19	,	,	PUNCT
ejpam-6638	43	20	and	and	CCONJ
ejpam-6638	43	21	multiscale	multiscale	ADJ
ejpam-6638	43	22	spatial	spatial	ADJ
ejpam-6638	43	23	-	-	PUNCT
ejpam-6638	43	24	channel	channel	NOUN
ejpam-6638	43	25	adaptivity	adaptivity	NOUN
ejpam-6638	43	26	-	-	PUNCT
ejpam-6638	43	27	based	base	VERB
ejpam-6638	43	28	attention	attention	NOUN
ejpam-6638	43	29	in	in	ADP
ejpam-6638	43	30	particular	particular	ADJ
ejpam-6638	43	31	,	,	PUNCT
ejpam-6638	43	32	have	have	AUX
ejpam-6638	43	33	been	be	AUX
ejpam-6638	43	34	proven	prove	VERB
ejpam-6638	43	35	to	to	PART
ejpam-6638	43	36	achieve	achieve	VERB
ejpam-6638	43	37	enhanced	enhanced	ADJ
ejpam-6638	43	38	edge	edge	NOUN
ejpam-6638	43	39	recovery	recovery	NOUN
ejpam-6638	43	40	and	and	CCONJ
ejpam-6638	43	41	texture	texture	ADJ
ejpam-6638	43	42	restoration	restoration	NOUN
ejpam-6638	43	43	[	[	X
ejpam-6638	43	44	11	11	NUM
ejpam-6638	43	45	,	,	PUNCT
ejpam-6638	43	46	12	12	NUM
ejpam-6638	43	47	]	]	PUNCT
ejpam-6638	43	48	.	.	PUNCT
ejpam-6638	44	1	these	these	DET
ejpam-6638	44	2	advances	advance	NOUN
ejpam-6638	44	3	parallel	parallel	ADJ
ejpam-6638	44	4	advances	advance	NOUN
ejpam-6638	44	5	in	in	ADP
ejpam-6638	44	6	transformer	transformer	NOUN
ejpam-6638	44	7	-	-	PUNCT
ejpam-6638	44	8	based	base	VERB
ejpam-6638	44	9	modules	module	NOUN
ejpam-6638	44	10	on	on	ADP
ejpam-6638	44	11	the	the	DET
ejpam-6638	44	12	low	low	ADJ
ejpam-6638	44	13	level	level	NOUN
ejpam-6638	44	14	vision	vision	NOUN
ejpam-6638	44	15	[	[	X
ejpam-6638	44	16	13	13	NUM
ejpam-6638	44	17	]	]	PUNCT
ejpam-6638	44	18	.	.	PUNCT
ejpam-6638	45	1	simultaneously	simultaneously	ADV
ejpam-6638	45	2	,	,	PUNCT
ejpam-6638	45	3	svd	svd	PROPN
ejpam-6638	45	4	pooling	pool	VERB
ejpam-6638	45	5	has	have	AUX
ejpam-6638	45	6	been	be	AUX
ejpam-6638	45	7	developed	develop	VERB
ejpam-6638	45	8	as	as	ADP
ejpam-6638	45	9	a	a	DET
ejpam-6638	45	10	successful	successful	ADJ
ejpam-6638	45	11	strategy	strategy	NOUN
ejpam-6638	45	12	to	to	PART
ejpam-6638	45	13	keep	keep	VERB
ejpam-6638	45	14	global	global	ADJ
ejpam-6638	45	15	image	image	NOUN
ejpam-6638	45	16	structure	structure	NOUN
ejpam-6638	45	17	with	with	ADP
ejpam-6638	45	18	the	the	DET
ejpam-6638	45	19	use	use	NOUN
ejpam-6638	45	20	of	of	ADP
ejpam-6638	45	21	the	the	DET
ejpam-6638	45	22	large	large	ADJ
ejpam-6638	45	23	singular	singular	ADJ
ejpam-6638	45	24	values	value	NOUN
ejpam-6638	45	25	of	of	ADP
ejpam-6638	45	26	feature	feature	NOUN
ejpam-6638	45	27	maps	map	NOUN
ejpam-6638	45	28	that	that	PRON
ejpam-6638	45	29	supersede	supersede	VERB
ejpam-6638	45	30	conventional	conventional	ADJ
ejpam-6638	45	31	pooling	pooling	NOUN
ejpam-6638	45	32	methods	method	NOUN
ejpam-6638	45	33	in	in	ADP
ejpam-6638	45	34	the	the	DET
ejpam-6638	45	35	way	way	NOUN
ejpam-6638	45	36	they	they	PRON
ejpam-6638	45	37	retain	retain	VERB
ejpam-6638	45	38	informative	informative	ADJ
ejpam-6638	45	39	data	datum	NOUN
ejpam-6638	45	40	[	[	X
ejpam-6638	45	41	14	14	NUM
ejpam-6638	45	42	]	]	PUNCT
ejpam-6638	45	43	.	.	PUNCT
ejpam-6638	46	1	these	these	DET
ejpam-6638	46	2	spectral	spectral	ADJ
ejpam-6638	46	3	representations	representation	NOUN
ejpam-6638	46	4	work	work	NOUN
ejpam-6638	46	5	as	as	ADP
ejpam-6638	46	6	complements	complement	NOUN
ejpam-6638	46	7	to	to	ADP
ejpam-6638	46	8	attention	attention	NOUN
ejpam-6638	46	9	networks	network	NOUN
ejpam-6638	46	10	because	because	SCONJ
ejpam-6638	46	11	long	long	ADJ
ejpam-6638	46	12	-	-	PUNCT
ejpam-6638	46	13	range	range	NOUN
ejpam-6638	46	14	correlations	correlation	NOUN
ejpam-6638	46	15	between	between	ADP
ejpam-6638	46	16	features	feature	NOUN
ejpam-6638	46	17	in	in	ADP
ejpam-6638	46	18	and	and	CCONJ
ejpam-6638	46	19	between	between	ADP
ejpam-6638	46	20	channels	channel	NOUN
ejpam-6638	46	21	can	can	AUX
ejpam-6638	46	22	be	be	AUX
ejpam-6638	46	23	learned	learn	VERB
ejpam-6638	46	24	[	[	X
ejpam-6638	46	25	15	15	NUM
ejpam-6638	46	26	]	]	PUNCT
ejpam-6638	46	27	.	.	PUNCT
ejpam-6638	47	1	recent	recent	ADJ
ejpam-6638	47	2	works	work	NOUN
ejpam-6638	47	3	such	such	ADJ
ejpam-6638	47	4	as	as	ADP
ejpam-6638	47	5	chen	chen	PROPN
ejpam-6638	47	6	et	et	PROPN
ejpam-6638	47	7	al	al	PROPN
ejpam-6638	47	8	.	.	PUNCT
ejpam-6638	48	1	(	(	PUNCT
ejpam-6638	48	2	hint	hint	NOUN
ejpam-6638	48	3	)	)	PUNCT
ejpam-6638	49	1	[	[	X
ejpam-6638	49	2	16	16	NUM
ejpam-6638	49	3	]	]	PUNCT
ejpam-6638	49	4	,	,	PUNCT
ejpam-6638	49	5	and	and	CCONJ
ejpam-6638	49	6	ran	run	VERB
ejpam-6638	49	7	et	et	PROPN
ejpam-6638	49	8	al	al	PROPN
ejpam-6638	49	9	.	.	PUNCT
ejpam-6638	50	1	[	[	X
ejpam-6638	50	2	17	17	NUM
ejpam-6638	50	3	]	]	PUNCT
ejpam-6638	50	4	further	far	ADV
ejpam-6638	50	5	illustrate	illustrate	VERB
ejpam-6638	50	6	that	that	PRON
ejpam-6638	50	7	transformerand	transformerand	ADP
ejpam-6638	50	8	attention	attention	NOUN
ejpam-6638	50	9	-	-	PUNCT
ejpam-6638	50	10	driven	drive	VERB
ejpam-6638	50	11	methods	method	NOUN
ejpam-6638	50	12	significantly	significantly	ADV
ejpam-6638	50	13	enhance	enhance	VERB
ejpam-6638	50	14	restoration	restoration	NOUN
ejpam-6638	50	15	of	of	ADP
ejpam-6638	50	16	large	large	ADJ
ejpam-6638	50	17	missing	missing	ADJ
ejpam-6638	50	18	regions	region	NOUN
ejpam-6638	50	19	and	and	CCONJ
ejpam-6638	50	20	complex	complex	ADJ
ejpam-6638	50	21	textures	texture	NOUN
ejpam-6638	50	22	.	.	PUNCT
ejpam-6638	51	1	moreover	moreover	ADV
ejpam-6638	51	2	,	,	PUNCT
ejpam-6638	51	3	martinho	martinho	NOUN
ejpam-6638	51	4	et	et	PROPN
ejpam-6638	51	5	al	al	PROPN
ejpam-6638	51	6	.	.	PUNCT
ejpam-6638	52	1	[	[	X
ejpam-6638	52	2	18	18	NUM
ejpam-6638	52	3	]	]	PUNCT
ejpam-6638	52	4	demonstrate	demonstrate	NOUN
ejpam-6638	52	5	that	that	SCONJ
ejpam-6638	52	6	deep	deep	ADJ
ejpam-6638	52	7	learning	learning	NOUN
ejpam-6638	52	8	models	model	NOUN
ejpam-6638	52	9	tailored	tailor	VERB
ejpam-6638	52	10	for	for	ADP
ejpam-6638	52	11	specialised	specialised	ADJ
ejpam-6638	52	12	domains	domain	NOUN
ejpam-6638	52	13	(	(	PUNCT
ejpam-6638	52	14	e.g.	e.g.	ADV
ejpam-6638	52	15	,	,	PUNCT
ejpam-6638	52	16	underwater	underwater	ADJ
ejpam-6638	52	17	image	image	NOUN
ejpam-6638	52	18	restoration	restoration	NOUN
ejpam-6638	52	19	)	)	PUNCT
ejpam-6638	52	20	push	push	VERB
ejpam-6638	52	21	the	the	DET
ejpam-6638	52	22	limits	limit	NOUN
ejpam-6638	52	23	of	of	ADP
ejpam-6638	52	24	general	general	ADJ
ejpam-6638	52	25	restoration	restoration	NOUN
ejpam-6638	52	26	frameworks	framework	NOUN
ejpam-6638	52	27	,	,	PUNCT
ejpam-6638	52	28	while	while	SCONJ
ejpam-6638	52	29	lin	lin	PROPN
ejpam-6638	52	30	et	et	PROPN
ejpam-6638	52	31	al	al	PROPN
ejpam-6638	52	32	.	.	PUNCT
ejpam-6638	53	1	[	[	X
ejpam-6638	53	2	19	19	NUM
ejpam-6638	53	3	]	]	AUX
ejpam-6638	53	4	survey	survey	NOUN
ejpam-6638	53	5	high	high	ADJ
ejpam-6638	53	6	-	-	PUNCT
ejpam-6638	53	7	resolution	resolution	NOUN
ejpam-6638	53	8	lightfield	lightfield	NOUN
ejpam-6638	53	9	reconstruction	reconstruction	NOUN
ejpam-6638	53	10	,	,	PUNCT
ejpam-6638	53	11	identifying	identify	VERB
ejpam-6638	53	12	the	the	DET
ejpam-6638	53	13	impact	impact	NOUN
ejpam-6638	53	14	of	of	ADP
ejpam-6638	53	15	architecture	architecture	NOUN
ejpam-6638	53	16	and	and	CCONJ
ejpam-6638	53	17	data	data	NOUN
ejpam-6638	53	18	-	-	PUNCT
ejpam-6638	53	19	sampling	sampling	NOUN
ejpam-6638	53	20	.	.	PUNCT
ejpam-6638	54	1	finally	finally	ADV
ejpam-6638	54	2	,	,	PUNCT
ejpam-6638	54	3	li	li	PROPN
ejpam-6638	54	4	et	et	PROPN
ejpam-6638	54	5	al	al	PROPN
ejpam-6638	54	6	.	.	PUNCT
ejpam-6638	55	1	[	[	X
ejpam-6638	55	2	20	20	NUM
ejpam-6638	55	3	]	]	PUNCT
ejpam-6638	55	4	provide	provide	VERB
ejpam-6638	55	5	a	a	DET
ejpam-6638	55	6	comprehensive	comprehensive	ADJ
ejpam-6638	55	7	review	review	NOUN
ejpam-6638	55	8	of	of	ADP
ejpam-6638	55	9	diffusion	diffusion	NOUN
ejpam-6638	55	10	-	-	PUNCT
ejpam-6638	55	11	based	base	VERB
ejpam-6638	55	12	image	image	NOUN
ejpam-6638	55	13	restoration	restoration	NOUN
ejpam-6638	55	14	,	,	PUNCT
ejpam-6638	55	15	emphasising	emphasise	VERB
ejpam-6638	55	16	future	future	ADJ
ejpam-6638	55	17	directions	direction	NOUN
ejpam-6638	55	18	in	in	ADP
ejpam-6638	55	19	restoration	restoration	NOUN
ejpam-6638	55	20	research	research	NOUN
ejpam-6638	55	21	.	.	PUNCT
ejpam-6638	56	1	the	the	DET
ejpam-6638	56	2	other	other	ADJ
ejpam-6638	56	3	important	important	ADJ
ejpam-6638	56	4	aspect	aspect	NOUN
ejpam-6638	56	5	towards	towards	ADP
ejpam-6638	56	6	improving	improve	VERB
ejpam-6638	56	7	dip	dip	NOUN
ejpam-6638	56	8	is	be	AUX
ejpam-6638	56	9	the	the	DET
ejpam-6638	56	10	optimization	optimization	NOUN
ejpam-6638	56	11	landscape	landscape	NOUN
ejpam-6638	56	12	.	.	PUNCT
ejpam-6638	57	1	conventionally	conventionally	ADV
ejpam-6638	57	2	used	use	VERB
ejpam-6638	57	3	optimizers	optimizer	NOUN
ejpam-6638	57	4	such	such	ADJ
ejpam-6638	57	5	as	as	ADP
ejpam-6638	57	6	adam	adam	PROPN
ejpam-6638	57	7	tend	tend	VERB
ejpam-6638	57	8	to	to	PART
ejpam-6638	57	9	be	be	AUX
ejpam-6638	57	10	suboptimal	suboptimal	ADJ
ejpam-6638	57	11	on	on	ADP
ejpam-6638	57	12	ill	ill	ADV
ejpam-6638	57	13	-	-	PUNCT
ejpam-6638	57	14	conditioned	condition	VERB
ejpam-6638	57	15	inverse	inverse	NOUN
ejpam-6638	57	16	problems	problem	NOUN
ejpam-6638	57	17	as	as	SCONJ
ejpam-6638	57	18	seen	see	VERB
ejpam-6638	57	19	in	in	ADP
ejpam-6638	57	20	image	image	NOUN
ejpam-6638	57	21	restoration	restoration	NOUN
ejpam-6638	57	22	.	.	PUNCT
ejpam-6638	58	1	geometries	geometry	NOUN
ejpam-6638	58	2	different	different	ADJ
ejpam-6638	58	3	than	than	ADP
ejpam-6638	58	4	gradient	gradient	ADJ
ejpam-6638	58	5	descent	descent	NOUN
ejpam-6638	58	6	,	,	PUNCT
ejpam-6638	58	7	including	include	VERB
ejpam-6638	58	8	shampoo	shampoo	NOUN
ejpam-6638	58	9	and	and	CCONJ
ejpam-6638	58	10	kfac	kfac	ADJ
ejpam-6638	58	11	preconditioned	preconditioned	ADJ
ejpam-6638	58	12	optimizers	optimizer	NOUN
ejpam-6638	58	13	,	,	PUNCT
ejpam-6638	58	14	allow	allow	VERB
ejpam-6638	58	15	learning	learning	NOUN
ejpam-6638	58	16	that	that	PRON
ejpam-6638	58	17	is	be	AUX
ejpam-6638	58	18	more	more	ADV
ejpam-6638	58	19	certain	certain	ADJ
ejpam-6638	58	20	and	and	CCONJ
ejpam-6638	58	21	efficient	efficient	ADJ
ejpam-6638	58	22	[	[	X
ejpam-6638	58	23	21	21	NUM
ejpam-6638	58	24	,	,	PUNCT
ejpam-6638	58	25	22	22	NUM
ejpam-6638	58	26	]	]	PUNCT
ejpam-6638	58	27	.	.	PUNCT
ejpam-6638	59	1	these	these	DET
ejpam-6638	59	2	strategies	strategy	NOUN
ejpam-6638	59	3	avoid	avoid	VERB
ejpam-6638	59	4	overfitting	overfitte	VERB
ejpam-6638	59	5	and	and	CCONJ
ejpam-6638	59	6	increase	increase	VERB
ejpam-6638	59	7	convergence	convergence	NOUN
ejpam-6638	59	8	in	in	ADP
ejpam-6638	59	9	models	model	NOUN
ejpam-6638	59	10	based	base	VERB
ejpam-6638	59	11	on	on	ADP
ejpam-6638	59	12	dip	dip	NOUN
ejpam-6638	59	13	[	[	X
ejpam-6638	59	14	23	23	NUM
ejpam-6638	59	15	]	]	PUNCT
ejpam-6638	59	16	.	.	PUNCT
ejpam-6638	60	1	there	there	PRON
ejpam-6638	60	2	are	be	VERB
ejpam-6638	60	3	studies	study	NOUN
ejpam-6638	60	4	which	which	PRON
ejpam-6638	60	5	have	have	AUX
ejpam-6638	60	6	also	also	ADV
ejpam-6638	60	7	added	add	VERB
ejpam-6638	60	8	these	these	DET
ejpam-6638	60	9	enhancements	enhancement	NOUN
ejpam-6638	60	10	together	together	ADV
ejpam-6638	60	11	.	.	PUNCT
ejpam-6638	61	1	as	as	ADP
ejpam-6638	61	2	an	an	DET
ejpam-6638	61	3	example	example	NOUN
ejpam-6638	61	4	,	,	PUNCT
ejpam-6638	61	5	attention	attention	NOUN
ejpam-6638	61	6	mechanisms	mechanism	NOUN
ejpam-6638	61	7	have	have	AUX
ejpam-6638	61	8	previously	previously	ADV
ejpam-6638	61	9	shown	show	VERB
ejpam-6638	61	10	to	to	PART
ejpam-6638	61	11	improve	improve	VERB
ejpam-6638	61	12	super	super	ADJ
ejpam-6638	61	13	-	-	NOUN
ejpam-6638	61	14	resolution	resolution	NOUN
ejpam-6638	61	15	and	and	CCONJ
ejpam-6638	61	16	inpainting	inpainte	VERB
ejpam-6638	61	17	in	in	ADP
ejpam-6638	61	18	dip	dip	NOUN
ejpam-6638	61	19	models	model	NOUN
ejpam-6638	61	20	[	[	X
ejpam-6638	61	21	24	24	NUM
ejpam-6638	61	22	]	]	PUNCT
ejpam-6638	61	23	,	,	PUNCT
ejpam-6638	61	24	preconditioned	preconditioned	ADJ
ejpam-6638	61	25	training	training	NOUN
ejpam-6638	61	26	was	be	AUX
ejpam-6638	61	27	also	also	ADV
ejpam-6638	61	28	used	use	VERB
ejpam-6638	61	29	to	to	PART
ejpam-6638	61	30	speed	speed	VERB
ejpam-6638	61	31	convergence	convergence	NOUN
ejpam-6638	61	32	in	in	ADP
ejpam-6638	61	33	large	large	ADJ
ejpam-6638	61	34	-	-	PUNCT
ejpam-6638	61	35	scale	scale	NOUN
ejpam-6638	61	36	medical	medical	ADJ
ejpam-6638	61	37	image	image	NOUN
ejpam-6638	61	38	reconstruction	reconstruction	NOUN
ejpam-6638	62	1	[	[	X
ejpam-6638	62	2	25	25	NUM
ejpam-6638	62	3	]	]	PUNCT
ejpam-6638	62	4	.	.	PUNCT
ejpam-6638	63	1	global	global	PROPN
ejpam-6638	63	2	svd	svd	PROPN
ejpam-6638	63	3	statistics	statistic	NOUN
ejpam-6638	63	4	have	have	AUX
ejpam-6638	63	5	proved	prove	VERB
ejpam-6638	63	6	to	to	PART
ejpam-6638	63	7	be	be	AUX
ejpam-6638	63	8	synergetic	synergetic	ADJ
ejpam-6638	63	9	when	when	SCONJ
ejpam-6638	63	10	coupled	couple	VERB
ejpam-6638	63	11	with	with	ADP
ejpam-6638	63	12	the	the	DET
ejpam-6638	63	13	m.	m.	NOUN
ejpam-6638	63	14	israr	israr	PROPN
ejpam-6638	63	15	et	et	PROPN
ejpam-6638	63	16	al	al	PROPN
ejpam-6638	63	17	.	.	PUNCT
ejpam-6638	63	18	/	/	SYM
ejpam-6638	63	19	eur	eur	PROPN
ejpam-6638	63	20	.	.	PUNCT
ejpam-6638	64	1	j.	j.	PROPN
ejpam-6638	64	2	pure	pure	PROPN
ejpam-6638	64	3	appl	appl	PROPN
ejpam-6638	64	4	.	.	PROPN
ejpam-6638	64	5	math	math	PROPN
ejpam-6638	64	6	,	,	PUNCT
ejpam-6638	64	7	18	18	NUM
ejpam-6638	64	8	(	(	PUNCT
ejpam-6638	64	9	4	4	NUM
ejpam-6638	64	10	)	)	PUNCT
ejpam-6638	64	11	(	(	PUNCT
ejpam-6638	64	12	2025	2025	NUM
ejpam-6638	64	13	)	)	PUNCT
ejpam-6638	64	14	,	,	PUNCT
ejpam-6638	64	15	6638	6638	NUM
ejpam-6638	64	16	3	3	NUM
ejpam-6638	64	17	of	of	ADP
ejpam-6638	64	18	22	22	NUM
ejpam-6638	64	19	attention	attention	NOUN
ejpam-6638	64	20	modules	module	NOUN
ejpam-6638	64	21	in	in	ADP
ejpam-6638	64	22	the	the	DET
ejpam-6638	64	23	retention	retention	NOUN
ejpam-6638	64	24	of	of	ADP
ejpam-6638	64	25	semantic	semantic	ADJ
ejpam-6638	64	26	components	component	NOUN
ejpam-6638	65	1	[	[	X
ejpam-6638	65	2	26	26	NUM
ejpam-6638	65	3	]	]	PUNCT
ejpam-6638	65	4	.	.	PUNCT
ejpam-6638	66	1	wang	wang	PROPN
ejpam-6638	66	2	et	et	PROPN
ejpam-6638	66	3	al	al	PROPN
ejpam-6638	66	4	.	.	PUNCT
ejpam-6638	67	1	[	[	X
ejpam-6638	67	2	27	27	NUM
ejpam-6638	67	3	]	]	PUNCT
ejpam-6638	67	4	introduced	introduce	VERB
ejpam-6638	67	5	the	the	DET
ejpam-6638	67	6	non	non	ADJ
ejpam-6638	67	7	-	-	ADJ
ejpam-6638	67	8	local	local	ADJ
ejpam-6638	67	9	neural	neural	ADJ
ejpam-6638	67	10	network	network	NOUN
ejpam-6638	67	11	,	,	PUNCT
ejpam-6638	67	12	which	which	PRON
ejpam-6638	67	13	is	be	AUX
ejpam-6638	67	14	capable	capable	ADJ
ejpam-6638	67	15	of	of	ADP
ejpam-6638	67	16	capturing	capture	VERB
ejpam-6638	67	17	long	long	ADJ
ejpam-6638	67	18	-	-	PUNCT
ejpam-6638	67	19	range	range	NOUN
ejpam-6638	67	20	dependencies	dependency	NOUN
ejpam-6638	67	21	in	in	ADP
ejpam-6638	67	22	feature	feature	NOUN
ejpam-6638	67	23	maps	map	NOUN
ejpam-6638	67	24	.	.	PUNCT
ejpam-6638	68	1	to	to	PART
ejpam-6638	68	2	enhance	enhance	VERB
ejpam-6638	68	3	feature	feature	NOUN
ejpam-6638	68	4	representation	representation	NOUN
ejpam-6638	68	5	,	,	PUNCT
ejpam-6638	68	6	fu	fu	PROPN
ejpam-6638	68	7	et	et	PROPN
ejpam-6638	68	8	al	al	PROPN
ejpam-6638	68	9	.	.	PUNCT
ejpam-6638	69	1	[	[	X
ejpam-6638	69	2	28	28	NUM
ejpam-6638	69	3	]	]	PUNCT
ejpam-6638	69	4	suggested	suggest	VERB
ejpam-6638	69	5	a	a	DET
ejpam-6638	69	6	dual	dual	ADJ
ejpam-6638	69	7	attention	attention	NOUN
ejpam-6638	69	8	network	network	NOUN
ejpam-6638	69	9	for	for	ADP
ejpam-6638	69	10	scene	scene	NOUN
ejpam-6638	69	11	segmentation	segmentation	NOUN
ejpam-6638	69	12	that	that	PRON
ejpam-6638	69	13	made	make	VERB
ejpam-6638	69	14	use	use	NOUN
ejpam-6638	69	15	of	of	ADP
ejpam-6638	69	16	spatial	spatial	ADJ
ejpam-6638	69	17	and	and	CCONJ
ejpam-6638	69	18	channel	channel	NOUN
ejpam-6638	69	19	attention	attention	NOUN
ejpam-6638	69	20	.	.	PUNCT
ejpam-6638	70	1	to	to	PART
ejpam-6638	70	2	effectively	effectively	ADV
ejpam-6638	70	3	control	control	VERB
ejpam-6638	70	4	feature	feature	NOUN
ejpam-6638	70	5	abstraction	abstraction	NOUN
ejpam-6638	70	6	,	,	PUNCT
ejpam-6638	70	7	simple	simple	ADJ
ejpam-6638	70	8	pooling	pooling	NOUN
ejpam-6638	70	9	techniques	technique	NOUN
ejpam-6638	70	10	are	be	AUX
ejpam-6638	70	11	crucial	crucial	ADJ
ejpam-6638	70	12	.	.	PUNCT
ejpam-6638	71	1	however	however	ADV
ejpam-6638	71	2	,	,	PUNCT
ejpam-6638	71	3	to	to	PART
ejpam-6638	71	4	preserve	preserve	VERB
ejpam-6638	71	5	the	the	DET
ejpam-6638	71	6	singular	singular	ADJ
ejpam-6638	71	7	value	value	NOUN
ejpam-6638	71	8	spectrum	spectrum	NOUN
ejpam-6638	71	9	and	and	CCONJ
ejpam-6638	71	10	generate	generate	VERB
ejpam-6638	71	11	compact	compact	ADJ
ejpam-6638	71	12	yet	yet	CCONJ
ejpam-6638	71	13	informative	informative	ADJ
ejpam-6638	71	14	representations	representation	NOUN
ejpam-6638	71	15	,	,	PUNCT
ejpam-6638	71	16	sun	sun	PROPN
ejpam-6638	71	17	et	et	PROPN
ejpam-6638	71	18	al	al	PROPN
ejpam-6638	71	19	.	.	PUNCT
ejpam-6638	72	1	[	[	X
ejpam-6638	72	2	29	29	NUM
ejpam-6638	72	3	]	]	PUNCT
ejpam-6638	72	4	proposed	propose	VERB
ejpam-6638	72	5	a	a	DET
ejpam-6638	72	6	pooling	pooling	NOUN
ejpam-6638	72	7	method	method	NOUN
ejpam-6638	72	8	based	base	VERB
ejpam-6638	72	9	on	on	ADP
ejpam-6638	72	10	singular	singular	ADJ
ejpam-6638	72	11	value	value	NOUN
ejpam-6638	72	12	decomposition	decomposition	NOUN
ejpam-6638	72	13	(	(	PUNCT
ejpam-6638	72	14	svd	svd	PROPN
ejpam-6638	72	15	)	)	PUNCT
ejpam-6638	72	16	.	.	PUNCT
ejpam-6638	73	1	zhang	zhang	PROPN
ejpam-6638	73	2	et	et	PROPN
ejpam-6638	73	3	al	al	PROPN
ejpam-6638	73	4	.	.	PUNCT
ejpam-6638	74	1	[	[	X
ejpam-6638	74	2	30	30	NUM
ejpam-6638	74	3	]	]	PUNCT
ejpam-6638	74	4	conducted	conduct	VERB
ejpam-6638	74	5	additional	additional	ADJ
ejpam-6638	74	6	research	research	NOUN
ejpam-6638	74	7	on	on	ADP
ejpam-6638	74	8	spectral	spectral	ADJ
ejpam-6638	74	9	pooling	pooling	NOUN
ejpam-6638	74	10	for	for	ADP
ejpam-6638	74	11	dense	dense	ADJ
ejpam-6638	74	12	prediction	prediction	NOUN
ejpam-6638	74	13	tasks	task	NOUN
ejpam-6638	74	14	,	,	PUNCT
ejpam-6638	74	15	emphasising	emphasise	VERB
ejpam-6638	74	16	its	its	PRON
ejpam-6638	74	17	usefulness	usefulness	NOUN
ejpam-6638	74	18	for	for	ADP
ejpam-6638	74	19	tasks	task	NOUN
ejpam-6638	74	20	that	that	PRON
ejpam-6638	74	21	require	require	VERB
ejpam-6638	74	22	structural	structural	ADJ
ejpam-6638	74	23	consistency	consistency	NOUN
ejpam-6638	74	24	.	.	PUNCT
ejpam-6638	75	1	image	image	NOUN
ejpam-6638	75	2	inpainting	inpainte	VERB
ejpam-6638	75	3	is	be	AUX
ejpam-6638	75	4	the	the	DET
ejpam-6638	75	5	problem	problem	NOUN
ejpam-6638	75	6	of	of	ADP
ejpam-6638	75	7	filling	fill	VERB
ejpam-6638	75	8	in	in	ADP
ejpam-6638	75	9	some	some	DET
ejpam-6638	75	10	area	area	NOUN
ejpam-6638	75	11	of	of	ADP
ejpam-6638	75	12	an	an	DET
ejpam-6638	75	13	image	image	NOUN
ejpam-6638	75	14	that	that	PRON
ejpam-6638	75	15	is	be	AUX
ejpam-6638	75	16	missing	miss	VERB
ejpam-6638	75	17	in	in	ADP
ejpam-6638	75	18	some	some	DET
ejpam-6638	75	19	way	way	NOUN
ejpam-6638	75	20	or	or	CCONJ
ejpam-6638	75	21	that	that	PRON
ejpam-6638	75	22	has	have	AUX
ejpam-6638	75	23	been	be	AUX
ejpam-6638	75	24	corrupted	corrupt	VERB
ejpam-6638	75	25	.	.	PUNCT
ejpam-6638	76	1	it	it	PRON
ejpam-6638	76	2	has	have	AUX
ejpam-6638	76	3	attracted	attract	VERB
ejpam-6638	76	4	much	much	ADJ
ejpam-6638	76	5	interest	interest	NOUN
ejpam-6638	76	6	within	within	ADP
ejpam-6638	76	7	the	the	DET
ejpam-6638	76	8	context	context	NOUN
ejpam-6638	76	9	of	of	ADP
ejpam-6638	76	10	computer	computer	NOUN
ejpam-6638	76	11	vision	vision	NOUN
ejpam-6638	76	12	because	because	SCONJ
ejpam-6638	76	13	it	it	PRON
ejpam-6638	76	14	has	have	VERB
ejpam-6638	76	15	numerous	numerous	ADJ
ejpam-6638	76	16	applications	application	NOUN
ejpam-6638	76	17	in	in	ADP
ejpam-6638	76	18	removal	removal	NOUN
ejpam-6638	76	19	of	of	ADP
ejpam-6638	76	20	objects	object	NOUN
ejpam-6638	76	21	,	,	PUNCT
ejpam-6638	76	22	restoration	restoration	NOUN
ejpam-6638	76	23	of	of	ADP
ejpam-6638	76	24	photos	photo	NOUN
ejpam-6638	76	25	,	,	PUNCT
ejpam-6638	76	26	editing	editing	NOUN
ejpam-6638	76	27	of	of	ADP
ejpam-6638	76	28	images	image	NOUN
ejpam-6638	76	29	,	,	PUNCT
ejpam-6638	76	30	and	and	CCONJ
ejpam-6638	76	31	augmented	augment	VERB
ejpam-6638	76	32	reality	reality	NOUN
ejpam-6638	76	33	.	.	PUNCT
ejpam-6638	77	1	some	some	PRON
ejpam-6638	77	2	of	of	ADP
ejpam-6638	77	3	the	the	DET
ejpam-6638	77	4	classical	classical	ADJ
ejpam-6638	77	5	algorithms	algorithm	NOUN
ejpam-6638	77	6	like	like	ADP
ejpam-6638	77	7	diffusion	diffusion	NOUN
ejpam-6638	77	8	-	-	PUNCT
ejpam-6638	77	9	based	base	VERB
ejpam-6638	77	10	and	and	CCONJ
ejpam-6638	77	11	patch	patch	NOUN
ejpam-6638	77	12	-	-	PUNCT
ejpam-6638	77	13	based	base	VERB
ejpam-6638	77	14	may	may	AUX
ejpam-6638	77	15	not	not	PART
ejpam-6638	77	16	work	work	VERB
ejpam-6638	77	17	well	well	ADV
ejpam-6638	77	18	with	with	ADP
ejpam-6638	77	19	respect	respect	NOUN
ejpam-6638	77	20	to	to	ADP
ejpam-6638	77	21	complex	complex	ADJ
ejpam-6638	77	22	structures	structure	NOUN
ejpam-6638	77	23	and	and	CCONJ
ejpam-6638	77	24	textures	texture	NOUN
ejpam-6638	77	25	particularly	particularly	ADV
ejpam-6638	77	26	in	in	ADP
ejpam-6638	77	27	a	a	DET
ejpam-6638	77	28	high	high	ADJ
ejpam-6638	77	29	-	-	PUNCT
ejpam-6638	77	30	resolution	resolution	NOUN
ejpam-6638	77	31	or	or	CCONJ
ejpam-6638	77	32	semantically	semantically	ADV
ejpam-6638	77	33	rich	rich	ADJ
ejpam-6638	77	34	image	image	NOUN
ejpam-6638	77	35	.	.	PUNCT
ejpam-6638	78	1	in	in	ADP
ejpam-6638	78	2	order	order	NOUN
ejpam-6638	78	3	to	to	PART
ejpam-6638	78	4	fill	fill	VERB
ejpam-6638	78	5	these	these	DET
ejpam-6638	78	6	shortcomings	shortcoming	NOUN
ejpam-6638	78	7	,	,	PUNCT
ejpam-6638	78	8	current	current	ADJ
ejpam-6638	78	9	inpainting	inpainting	ADJ
ejpam-6638	78	10	methods	method	NOUN
ejpam-6638	78	11	exploit	exploit	VERB
ejpam-6638	78	12	deep	deep	ADJ
ejpam-6638	78	13	learning	learning	NOUN
ejpam-6638	78	14	and	and	CCONJ
ejpam-6638	78	15	attention	attention	NOUN
ejpam-6638	78	16	to	to	PART
ejpam-6638	78	17	produce	produce	VERB
ejpam-6638	78	18	quality	quality	NOUN
ejpam-6638	78	19	output	output	NOUN
ejpam-6638	78	20	[	[	X
ejpam-6638	78	21	31–33	31–33	NUM
ejpam-6638	78	22	]	]	PUNCT
ejpam-6638	78	23	.	.	PUNCT
ejpam-6638	79	1	new	new	ADJ
ejpam-6638	79	2	deep	deep	ADJ
ejpam-6638	79	3	convolutional	convolutional	ADJ
ejpam-6638	79	4	neural	neural	ADJ
ejpam-6638	79	5	network	network	NOUN
ejpam-6638	79	6	(	(	PUNCT
ejpam-6638	79	7	cnn)-related	cnn)-related	ADJ
ejpam-6638	79	8	architectures	architecture	NOUN
ejpam-6638	79	9	have	have	AUX
ejpam-6638	79	10	achieved	achieve	VERB
ejpam-6638	79	11	better	well	ADJ
ejpam-6638	79	12	structural	structural	ADJ
ejpam-6638	79	13	consistency	consistency	NOUN
ejpam-6638	79	14	as	as	ADV
ejpam-6638	79	15	well	well	ADV
ejpam-6638	79	16	as	as	ADP
ejpam-6638	79	17	semantic	semantic	ADJ
ejpam-6638	79	18	coherence	coherence	NOUN
ejpam-6638	79	19	of	of	ADP
ejpam-6638	79	20	the	the	DET
ejpam-6638	79	21	reconstructed	reconstructed	ADJ
ejpam-6638	79	22	area	area	NOUN
ejpam-6638	79	23	.	.	PUNCT
ejpam-6638	80	1	as	as	ADP
ejpam-6638	80	2	an	an	DET
ejpam-6638	80	3	example	example	NOUN
ejpam-6638	80	4	,	,	PUNCT
ejpam-6638	80	5	the	the	DET
ejpam-6638	80	6	partial	partial	ADJ
ejpam-6638	80	7	convolutions	convolution	NOUN
ejpam-6638	80	8	dynamically	dynamically	ADV
ejpam-6638	80	9	block	block	VERB
ejpam-6638	80	10	certain	certain	ADJ
ejpam-6638	80	11	features	feature	NOUN
ejpam-6638	80	12	of	of	ADP
ejpam-6638	80	13	the	the	DET
ejpam-6638	80	14	input	input	NOUN
ejpam-6638	80	15	to	to	PART
ejpam-6638	80	16	allow	allow	VERB
ejpam-6638	80	17	networks	network	NOUN
ejpam-6638	80	18	to	to	PART
ejpam-6638	80	19	target	target	VERB
ejpam-6638	80	20	only	only	ADV
ejpam-6638	80	21	genuine	genuine	ADJ
ejpam-6638	80	22	pixels	pixel	NOUN
ejpam-6638	80	23	[	[	X
ejpam-6638	80	24	31	31	NUM
ejpam-6638	80	25	]	]	PUNCT
ejpam-6638	80	26	.	.	PUNCT
ejpam-6638	81	1	in	in	ADP
ejpam-6638	81	2	the	the	DET
ejpam-6638	81	3	meantime	meantime	NOUN
ejpam-6638	81	4	,	,	PUNCT
ejpam-6638	81	5	there	there	PRON
ejpam-6638	81	6	has	have	AUX
ejpam-6638	81	7	been	be	AUX
ejpam-6638	81	8	an	an	DET
ejpam-6638	81	9	introduction	introduction	NOUN
ejpam-6638	81	10	of	of	ADP
ejpam-6638	81	11	attention	attention	NOUN
ejpam-6638	81	12	-	-	PUNCT
ejpam-6638	81	13	based	base	VERB
ejpam-6638	81	14	mechanisms	mechanism	NOUN
ejpam-6638	81	15	which	which	PRON
ejpam-6638	81	16	are	be	AUX
ejpam-6638	81	17	aimed	aim	VERB
ejpam-6638	81	18	at	at	ADP
ejpam-6638	81	19	increasing	increase	VERB
ejpam-6638	81	20	the	the	DET
ejpam-6638	81	21	contextual	contextual	ADJ
ejpam-6638	81	22	meaning	meaning	NOUN
ejpam-6638	81	23	of	of	ADP
ejpam-6638	81	24	the	the	DET
ejpam-6638	81	25	image	image	NOUN
ejpam-6638	81	26	[	[	X
ejpam-6638	81	27	34	34	NUM
ejpam-6638	81	28	]	]	PUNCT
ejpam-6638	81	29	.	.	PUNCT
ejpam-6638	82	1	these	these	DET
ejpam-6638	82	2	mechanisms	mechanism	NOUN
ejpam-6638	82	3	simulate	simulate	VERB
ejpam-6638	82	4	long	long	ADJ
ejpam-6638	82	5	-	-	PUNCT
ejpam-6638	82	6	term	term	NOUN
ejpam-6638	82	7	dependency	dependency	NOUN
ejpam-6638	82	8	and	and	CCONJ
ejpam-6638	82	9	are	be	AUX
ejpam-6638	82	10	more	more	ADV
ejpam-6638	82	11	likely	likely	ADJ
ejpam-6638	82	12	to	to	PART
ejpam-6638	82	13	retain	retain	VERB
ejpam-6638	82	14	semantics	semantic	NOUN
ejpam-6638	82	15	information	information	NOUN
ejpam-6638	82	16	on	on	ADP
ejpam-6638	82	17	masked	masked	ADJ
ejpam-6638	82	18	areas	area	NOUN
ejpam-6638	82	19	[	[	X
ejpam-6638	82	20	35	35	NUM
ejpam-6638	82	21	]	]	PUNCT
ejpam-6638	82	22	.	.	PUNCT
ejpam-6638	83	1	in	in	ADP
ejpam-6638	83	2	addition	addition	NOUN
ejpam-6638	83	3	,	,	PUNCT
ejpam-6638	83	4	generative	generative	ADJ
ejpam-6638	83	5	adversarial	adversarial	ADJ
ejpam-6638	83	6	networks	network	NOUN
ejpam-6638	83	7	(	(	PUNCT
ejpam-6638	83	8	gans	gan	NOUN
ejpam-6638	83	9	)	)	PUNCT
ejpam-6638	83	10	have	have	AUX
ejpam-6638	83	11	been	be	AUX
ejpam-6638	83	12	applied	apply	VERB
ejpam-6638	83	13	vastly	vastly	ADV
ejpam-6638	83	14	to	to	PART
ejpam-6638	83	15	enhance	enhance	VERB
ejpam-6638	83	16	the	the	DET
ejpam-6638	83	17	realism	realism	NOUN
ejpam-6638	83	18	of	of	ADP
ejpam-6638	83	19	inpainted	inpainte	VERB
ejpam-6638	83	20	outputs	output	NOUN
ejpam-6638	83	21	[	[	X
ejpam-6638	83	22	33	33	NUM
ejpam-6638	83	23	]	]	PUNCT
ejpam-6638	83	24	.	.	PUNCT
ejpam-6638	84	1	special	special	ADJ
ejpam-6638	84	2	in	in	ADP
ejpam-6638	84	3	these	these	DET
ejpam-6638	84	4	models	model	NOUN
ejpam-6638	84	5	is	be	AUX
ejpam-6638	84	6	their	their	PRON
ejpam-6638	84	7	capability	capability	NOUN
ejpam-6638	84	8	to	to	PART
ejpam-6638	84	9	create	create	VERB
ejpam-6638	84	10	fine	fine	ADJ
ejpam-6638	84	11	textures	texture	NOUN
ejpam-6638	84	12	and	and	CCONJ
ejpam-6638	84	13	give	give	VERB
ejpam-6638	84	14	smooth	smooth	ADJ
ejpam-6638	84	15	transitions	transition	NOUN
ejpam-6638	84	16	between	between	ADP
ejpam-6638	84	17	familiar	familiar	ADJ
ejpam-6638	84	18	and	and	CCONJ
ejpam-6638	84	19	unfamiliar	unfamiliar	ADJ
ejpam-6638	84	20	space	space	NOUN
ejpam-6638	84	21	.	.	PUNCT
ejpam-6638	85	1	there	there	PRON
ejpam-6638	85	2	also	also	ADV
ejpam-6638	85	3	has	have	AUX
ejpam-6638	85	4	been	be	AUX
ejpam-6638	85	5	the	the	DET
ejpam-6638	85	6	emergence	emergence	NOUN
ejpam-6638	85	7	of	of	ADP
ejpam-6638	85	8	hybrid	hybrid	ADJ
ejpam-6638	85	9	models	model	NOUN
ejpam-6638	85	10	that	that	PRON
ejpam-6638	85	11	merge	merge	VERB
ejpam-6638	85	12	structural	structural	ADJ
ejpam-6638	85	13	priors	prior	NOUN
ejpam-6638	85	14	and	and	CCONJ
ejpam-6638	85	15	multiscale	multiscale	ADJ
ejpam-6638	85	16	representations	representation	NOUN
ejpam-6638	85	17	could	could	AUX
ejpam-6638	85	18	be	be	AUX
ejpam-6638	85	19	robust	robust	ADJ
ejpam-6638	85	20	to	to	ADP
ejpam-6638	85	21	degraded	degraded	ADJ
ejpam-6638	85	22	images	image	NOUN
ejpam-6638	85	23	in	in	ADP
ejpam-6638	85	24	the	the	DET
ejpam-6638	85	25	real	real	ADJ
ejpam-6638	85	26	world	world	NOUN
ejpam-6638	86	1	[	[	X
ejpam-6638	86	2	36	36	NUM
ejpam-6638	86	3	,	,	PUNCT
ejpam-6638	86	4	37	37	NUM
ejpam-6638	86	5	]	]	PUNCT
ejpam-6638	86	6	.	.	PUNCT
ejpam-6638	87	1	transformer	transformer	NOUN
ejpam-6638	87	2	based	base	VERB
ejpam-6638	87	3	networks	network	NOUN
ejpam-6638	87	4	have	have	AUX
ejpam-6638	87	5	also	also	ADV
ejpam-6638	87	6	lately	lately	ADV
ejpam-6638	87	7	been	be	AUX
ejpam-6638	87	8	applied	apply	VERB
ejpam-6638	87	9	to	to	ADP
ejpam-6638	87	10	inpainting	inpainte	VERB
ejpam-6638	87	11	where	where	SCONJ
ejpam-6638	87	12	they	they	PRON
ejpam-6638	87	13	provide	provide	VERB
ejpam-6638	87	14	a	a	DET
ejpam-6638	87	15	better	well	ADJ
ejpam-6638	87	16	global	global	ADJ
ejpam-6638	87	17	reasoning	reasoning	NOUN
ejpam-6638	87	18	abilities	ability	NOUN
ejpam-6638	87	19	[	[	X
ejpam-6638	87	20	38	38	NUM
ejpam-6638	87	21	]	]	PUNCT
ejpam-6638	87	22	.	.	PUNCT
ejpam-6638	88	1	key	key	ADJ
ejpam-6638	88	2	contributions	contribution	NOUN
ejpam-6638	88	3	in	in	ADP
ejpam-6638	88	4	comparison	comparison	NOUN
ejpam-6638	88	5	with	with	ADP
ejpam-6638	88	6	existing	exist	VERB
ejpam-6638	88	7	literature	literature	NOUN
ejpam-6638	88	8	:	:	PUNCT
ejpam-6638	88	9	•	•	NUM
ejpam-6638	88	10	svd	svd	PROPN
ejpam-6638	88	11	pooling	pool	VERB
ejpam-6638	88	12	for	for	ADP
ejpam-6638	88	13	enhanced	enhanced	ADJ
ejpam-6638	88	14	structural	structural	ADJ
ejpam-6638	88	15	representation	representation	NOUN
ejpam-6638	88	16	:	:	PUNCT
ejpam-6638	88	17	we	we	PRON
ejpam-6638	88	18	incorporate	incorporate	VERB
ejpam-6638	88	19	svdbased	svdbase	VERB
ejpam-6638	88	20	pooling	pool	VERB
ejpam-6638	88	21	to	to	PART
ejpam-6638	88	22	robustly	robustly	ADV
ejpam-6638	88	23	capture	capture	VERB
ejpam-6638	88	24	global	global	ADJ
ejpam-6638	88	25	contextual	contextual	ADJ
ejpam-6638	88	26	structures	structure	NOUN
ejpam-6638	88	27	through	through	ADP
ejpam-6638	88	28	low	low	ADJ
ejpam-6638	88	29	-	-	PUNCT
ejpam-6638	88	30	rank	rank	NOUN
ejpam-6638	88	31	feature	feature	NOUN
ejpam-6638	88	32	approximations	approximation	NOUN
ejpam-6638	89	1	[	[	X
ejpam-6638	89	2	39	39	NUM
ejpam-6638	89	3	]	]	PUNCT
ejpam-6638	89	4	,	,	PUNCT
ejpam-6638	89	5	thereby	thereby	ADV
ejpam-6638	89	6	enhancing	enhance	VERB
ejpam-6638	89	7	the	the	DET
ejpam-6638	89	8	semantic	semantic	ADJ
ejpam-6638	89	9	expressiveness	expressiveness	NOUN
ejpam-6638	89	10	of	of	ADP
ejpam-6638	89	11	attention	attention	NOUN
ejpam-6638	89	12	modules	module	NOUN
ejpam-6638	89	13	[	[	X
ejpam-6638	89	14	40	40	NUM
ejpam-6638	89	15	]	]	PUNCT
ejpam-6638	89	16	.	.	PUNCT
ejpam-6638	90	1	this	this	DET
ejpam-6638	90	2	spectral	spectral	ADJ
ejpam-6638	90	3	pooling	pooling	NOUN
ejpam-6638	90	4	technique	technique	NOUN
ejpam-6638	90	5	also	also	ADV
ejpam-6638	90	6	serves	serve	VERB
ejpam-6638	90	7	as	as	ADP
ejpam-6638	90	8	a	a	DET
ejpam-6638	90	9	natural	natural	ADJ
ejpam-6638	90	10	regularizer	regularizer	NOUN
ejpam-6638	90	11	,	,	PUNCT
ejpam-6638	90	12	helping	help	VERB
ejpam-6638	90	13	to	to	PART
ejpam-6638	90	14	mitigate	mitigate	VERB
ejpam-6638	90	15	overfitting	overfitte	VERB
ejpam-6638	90	16	in	in	ADP
ejpam-6638	90	17	inverse	inverse	ADJ
ejpam-6638	90	18	imaging	imaging	NOUN
ejpam-6638	90	19	tasks	task	NOUN
ejpam-6638	90	20	.	.	PUNCT
ejpam-6638	91	1	•	•	NUM
ejpam-6638	91	2	softplus	softplus	ADJ
ejpam-6638	91	3	activation	activation	NOUN
ejpam-6638	91	4	for	for	ADP
ejpam-6638	91	5	gradient	gradient	ADJ
ejpam-6638	91	6	stability	stability	NOUN
ejpam-6638	91	7	:	:	PUNCT
ejpam-6638	91	8	replacing	replace	VERB
ejpam-6638	91	9	conventional	conventional	ADJ
ejpam-6638	91	10	activations	activation	NOUN
ejpam-6638	91	11	,	,	PUNCT
ejpam-6638	91	12	we	we	PRON
ejpam-6638	91	13	employ	employ	VERB
ejpam-6638	91	14	the	the	DET
ejpam-6638	91	15	softplus	softplus	ADJ
ejpam-6638	91	16	function	function	NOUN
ejpam-6638	91	17	to	to	PART
ejpam-6638	91	18	ensure	ensure	VERB
ejpam-6638	91	19	smooth	smooth	ADJ
ejpam-6638	91	20	,	,	PUNCT
ejpam-6638	91	21	non	non	ADJ
ejpam-6638	91	22	-	-	ADJ
ejpam-6638	91	23	saturating	saturate	VERB
ejpam-6638	91	24	gradients	gradient	NOUN
ejpam-6638	91	25	that	that	PRON
ejpam-6638	91	26	promote	promote	VERB
ejpam-6638	91	27	stable	stable	ADJ
ejpam-6638	91	28	training	training	NOUN
ejpam-6638	91	29	across	across	ADP
ejpam-6638	91	30	deeper	deep	ADJ
ejpam-6638	91	31	layers	layer	NOUN
ejpam-6638	92	1	[	[	X
ejpam-6638	92	2	41	41	NUM
ejpam-6638	92	3	,	,	PUNCT
ejpam-6638	92	4	42	42	NUM
ejpam-6638	92	5	]	]	PUNCT
ejpam-6638	92	6	.	.	PUNCT
ejpam-6638	93	1	this	this	DET
ejpam-6638	93	2	results	result	NOUN
ejpam-6638	93	3	in	in	ADP
ejpam-6638	93	4	improved	improved	ADJ
ejpam-6638	93	5	convergence	convergence	NOUN
ejpam-6638	93	6	and	and	CCONJ
ejpam-6638	93	7	a	a	DET
ejpam-6638	93	8	more	more	ADV
ejpam-6638	93	9	robust	robust	ADJ
ejpam-6638	93	10	gradient	gradient	NOUN
ejpam-6638	93	11	flow	flow	NOUN
ejpam-6638	93	12	,	,	PUNCT
ejpam-6638	93	13	especially	especially	ADV
ejpam-6638	93	14	in	in	ADP
ejpam-6638	93	15	noise	noise	NOUN
ejpam-6638	93	16	-	-	PUNCT
ejpam-6638	93	17	sensitive	sensitive	ADJ
ejpam-6638	93	18	scenarios	scenario	NOUN
ejpam-6638	93	19	.	.	PUNCT
ejpam-6638	94	1	•	•	NUM
ejpam-6638	94	2	multiscale	multiscale	ADJ
ejpam-6638	94	3	attention	attention	NOUN
ejpam-6638	94	4	for	for	ADP
ejpam-6638	94	5	fine	fine	ADJ
ejpam-6638	94	6	-	-	PUNCT
ejpam-6638	94	7	detail	detail	NOUN
ejpam-6638	94	8	recovery	recovery	NOUN
ejpam-6638	94	9	:	:	PUNCT
ejpam-6638	94	10	a	a	DET
ejpam-6638	94	11	hierarchical	hierarchical	ADJ
ejpam-6638	94	12	soft	soft	ADJ
ejpam-6638	94	13	attention	attention	NOUN
ejpam-6638	94	14	mechanism	mechanism	NOUN
ejpam-6638	94	15	is	be	AUX
ejpam-6638	94	16	introduced	introduce	VERB
ejpam-6638	94	17	to	to	PART
ejpam-6638	94	18	selectively	selectively	ADV
ejpam-6638	94	19	emphasize	emphasize	VERB
ejpam-6638	94	20	features	feature	NOUN
ejpam-6638	94	21	across	across	ADP
ejpam-6638	94	22	multiple	multiple	ADJ
ejpam-6638	94	23	resolutions	resolution	NOUN
ejpam-6638	94	24	.	.	PUNCT
ejpam-6638	95	1	this	this	DET
ejpam-6638	95	2	design	design	NOUN
ejpam-6638	95	3	significantly	significantly	ADV
ejpam-6638	95	4	enhances	enhance	VERB
ejpam-6638	95	5	the	the	DET
ejpam-6638	95	6	model	model	NOUN
ejpam-6638	95	7	’s	’s	PART
ejpam-6638	95	8	capacity	capacity	NOUN
ejpam-6638	95	9	to	to	PART
ejpam-6638	95	10	recover	recover	VERB
ejpam-6638	95	11	fine	fine	ADJ
ejpam-6638	95	12	textures	texture	NOUN
ejpam-6638	95	13	and	and	CCONJ
ejpam-6638	95	14	intricate	intricate	ADJ
ejpam-6638	95	15	image	image	NOUN
ejpam-6638	95	16	details	detail	NOUN
ejpam-6638	95	17	[	[	X
ejpam-6638	95	18	43	43	NUM
ejpam-6638	95	19	,	,	PUNCT
ejpam-6638	95	20	44	44	NUM
ejpam-6638	95	21	]	]	PUNCT
ejpam-6638	95	22	.	.	PUNCT
ejpam-6638	96	1	•	•	NUM
ejpam-6638	96	2	preconditioned	precondition	VERB
ejpam-6638	96	3	optimization	optimization	NOUN
ejpam-6638	96	4	for	for	ADP
ejpam-6638	96	5	accelerated	accelerated	ADJ
ejpam-6638	96	6	convergence	convergence	NOUN
ejpam-6638	96	7	:	:	PUNCT
ejpam-6638	96	8	leveraging	leverage	VERB
ejpam-6638	96	9	advanced	advanced	ADJ
ejpam-6638	96	10	preconditioned	precondition	VERB
ejpam-6638	96	11	optimizers	optimizer	NOUN
ejpam-6638	96	12	such	such	ADJ
ejpam-6638	96	13	as	as	ADP
ejpam-6638	96	14	diffgrad	diffgrad	PROPN
ejpam-6638	96	15	and	and	CCONJ
ejpam-6638	96	16	apollo	apollo	PROPN
ejpam-6638	97	1	,	,	PUNCT
ejpam-6638	97	2	our	our	PRON
ejpam-6638	97	3	framework	framework	NOUN
ejpam-6638	97	4	achieves	achieve	VERB
ejpam-6638	97	5	faster	fast	ADJ
ejpam-6638	97	6	convergence	convergence	NOUN
ejpam-6638	97	7	and	and	CCONJ
ejpam-6638	97	8	greater	great	ADJ
ejpam-6638	97	9	numerical	numerical	ADJ
ejpam-6638	97	10	stability	stability	NOUN
ejpam-6638	97	11	by	by	ADP
ejpam-6638	97	12	dynamically	dynamically	ADV
ejpam-6638	97	13	adapting	adapt	VERB
ejpam-6638	97	14	to	to	PART
ejpam-6638	97	15	curvature	curvature	VERB
ejpam-6638	97	16	and	and	CCONJ
ejpam-6638	97	17	gradient	gradient	ADJ
ejpam-6638	97	18	variance	variance	NOUN
ejpam-6638	97	19	[	[	X
ejpam-6638	97	20	45–47	45–47	NOUN
ejpam-6638	97	21	]	]	PUNCT
ejpam-6638	97	22	.	.	PUNCT
ejpam-6638	98	1	these	these	DET
ejpam-6638	98	2	methods	method	NOUN
ejpam-6638	98	3	outperform	outperform	VERB
ejpam-6638	98	4	traditional	traditional	ADJ
ejpam-6638	98	5	optimizers	optimizer	NOUN
ejpam-6638	98	6	in	in	ADP
ejpam-6638	98	7	both	both	DET
ejpam-6638	98	8	convergence	convergence	NOUN
ejpam-6638	98	9	speed	speed	NOUN
ejpam-6638	98	10	and	and	CCONJ
ejpam-6638	98	11	final	final	ADJ
ejpam-6638	98	12	accuracy	accuracy	NOUN
ejpam-6638	98	13	.	.	PUNCT
ejpam-6638	99	1	m.	m.	NOUN
ejpam-6638	99	2	israr	israr	PROPN
ejpam-6638	99	3	et	et	PROPN
ejpam-6638	99	4	al	al	PROPN
ejpam-6638	99	5	.	.	PUNCT
ejpam-6638	99	6	/	/	SYM
ejpam-6638	99	7	eur	eur	PROPN
ejpam-6638	99	8	.	.	PUNCT
ejpam-6638	100	1	j.	j.	PROPN
ejpam-6638	100	2	pure	pure	PROPN
ejpam-6638	100	3	appl	appl	PROPN
ejpam-6638	100	4	.	.	PROPN
ejpam-6638	100	5	math	math	PROPN
ejpam-6638	100	6	,	,	PUNCT
ejpam-6638	100	7	18	18	NUM
ejpam-6638	100	8	(	(	PUNCT
ejpam-6638	100	9	4	4	NUM
ejpam-6638	100	10	)	)	PUNCT
ejpam-6638	100	11	(	(	PUNCT
ejpam-6638	100	12	2025	2025	NUM
ejpam-6638	100	13	)	)	PUNCT
ejpam-6638	100	14	,	,	PUNCT
ejpam-6638	100	15	6638	6638	NUM
ejpam-6638	100	16	4	4	NUM
ejpam-6638	100	17	of	of	ADP
ejpam-6638	100	18	22	22	NUM
ejpam-6638	100	19	•	•	NOUN
ejpam-6638	100	20	comprehensive	comprehensive	ADJ
ejpam-6638	100	21	quantitative	quantitative	ADJ
ejpam-6638	100	22	evaluation	evaluation	NOUN
ejpam-6638	100	23	:	:	PUNCT
ejpam-6638	100	24	we	we	PRON
ejpam-6638	100	25	rigorously	rigorously	ADV
ejpam-6638	100	26	evaluate	evaluate	VERB
ejpam-6638	100	27	our	our	PRON
ejpam-6638	100	28	method	method	NOUN
ejpam-6638	100	29	on	on	ADP
ejpam-6638	100	30	standard	standard	ADJ
ejpam-6638	100	31	image	image	NOUN
ejpam-6638	100	32	denoising	denoise	VERB
ejpam-6638	100	33	and	and	CCONJ
ejpam-6638	100	34	inpainting	inpainte	VERB
ejpam-6638	100	35	benchmarks	benchmark	NOUN
ejpam-6638	100	36	using	use	VERB
ejpam-6638	100	37	objective	objective	ADJ
ejpam-6638	100	38	metrics	metric	NOUN
ejpam-6638	100	39	such	such	ADJ
ejpam-6638	100	40	as	as	ADP
ejpam-6638	100	41	mse	mse	NOUN
ejpam-6638	100	42	,	,	PUNCT
ejpam-6638	100	43	psnr	psnr	NOUN
ejpam-6638	100	44	,	,	PUNCT
ejpam-6638	100	45	and	and	CCONJ
ejpam-6638	100	46	ssim	ssim	NOUN
ejpam-6638	100	47	.	.	PUNCT
ejpam-6638	101	1	the	the	DET
ejpam-6638	101	2	proposed	propose	VERB
ejpam-6638	101	3	approach	approach	NOUN
ejpam-6638	101	4	consistently	consistently	ADV
ejpam-6638	101	5	surpasses	surpass	VERB
ejpam-6638	101	6	classical	classical	ADJ
ejpam-6638	101	7	dip	dip	NOUN
ejpam-6638	101	8	and	and	CCONJ
ejpam-6638	101	9	several	several	ADJ
ejpam-6638	101	10	state	state	NOUN
ejpam-6638	101	11	-	-	PUNCT
ejpam-6638	101	12	of	of	ADP
ejpam-6638	101	13	-	-	PUNCT
ejpam-6638	101	14	the	the	DET
ejpam-6638	101	15	-	-	PUNCT
ejpam-6638	101	16	art	art	NOUN
ejpam-6638	101	17	restoration	restoration	NOUN
ejpam-6638	101	18	frameworks	framework	NOUN
ejpam-6638	101	19	.	.	PUNCT
ejpam-6638	102	1	•	•	ADJ
ejpam-6638	102	2	in	in	ADP
ejpam-6638	102	3	-	-	PUNCT
ejpam-6638	102	4	depth	depth	NOUN
ejpam-6638	102	5	visual	visual	ADJ
ejpam-6638	102	6	and	and	CCONJ
ejpam-6638	102	7	optimizer	optimizer	NOUN
ejpam-6638	102	8	performance	performance	NOUN
ejpam-6638	102	9	analysis	analysis	NOUN
ejpam-6638	102	10	:	:	PUNCT
ejpam-6638	102	11	extensive	extensive	ADJ
ejpam-6638	102	12	visual	visual	ADJ
ejpam-6638	102	13	comparisons	comparison	NOUN
ejpam-6638	102	14	including	include	VERB
ejpam-6638	102	15	qualitative	qualitative	ADJ
ejpam-6638	102	16	outputs	output	NOUN
ejpam-6638	102	17	,	,	PUNCT
ejpam-6638	102	18	psnr	psnr	NOUN
ejpam-6638	102	19	/	/	SYM
ejpam-6638	102	20	ssim	ssim	NOUN
ejpam-6638	102	21	evolution	evolution	NOUN
ejpam-6638	102	22	curves	curve	NOUN
ejpam-6638	102	23	,	,	PUNCT
ejpam-6638	102	24	and	and	CCONJ
ejpam-6638	102	25	3d	3d	NUM
ejpam-6638	102	26	bar	bar	NOUN
ejpam-6638	102	27	plots	plot	NOUN
ejpam-6638	102	28	are	be	AUX
ejpam-6638	102	29	provided	provide	VERB
ejpam-6638	102	30	to	to	PART
ejpam-6638	102	31	demonstrate	demonstrate	VERB
ejpam-6638	102	32	the	the	DET
ejpam-6638	102	33	efficiency	efficiency	NOUN
ejpam-6638	102	34	and	and	CCONJ
ejpam-6638	102	35	convergence	convergence	NOUN
ejpam-6638	102	36	behavior	behavior	NOUN
ejpam-6638	102	37	of	of	ADP
ejpam-6638	102	38	different	different	ADJ
ejpam-6638	102	39	optimizers	optimizer	NOUN
ejpam-6638	102	40	,	,	PUNCT
ejpam-6638	102	41	emphasizing	emphasize	VERB
ejpam-6638	102	42	the	the	DET
ejpam-6638	102	43	impact	impact	NOUN
ejpam-6638	102	44	of	of	ADP
ejpam-6638	102	45	optimization	optimization	NOUN
ejpam-6638	102	46	strategies	strategy	NOUN
ejpam-6638	102	47	on	on	ADP
ejpam-6638	102	48	restoration	restoration	NOUN
ejpam-6638	102	49	quality	quality	NOUN
ejpam-6638	102	50	.	.	PUNCT
ejpam-6638	103	1	the	the	DET
ejpam-6638	103	2	remainder	remainder	NOUN
ejpam-6638	103	3	of	of	ADP
ejpam-6638	103	4	this	this	DET
ejpam-6638	103	5	article	article	NOUN
ejpam-6638	103	6	is	be	AUX
ejpam-6638	103	7	organized	organize	VERB
ejpam-6638	103	8	as	as	SCONJ
ejpam-6638	103	9	follows	follow	VERB
ejpam-6638	103	10	:	:	PUNCT
ejpam-6638	103	11	section	section	NOUN
ejpam-6638	103	12	2	2	NUM
ejpam-6638	103	13	provides	provide	VERB
ejpam-6638	103	14	an	an	DET
ejpam-6638	103	15	overview	overview	NOUN
ejpam-6638	103	16	of	of	ADP
ejpam-6638	103	17	key	key	ADJ
ejpam-6638	103	18	concepts	concept	NOUN
ejpam-6638	103	19	in	in	ADP
ejpam-6638	103	20	image	image	NOUN
ejpam-6638	103	21	restoration	restoration	NOUN
ejpam-6638	103	22	and	and	CCONJ
ejpam-6638	103	23	introduces	introduce	VERB
ejpam-6638	103	24	the	the	DET
ejpam-6638	103	25	deep	deep	ADJ
ejpam-6638	103	26	image	image	NOUN
ejpam-6638	103	27	prior	prior	ADJ
ejpam-6638	103	28	framework	framework	NOUN
ejpam-6638	103	29	.	.	PUNCT
ejpam-6638	104	1	section	section	NOUN
ejpam-6638	104	2	3	3	NUM
ejpam-6638	104	3	describes	describe	VERB
ejpam-6638	104	4	our	our	PRON
ejpam-6638	104	5	proposed	propose	VERB
ejpam-6638	104	6	improvements	improvement	NOUN
ejpam-6638	104	7	,	,	PUNCT
ejpam-6638	104	8	including	include	VERB
ejpam-6638	104	9	svd	svd	PROPN
ejpam-6638	104	10	-	-	PUNCT
ejpam-6638	104	11	based	base	VERB
ejpam-6638	104	12	pooling	pooling	NOUN
ejpam-6638	104	13	,	,	PUNCT
ejpam-6638	104	14	softplus	softplus	ADJ
ejpam-6638	104	15	activation	activation	NOUN
ejpam-6638	104	16	,	,	PUNCT
ejpam-6638	104	17	and	and	CCONJ
ejpam-6638	104	18	multiscale	multiscale	ADJ
ejpam-6638	104	19	attention	attention	NOUN
ejpam-6638	104	20	mechanisms	mechanism	NOUN
ejpam-6638	104	21	.	.	PUNCT
ejpam-6638	105	1	section	section	NOUN
ejpam-6638	105	2	4	4	NUM
ejpam-6638	105	3	presents	present	VERB
ejpam-6638	105	4	experimental	experimental	ADJ
ejpam-6638	105	5	results	result	NOUN
ejpam-6638	105	6	on	on	ADP
ejpam-6638	105	7	image	image	NOUN
ejpam-6638	105	8	denoising	denoise	VERB
ejpam-6638	105	9	and	and	CCONJ
ejpam-6638	105	10	inpainting	inpainte	VERB
ejpam-6638	105	11	tasks	task	NOUN
ejpam-6638	105	12	,	,	PUNCT
ejpam-6638	105	13	along	along	ADP
ejpam-6638	105	14	with	with	ADP
ejpam-6638	105	15	ablation	ablation	NOUN
ejpam-6638	105	16	studies	study	NOUN
ejpam-6638	105	17	evaluating	evaluate	VERB
ejpam-6638	105	18	the	the	DET
ejpam-6638	105	19	impact	impact	NOUN
ejpam-6638	105	20	of	of	ADP
ejpam-6638	105	21	different	different	ADJ
ejpam-6638	105	22	optimizers	optimizer	NOUN
ejpam-6638	105	23	.	.	PUNCT
ejpam-6638	106	1	section	section	NOUN
ejpam-6638	106	2	5	5	NUM
ejpam-6638	106	3	concludes	conclude	VERB
ejpam-6638	106	4	the	the	DET
ejpam-6638	106	5	paper	paper	NOUN
ejpam-6638	106	6	by	by	ADP
ejpam-6638	106	7	summarizing	summarize	VERB
ejpam-6638	106	8	the	the	DET
ejpam-6638	106	9	main	main	ADJ
ejpam-6638	106	10	findings	finding	NOUN
ejpam-6638	106	11	and	and	CCONJ
ejpam-6638	106	12	outlining	outline	VERB
ejpam-6638	106	13	potential	potential	ADJ
ejpam-6638	106	14	future	future	ADJ
ejpam-6638	106	15	work	work	NOUN
ejpam-6638	106	16	.	.	PUNCT
ejpam-6638	107	1	supplementary	supplementary	ADJ
ejpam-6638	107	2	materials	material	NOUN
ejpam-6638	107	3	and	and	CCONJ
ejpam-6638	107	4	references	reference	NOUN
ejpam-6638	107	5	are	be	AUX
ejpam-6638	107	6	provided	provide	VERB
ejpam-6638	107	7	at	at	ADP
ejpam-6638	107	8	the	the	DET
ejpam-6638	107	9	end	end	NOUN
ejpam-6638	107	10	.	.	PUNCT
ejpam-6638	108	1	2	2	X
ejpam-6638	108	2	.	.	X
ejpam-6638	108	3	background	background	NOUN
ejpam-6638	108	4	let	let	VERB
ejpam-6638	108	5	’s	’s	NOUN
ejpam-6638	108	6	explore	explore	VERB
ejpam-6638	108	7	the	the	DET
ejpam-6638	108	8	topic	topic	NOUN
ejpam-6638	108	9	of	of	ADP
ejpam-6638	108	10	image	image	NOUN
ejpam-6638	108	11	restoration	restoration	NOUN
ejpam-6638	108	12	,	,	PUNCT
ejpam-6638	108	13	with	with	ADP
ejpam-6638	108	14	a	a	DET
ejpam-6638	108	15	particular	particular	ADJ
ejpam-6638	108	16	focus	focus	NOUN
ejpam-6638	108	17	on	on	ADP
ejpam-6638	108	18	image	image	NOUN
ejpam-6638	108	19	denoising	denoising	NOUN
ejpam-6638	108	20	and	and	CCONJ
ejpam-6638	108	21	image	image	NOUN
ejpam-6638	108	22	deblurring	deblurring	NOUN
ejpam-6638	108	23	as	as	ADP
ejpam-6638	108	24	part	part	NOUN
ejpam-6638	108	25	of	of	ADP
ejpam-6638	108	26	a	a	DET
ejpam-6638	108	27	linear	linear	ADJ
ejpam-6638	108	28	inverse	inverse	NOUN
ejpam-6638	108	29	problem	problem	NOUN
ejpam-6638	108	30	.	.	PUNCT
ejpam-6638	109	1	inverse	inverse	NOUN
ejpam-6638	109	2	problems	problem	NOUN
ejpam-6638	109	3	of	of	ADP
ejpam-6638	109	4	this	this	DET
ejpam-6638	109	5	form	form	NOUN
ejpam-6638	109	6	are	be	AUX
ejpam-6638	109	7	typically	typically	ADV
ejpam-6638	109	8	ill	ill	ADV
ejpam-6638	109	9	-	-	PUNCT
ejpam-6638	109	10	posed	pose	VERB
ejpam-6638	109	11	in	in	ADP
ejpam-6638	109	12	the	the	DET
ejpam-6638	109	13	sense	sense	NOUN
ejpam-6638	109	14	of	of	ADP
ejpam-6638	109	15	hadamard	hadamard	NOUN
ejpam-6638	109	16	and	and	CCONJ
ejpam-6638	109	17	require	require	VERB
ejpam-6638	109	18	the	the	DET
ejpam-6638	109	19	use	use	NOUN
ejpam-6638	109	20	of	of	ADP
ejpam-6638	109	21	regularization	regularization	NOUN
ejpam-6638	109	22	techniques	technique	NOUN
ejpam-6638	109	23	[	[	X
ejpam-6638	109	24	48	48	NUM
ejpam-6638	109	25	]	]	PUNCT
ejpam-6638	109	26	.	.	PUNCT
ejpam-6638	110	1	g	g	PROPN
ejpam-6638	110	2	=	=	PROPN
ejpam-6638	110	3	dv	dv	PROPN
ejpam-6638	110	4	+	+	CCONJ
ejpam-6638	110	5	γ	γ	X
ejpam-6638	110	6	(	(	PUNCT
ejpam-6638	110	7	2.1	2.1	NUM
ejpam-6638	110	8	)	)	PUNCT
ejpam-6638	110	9	the	the	DET
ejpam-6638	110	10	objective	objective	NOUN
ejpam-6638	110	11	is	be	AUX
ejpam-6638	110	12	to	to	PART
ejpam-6638	110	13	retrieve	retrieve	VERB
ejpam-6638	110	14	an	an	DET
ejpam-6638	110	15	unknown	unknown	ADJ
ejpam-6638	110	16	image	image	NOUN
ejpam-6638	110	17	v	v	ADP
ejpam-6638	110	18	∈	∈	PROPN
ejpam-6638	110	19	rn	rn	NOUN
ejpam-6638	110	20	using	use	VERB
ejpam-6638	110	21	the	the	DET
ejpam-6638	110	22	collected	collect	VERB
ejpam-6638	110	23	measurements	measurement	NOUN
ejpam-6638	110	24	g	g	PROPN
ejpam-6638	110	25	∈	∈	PROPN
ejpam-6638	110	26	rm	rm	PROPN
ejpam-6638	110	27	.	.	PUNCT
ejpam-6638	111	1	in	in	ADP
ejpam-6638	111	2	this	this	DET
ejpam-6638	111	3	scenario	scenario	NOUN
ejpam-6638	111	4	,	,	PUNCT
ejpam-6638	111	5	d	d	PROPN
ejpam-6638	111	6	∈	∈	PROPN
ejpam-6638	111	7	rm×n	rm×n	NOUN
ejpam-6638	111	8	denotes	denote	VERB
ejpam-6638	111	9	a	a	DET
ejpam-6638	111	10	degradation	degradation	NOUN
ejpam-6638	111	11	matrix	matrix	NOUN
ejpam-6638	111	12	,	,	PUNCT
ejpam-6638	111	13	while	while	SCONJ
ejpam-6638	111	14	γ	γ	PROPN
ejpam-6638	111	15	∈	∈	PROPN
ejpam-6638	111	16	rm	rm	NOUN
ejpam-6638	111	17	represents	represent	VERB
ejpam-6638	111	18	the	the	DET
ejpam-6638	111	19	measurement	measurement	NOUN
ejpam-6638	111	20	noise	noise	NOUN
ejpam-6638	111	21	,	,	PUNCT
ejpam-6638	111	22	which	which	PRON
ejpam-6638	111	23	is	be	AUX
ejpam-6638	111	24	assumed	assume	VERB
ejpam-6638	111	25	to	to	PART
ejpam-6638	111	26	follow	follow	VERB
ejpam-6638	111	27	an	an	DET
ejpam-6638	111	28	additive	additive	ADJ
ejpam-6638	111	29	white	white	ADJ
ejpam-6638	111	30	gaussian	gaussian	NOUN
ejpam-6638	111	31	(	(	PUNCT
ejpam-6638	111	32	awgn	awgn	NOUN
ejpam-6638	111	33	)	)	PUNCT
ejpam-6638	111	34	distribution	distribution	NOUN
ejpam-6638	111	35	with	with	ADP
ejpam-6638	111	36	a	a	DET
ejpam-6638	111	37	variance	variance	NOUN
ejpam-6638	111	38	of	of	ADP
ejpam-6638	111	39	σ2	σ2	PROPN
ejpam-6638	111	40	.	.	PUNCT
ejpam-6638	112	1	inverse	inverse	NOUN
ejpam-6638	112	2	problems	problem	NOUN
ejpam-6638	112	3	in	in	ADP
ejpam-6638	112	4	real	real	ADJ
ejpam-6638	112	5	-	-	PUNCT
ejpam-6638	112	6	world	world	NOUN
ejpam-6638	112	7	applications	application	NOUN
ejpam-6638	112	8	are	be	AUX
ejpam-6638	112	9	often	often	ADV
ejpam-6638	112	10	ill	ill	ADV
ejpam-6638	112	11	-	-	PUNCT
ejpam-6638	112	12	posed	pose	VERB
ejpam-6638	112	13	;	;	PUNCT
ejpam-6638	112	14	hence	hence	ADV
ejpam-6638	112	15	,	,	PUNCT
ejpam-6638	112	16	it	it	PRON
ejpam-6638	112	17	is	be	AUX
ejpam-6638	112	18	common	common	ADJ
ejpam-6638	112	19	to	to	PART
ejpam-6638	112	20	apply	apply	VERB
ejpam-6638	112	21	regularization	regularization	NOUN
ejpam-6638	112	22	to	to	ADP
ejpam-6638	112	23	the	the	DET
ejpam-6638	112	24	solution	solution	NOUN
ejpam-6638	112	25	utilizing	utilize	VERB
ejpam-6638	112	26	existing	exist	VERB
ejpam-6638	112	27	prior	prior	ADJ
ejpam-6638	112	28	knowledge	knowledge	NOUN
ejpam-6638	112	29	.	.	PUNCT
ejpam-6638	113	1	in	in	ADP
ejpam-6638	113	2	practical	practical	ADJ
ejpam-6638	113	3	scenarios	scenario	NOUN
ejpam-6638	113	4	,	,	PUNCT
ejpam-6638	113	5	the	the	DET
ejpam-6638	113	6	reconstruction	reconstruction	NOUN
ejpam-6638	113	7	usually	usually	ADV
ejpam-6638	113	8	relies	rely	VERB
ejpam-6638	113	9	on	on	ADP
ejpam-6638	113	10	a	a	DET
ejpam-6638	113	11	regularized	regularize	VERB
ejpam-6638	113	12	least	least	ADJ
ejpam-6638	113	13	-	-	PUNCT
ejpam-6638	113	14	squares	square	NOUN
ejpam-6638	113	15	method	method	NOUN
ejpam-6638	113	16	.	.	PUNCT
ejpam-6638	114	1	v∗	v∗	NOUN
ejpam-6638	114	2	=	=	PUNCT
ejpam-6638	114	3	argxmin||g	argxmin||g	NOUN
ejpam-6638	114	4	−	−	PROPN
ejpam-6638	114	5	dv||2l2	dv||2l2	PROPN
ejpam-6638	114	6	+	+	CCONJ
ejpam-6638	114	7	βη(v	βη(v	NUM
ejpam-6638	114	8	)	)	PUNCT
ejpam-6638	114	9	(	(	PUNCT
ejpam-6638	114	10	2.2	2.2	NUM
ejpam-6638	114	11	)	)	PUNCT
ejpam-6638	114	12	in	in	ADP
ejpam-6638	114	13	this	this	DET
ejpam-6638	114	14	setup	setup	NOUN
ejpam-6638	114	15	,	,	PUNCT
ejpam-6638	114	16	the	the	DET
ejpam-6638	114	17	data	datum	NOUN
ejpam-6638	114	18	fidelity	fidelity	PROPN
ejpam-6638	114	19	component	component	NOUN
ejpam-6638	114	20	ensures	ensure	VERB
ejpam-6638	114	21	consistency	consistency	NOUN
ejpam-6638	114	22	with	with	ADP
ejpam-6638	114	23	the	the	DET
ejpam-6638	114	24	measurements	measurement	NOUN
ejpam-6638	114	25	,	,	PUNCT
ejpam-6638	114	26	whereas	whereas	SCONJ
ejpam-6638	114	27	the	the	DET
ejpam-6638	114	28	regularizer	regularizer	NOUN
ejpam-6638	114	29	η	η	PROPN
ejpam-6638	114	30	confines	confine	VERB
ejpam-6638	114	31	the	the	DET
ejpam-6638	114	32	solution	solution	NOUN
ejpam-6638	114	33	to	to	ADP
ejpam-6638	114	34	the	the	DET
ejpam-6638	114	35	defined	define	VERB
ejpam-6638	114	36	image	image	NOUN
ejpam-6638	114	37	category	category	NOUN
ejpam-6638	114	38	.	.	PUNCT
ejpam-6638	115	1	the	the	DET
ejpam-6638	115	2	parameter	parameter	PROPN
ejpam-6638	115	3	β	β	PROPN
ejpam-6638	115	4	is	be	AUX
ejpam-6638	115	5	a	a	DET
ejpam-6638	115	6	positive	positive	ADJ
ejpam-6638	115	7	value	value	NOUN
ejpam-6638	115	8	that	that	PRON
ejpam-6638	115	9	controls	control	VERB
ejpam-6638	115	10	the	the	DET
ejpam-6638	115	11	strength	strength	NOUN
ejpam-6638	115	12	of	of	ADP
ejpam-6638	115	13	the	the	DET
ejpam-6638	115	14	regularization	regularization	NOUN
ejpam-6638	115	15	.	.	PUNCT
ejpam-6638	116	1	at	at	ADP
ejpam-6638	116	2	present	present	ADJ
ejpam-6638	116	3	,	,	PUNCT
ejpam-6638	116	4	deep	deep	ADJ
ejpam-6638	116	5	learning	learning	NOUN
ejpam-6638	116	6	reaches	reach	VERB
ejpam-6638	116	7	cutting	cutting	NOUN
ejpam-6638	116	8	-	-	PUNCT
ejpam-6638	116	9	edge	edge	NOUN
ejpam-6638	116	10	results	result	NOUN
ejpam-6638	116	11	for	for	ADP
ejpam-6638	116	12	various	various	ADJ
ejpam-6638	116	13	issues	issue	NOUN
ejpam-6638	116	14	related	relate	VERB
ejpam-6638	116	15	to	to	ADP
ejpam-6638	116	16	image	image	NOUN
ejpam-6638	116	17	restoration	restoration	NOUN
ejpam-6638	116	18	[	[	X
ejpam-6638	116	19	49	49	NUM
ejpam-6638	116	20	]	]	PUNCT
ejpam-6638	116	21	.	.	PUNCT
ejpam-6638	117	1	the	the	DET
ejpam-6638	117	2	fundamental	fundamental	ADJ
ejpam-6638	117	3	concept	concept	NOUN
ejpam-6638	117	4	involves	involve	VERB
ejpam-6638	117	5	training	train	VERB
ejpam-6638	117	6	a	a	DET
ejpam-6638	117	7	cnn	cnn	NOUN
ejpam-6638	117	8	through	through	ADP
ejpam-6638	117	9	the	the	DET
ejpam-6638	117	10	subsequent	subsequent	ADJ
ejpam-6638	117	11	optimization	optimization	NOUN
ejpam-6638	117	12	.	.	PUNCT
ejpam-6638	118	1	θ∗	θ∗	NOUN
ejpam-6638	118	2	=	=	SYM
ejpam-6638	118	3	argθminl(fθ(g	argθminl(fθ(g	X
ejpam-6638	118	4	)	)	PUNCT
ejpam-6638	118	5	,	,	PUNCT
ejpam-6638	118	6	v	v	NOUN
ejpam-6638	118	7	)	)	PUNCT
ejpam-6638	118	8	(	(	PUNCT
ejpam-6638	118	9	2.3	2.3	NUM
ejpam-6638	118	10	)	)	PUNCT
ejpam-6638	118	11	where	where	SCONJ
ejpam-6638	118	12	fθ	fθ	ADJ
ejpam-6638	118	13	(	(	PUNCT
ejpam-6638	118	14	.	.	PUNCT
ejpam-6638	118	15	)	)	PUNCT
ejpam-6638	118	16	denotes	denote	VERB
ejpam-6638	118	17	the	the	DET
ejpam-6638	118	18	convolutional	convolutional	ADJ
ejpam-6638	118	19	neural	neural	ADJ
ejpam-6638	118	20	network	network	NOUN
ejpam-6638	118	21	characterized	characterize	VERB
ejpam-6638	118	22	by	by	ADP
ejpam-6638	118	23	the	the	DET
ejpam-6638	118	24	parameters	parameter	NOUN
ejpam-6638	118	25	θ	θ	PROPN
ejpam-6638	118	26	.	.	PUNCT
ejpam-6638	119	1	the	the	DET
ejpam-6638	119	2	symbol	symbol	NOUN
ejpam-6638	119	3	l	l	NOUN
ejpam-6638	119	4	signifies	signify	VERB
ejpam-6638	119	5	the	the	DET
ejpam-6638	119	6	loss	loss	NOUN
ejpam-6638	119	7	function	function	NOUN
ejpam-6638	119	8	.	.	PUNCT
ejpam-6638	120	1	in	in	ADP
ejpam-6638	120	2	practical	practical	ADJ
ejpam-6638	120	3	applications	application	NOUN
ejpam-6638	120	4	,	,	PUNCT
ejpam-6638	120	5	(	(	PUNCT
ejpam-6638	120	6	2.3	2.3	NUM
ejpam-6638	120	7	)	)	PUNCT
ejpam-6638	120	8	can	can	AUX
ejpam-6638	120	9	be	be	AUX
ejpam-6638	120	10	efficiently	efficiently	ADV
ejpam-6638	120	11	optimized	optimize	VERB
ejpam-6638	120	12	by	by	ADP
ejpam-6638	120	13	utilizing	utilize	VERB
ejpam-6638	120	14	various	various	ADJ
ejpam-6638	120	15	methods	method	NOUN
ejpam-6638	120	16	from	from	ADP
ejpam-6638	120	17	the	the	DET
ejpam-6638	120	18	family	family	NOUN
ejpam-6638	120	19	of	of	ADP
ejpam-6638	120	20	stochastic	stochastic	ADJ
ejpam-6638	120	21	gradient	gradient	ADJ
ejpam-6638	120	22	descent	descent	NOUN
ejpam-6638	120	23	(	(	PUNCT
ejpam-6638	120	24	sgd	sgd	PROPN
ejpam-6638	120	25	)	)	PUNCT
ejpam-6638	120	26	,	,	PUNCT
ejpam-6638	120	27	including	include	VERB
ejpam-6638	120	28	adaptive	adaptive	ADJ
ejpam-6638	120	29	moment	moment	NOUN
ejpam-6638	120	30	estimation	estimation	NOUN
ejpam-6638	120	31	(	(	PUNCT
ejpam-6638	120	32	adam	adam	PROPN
ejpam-6638	120	33	)	)	PUNCT
ejpam-6638	121	1	[	[	X
ejpam-6638	121	2	50	50	NUM
ejpam-6638	121	3	]	]	PUNCT
ejpam-6638	121	4	.	.	PUNCT
ejpam-6638	122	1	recently	recently	ADV
ejpam-6638	122	2	,	,	PUNCT
ejpam-6638	122	3	[	[	X
ejpam-6638	122	4	51	51	NUM
ejpam-6638	122	5	]	]	PUNCT
ejpam-6638	122	6	proposed	propose	VERB
ejpam-6638	122	7	a	a	DET
ejpam-6638	122	8	different	different	ADJ
ejpam-6638	122	9	strategy	strategy	NOUN
ejpam-6638	122	10	that	that	PRON
ejpam-6638	122	11	utilizes	utilize	VERB
ejpam-6638	122	12	cnn	cnn	PROPN
ejpam-6638	122	13	-	-	PUNCT
ejpam-6638	122	14	based	base	VERB
ejpam-6638	122	15	techniques	technique	NOUN
ejpam-6638	122	16	.	.	PUNCT
ejpam-6638	123	1	their	their	PRON
ejpam-6638	123	2	research	research	NOUN
ejpam-6638	123	3	indicated	indicate	VERB
ejpam-6638	123	4	that	that	SCONJ
ejpam-6638	123	5	deep	deep	ADJ
ejpam-6638	123	6	cnn	cnn	PROPN
ejpam-6638	123	7	models	model	NOUN
ejpam-6638	123	8	excel	excel	VERB
ejpam-6638	123	9	at	at	ADP
ejpam-6638	123	10	understanding	understand	VERB
ejpam-6638	123	11	the	the	DET
ejpam-6638	123	12	organization	organization	NOUN
ejpam-6638	123	13	of	of	ADP
ejpam-6638	123	14	natural	natural	ADJ
ejpam-6638	123	15	images	image	NOUN
ejpam-6638	123	16	,	,	PUNCT
ejpam-6638	123	17	yet	yet	CCONJ
ejpam-6638	123	18	they	they	PRON
ejpam-6638	123	19	struggle	struggle	VERB
ejpam-6638	123	20	with	with	ADP
ejpam-6638	123	21	random	random	ADJ
ejpam-6638	123	22	noise	noise	NOUN
ejpam-6638	123	23	.	.	PUNCT
ejpam-6638	124	1	when	when	SCONJ
ejpam-6638	124	2	presented	present	VERB
ejpam-6638	124	3	with	with	ADP
ejpam-6638	124	4	a	a	DET
ejpam-6638	124	5	random	random	ADJ
ejpam-6638	124	6	input	input	NOUN
ejpam-6638	124	7	vector	vector	NOUN
ejpam-6638	124	8	,	,	PUNCT
ejpam-6638	124	9	a	a	DET
ejpam-6638	124	10	cnn	cnn	PROPN
ejpam-6638	124	11	is	be	AUX
ejpam-6638	124	12	capable	capable	ADJ
ejpam-6638	124	13	of	of	ADP
ejpam-6638	124	14	producing	produce	VERB
ejpam-6638	124	15	a	a	DET
ejpam-6638	124	16	distinct	distinct	ADJ
ejpam-6638	124	17	image	image	NOUN
ejpam-6638	124	18	without	without	ADP
ejpam-6638	124	19	requiring	require	VERB
ejpam-6638	124	20	extensive	extensive	ADJ
ejpam-6638	124	21	supervised	supervised	ADJ
ejpam-6638	124	22	training	training	NOUN
ejpam-6638	124	23	m.	m.	NOUN
ejpam-6638	124	24	israr	israr	PROPN
ejpam-6638	124	25	et	et	PROPN
ejpam-6638	124	26	al	al	PROPN
ejpam-6638	124	27	.	.	PUNCT
ejpam-6638	124	28	/	/	SYM
ejpam-6638	124	29	eur	eur	PROPN
ejpam-6638	124	30	.	.	PUNCT
ejpam-6638	125	1	j.	j.	PROPN
ejpam-6638	125	2	pure	pure	PROPN
ejpam-6638	125	3	appl	appl	PROPN
ejpam-6638	125	4	.	.	PROPN
ejpam-6638	125	5	math	math	PROPN
ejpam-6638	125	6	,	,	PUNCT
ejpam-6638	125	7	18	18	NUM
ejpam-6638	125	8	(	(	PUNCT
ejpam-6638	125	9	4	4	NUM
ejpam-6638	125	10	)	)	PUNCT
ejpam-6638	125	11	(	(	PUNCT
ejpam-6638	125	12	2025	2025	NUM
ejpam-6638	125	13	)	)	PUNCT
ejpam-6638	125	14	,	,	PUNCT
ejpam-6638	125	15	6638	6638	NUM
ejpam-6638	125	16	5	5	NUM
ejpam-6638	125	17	of	of	ADP
ejpam-6638	125	18	22	22	NUM
ejpam-6638	125	19	on	on	ADP
ejpam-6638	125	20	a	a	DET
ejpam-6638	125	21	large	large	ADJ
ejpam-6638	125	22	dataset	dataset	NOUN
ejpam-6638	125	23	.	.	PUNCT
ejpam-6638	126	1	regarding	regard	VERB
ejpam-6638	126	2	image	image	NOUN
ejpam-6638	126	3	restoration	restoration	NOUN
ejpam-6638	126	4	,	,	PUNCT
ejpam-6638	126	5	the	the	DET
ejpam-6638	126	6	optimization	optimization	NOUN
ejpam-6638	126	7	associated	associate	VERB
ejpam-6638	126	8	with	with	ADP
ejpam-6638	126	9	dip	dip	NOUN
ejpam-6638	126	10	can	can	AUX
ejpam-6638	126	11	be	be	AUX
ejpam-6638	126	12	expressed	express	VERB
ejpam-6638	126	13	as	as	ADP
ejpam-6638	126	14	θ∗	θ∗	NOUN
ejpam-6638	126	15	=	=	PROPN
ejpam-6638	126	16	argθmin||g	argθmin||g	PROPN
ejpam-6638	126	17	−hfθ(x)||2l2	−hfθ(x)||2l2	PROPN
ejpam-6638	126	18	(	(	PUNCT
ejpam-6638	126	19	2.4	2.4	NUM
ejpam-6638	126	20	)	)	PUNCT
ejpam-6638	126	21	such	such	ADJ
ejpam-6638	126	22	that	that	DET
ejpam-6638	126	23	v∗	v∗	ADJ
ejpam-6638	126	24	=	=	PUNCT
ejpam-6638	126	25	fθ∗(x	fθ∗(x	NOUN
ejpam-6638	126	26	)	)	PUNCT
ejpam-6638	126	27	.	.	PUNCT
ejpam-6638	127	1	where	where	SCONJ
ejpam-6638	127	2	x	x	SYM
ejpam-6638	127	3	∈	∈	PROPN
ejpam-6638	127	4	rn	rn	PROPN
ejpam-6638	127	5	denotes	denote	VERB
ejpam-6638	127	6	the	the	DET
ejpam-6638	127	7	random	random	ADJ
ejpam-6638	127	8	input	input	NOUN
ejpam-6638	127	9	vector	vector	NOUN
ejpam-6638	127	10	.	.	PUNCT
ejpam-6638	128	1	the	the	DET
ejpam-6638	128	2	cnn	cnn	PROPN
ejpam-6638	128	3	generator	generator	NOUN
ejpam-6638	128	4	is	be	AUX
ejpam-6638	128	5	improved	improve	VERB
ejpam-6638	128	6	by	by	ADP
ejpam-6638	128	7	minimizing	minimize	VERB
ejpam-6638	128	8	the	the	DET
ejpam-6638	128	9	random	random	ADJ
ejpam-6638	128	10	variable	variable	ADJ
ejpam-6638	128	11	θ	θ	NOUN
ejpam-6638	128	12	,	,	PUNCT
ejpam-6638	128	13	and	and	CCONJ
ejpam-6638	128	14	these	these	DET
ejpam-6638	128	15	variables	variable	NOUN
ejpam-6638	128	16	are	be	AUX
ejpam-6638	128	17	modified	modify	VERB
ejpam-6638	128	18	in	in	ADP
ejpam-6638	128	19	an	an	DET
ejpam-6638	128	20	iterative	iterative	NOUN
ejpam-6638	128	21	manner	manner	NOUN
ejpam-6638	128	22	to	to	PART
ejpam-6638	128	23	guarantee	guarantee	VERB
ejpam-6638	128	24	that	that	SCONJ
ejpam-6638	128	25	the	the	DET
ejpam-6638	128	26	network	network	NOUN
ejpam-6638	128	27	’s	’s	PART
ejpam-6638	128	28	output	output	NOUN
ejpam-6638	128	29	closely	closely	ADV
ejpam-6638	128	30	corresponds	correspond	VERB
ejpam-6638	128	31	with	with	ADP
ejpam-6638	128	32	the	the	DET
ejpam-6638	128	33	desired	desire	VERB
ejpam-6638	128	34	measurement	measurement	NOUN
ejpam-6638	128	35	.	.	PUNCT
ejpam-6638	129	1	3	3	X
ejpam-6638	129	2	.	.	NUM
ejpam-6638	129	3	proposed	propose	VERB
ejpam-6638	129	4	method	method	NOUN
ejpam-6638	129	5	:	:	PUNCT
ejpam-6638	129	6	deep	deep	ADJ
ejpam-6638	129	7	image	image	NOUN
ejpam-6638	129	8	prior	prior	ADV
ejpam-6638	129	9	with	with	ADP
ejpam-6638	129	10	svd	svd	PROPN
ejpam-6638	129	11	pooling	pool	VERB
ejpam-6638	129	12	in	in	ADP
ejpam-6638	129	13	this	this	DET
ejpam-6638	129	14	section	section	NOUN
ejpam-6638	129	15	,	,	PUNCT
ejpam-6638	129	16	we	we	PRON
ejpam-6638	129	17	introduce	introduce	VERB
ejpam-6638	129	18	our	our	PRON
ejpam-6638	129	19	proposed	propose	VERB
ejpam-6638	129	20	framework	framework	NOUN
ejpam-6638	129	21	for	for	ADP
ejpam-6638	129	22	image	image	NOUN
ejpam-6638	129	23	restoration	restoration	NOUN
ejpam-6638	129	24	that	that	PRON
ejpam-6638	129	25	builds	build	VERB
ejpam-6638	129	26	on	on	ADP
ejpam-6638	129	27	the	the	DET
ejpam-6638	129	28	deep	deep	ADJ
ejpam-6638	129	29	image	image	NOUN
ejpam-6638	129	30	prior	prior	ADV
ejpam-6638	129	31	(	(	PUNCT
ejpam-6638	129	32	dip	dip	NOUN
ejpam-6638	129	33	)	)	PUNCT
ejpam-6638	129	34	architecture	architecture	NOUN
ejpam-6638	129	35	,	,	PUNCT
ejpam-6638	129	36	enhanced	enhance	VERB
ejpam-6638	129	37	with	with	ADP
ejpam-6638	129	38	singular	singular	ADJ
ejpam-6638	129	39	value	value	NOUN
ejpam-6638	129	40	decomposition	decomposition	NOUN
ejpam-6638	129	41	(	(	PUNCT
ejpam-6638	129	42	svd	svd	PROPN
ejpam-6638	129	43	)	)	PUNCT
ejpam-6638	129	44	pooling	pooling	NOUN
ejpam-6638	129	45	.	.	PUNCT
ejpam-6638	130	1	the	the	DET
ejpam-6638	130	2	main	main	ADJ
ejpam-6638	130	3	goal	goal	NOUN
ejpam-6638	130	4	of	of	ADP
ejpam-6638	130	5	this	this	DET
ejpam-6638	130	6	enhancement	enhancement	NOUN
ejpam-6638	130	7	is	be	AUX
ejpam-6638	130	8	to	to	PART
ejpam-6638	130	9	take	take	VERB
ejpam-6638	130	10	advantage	advantage	NOUN
ejpam-6638	130	11	of	of	ADP
ejpam-6638	130	12	the	the	DET
ejpam-6638	130	13	natural	natural	ADJ
ejpam-6638	130	14	lowrank	lowrank	ADJ
ejpam-6638	130	15	structure	structure	NOUN
ejpam-6638	130	16	found	find	VERB
ejpam-6638	130	17	in	in	ADP
ejpam-6638	130	18	images	image	NOUN
ejpam-6638	130	19	and	and	CCONJ
ejpam-6638	130	20	to	to	PART
ejpam-6638	130	21	further	far	ADV
ejpam-6638	130	22	regularize	regularize	VERB
ejpam-6638	130	23	the	the	DET
ejpam-6638	130	24	learning	learning	NOUN
ejpam-6638	130	25	process	process	NOUN
ejpam-6638	130	26	by	by	ADP
ejpam-6638	130	27	using	use	VERB
ejpam-6638	130	28	global	global	ADJ
ejpam-6638	130	29	structural	structural	ADJ
ejpam-6638	130	30	information	information	NOUN
ejpam-6638	130	31	through	through	ADP
ejpam-6638	130	32	svd	svd	PROPN
ejpam-6638	130	33	.	.	PROPN
ejpam-6638	130	34	3.1	3.1	NUM
ejpam-6638	130	35	.	.	PUNCT
ejpam-6638	131	1	deep	deep	ADJ
ejpam-6638	131	2	image	image	NOUN
ejpam-6638	131	3	prior	prior	ADJ
ejpam-6638	131	4	framework	framework	NOUN
ejpam-6638	131	5	by	by	ADP
ejpam-6638	131	6	using	use	VERB
ejpam-6638	131	7	the	the	DET
ejpam-6638	131	8	architecture	architecture	NOUN
ejpam-6638	131	9	of	of	ADP
ejpam-6638	131	10	convolutional	convolutional	ADJ
ejpam-6638	131	11	neural	neural	ADJ
ejpam-6638	131	12	networks	network	NOUN
ejpam-6638	131	13	(	(	PUNCT
ejpam-6638	131	14	cnns	cnns	PROPN
ejpam-6638	131	15	)	)	PUNCT
ejpam-6638	131	16	as	as	ADP
ejpam-6638	131	17	an	an	DET
ejpam-6638	131	18	intrinsic	intrinsic	ADJ
ejpam-6638	131	19	image	image	NOUN
ejpam-6638	131	20	prior	prior	ADV
ejpam-6638	131	21	,	,	PUNCT
ejpam-6638	131	22	the	the	DET
ejpam-6638	131	23	deep	deep	ADJ
ejpam-6638	131	24	image	image	NOUN
ejpam-6638	131	25	prior	prior	ADV
ejpam-6638	131	26	(	(	PUNCT
ejpam-6638	131	27	dip	dip	NOUN
ejpam-6638	131	28	)	)	PUNCT
ejpam-6638	131	29	technique	technique	NOUN
ejpam-6638	131	30	,	,	PUNCT
ejpam-6638	131	31	which	which	PRON
ejpam-6638	131	32	was	be	AUX
ejpam-6638	131	33	first	first	ADV
ejpam-6638	131	34	presented	present	VERB
ejpam-6638	131	35	by	by	ADP
ejpam-6638	131	36	ulyanov	ulyanov	PROPN
ejpam-6638	131	37	et	et	PROPN
ejpam-6638	131	38	al	al	PROPN
ejpam-6638	131	39	.	.	PUNCT
ejpam-6638	132	1	[	[	X
ejpam-6638	132	2	51	51	NUM
ejpam-6638	132	3	]	]	PUNCT
ejpam-6638	132	4	,	,	PUNCT
ejpam-6638	132	5	does	do	AUX
ejpam-6638	132	6	not	not	PART
ejpam-6638	132	7	require	require	VERB
ejpam-6638	132	8	any	any	DET
ejpam-6638	132	9	additional	additional	ADJ
ejpam-6638	132	10	training	training	NOUN
ejpam-6638	132	11	data	datum	NOUN
ejpam-6638	132	12	.	.	PUNCT
ejpam-6638	133	1	a	a	DET
ejpam-6638	133	2	cnn	cnn	NOUN
ejpam-6638	133	3	that	that	PRON
ejpam-6638	133	4	has	have	AUX
ejpam-6638	133	5	been	be	AUX
ejpam-6638	133	6	randomly	randomly	ADV
ejpam-6638	133	7	initialized	initialize	VERB
ejpam-6638	133	8	is	be	AUX
ejpam-6638	133	9	used	use	VERB
ejpam-6638	133	10	in	in	ADP
ejpam-6638	133	11	the	the	DET
ejpam-6638	133	12	basic	basic	ADJ
ejpam-6638	133	13	dip	dip	NOUN
ejpam-6638	133	14	design	design	NOUN
ejpam-6638	133	15	to	to	PART
ejpam-6638	133	16	convert	convert	VERB
ejpam-6638	133	17	a	a	DET
ejpam-6638	133	18	constant	constant	ADJ
ejpam-6638	133	19	input	input	NOUN
ejpam-6638	133	20	(	(	PUNCT
ejpam-6638	133	21	like	like	ADP
ejpam-6638	133	22	uniform	uniform	ADJ
ejpam-6638	133	23	noise	noise	NOUN
ejpam-6638	133	24	)	)	PUNCT
ejpam-6638	133	25	into	into	ADP
ejpam-6638	133	26	a	a	DET
ejpam-6638	133	27	target	target	NOUN
ejpam-6638	133	28	image	image	NOUN
ejpam-6638	133	29	through	through	ADP
ejpam-6638	133	30	an	an	DET
ejpam-6638	133	31	optimization	optimization	NOUN
ejpam-6638	133	32	process	process	NOUN
ejpam-6638	133	33	that	that	PRON
ejpam-6638	133	34	is	be	AUX
ejpam-6638	133	35	motivated	motivate	VERB
ejpam-6638	133	36	by	by	ADP
ejpam-6638	133	37	a	a	DET
ejpam-6638	133	38	reconstruction	reconstruction	NOUN
ejpam-6638	133	39	loss	loss	NOUN
ejpam-6638	133	40	,	,	PUNCT
ejpam-6638	133	41	which	which	PRON
ejpam-6638	133	42	is	be	AUX
ejpam-6638	133	43	frequently	frequently	ADV
ejpam-6638	133	44	the	the	DET
ejpam-6638	133	45	mean	mean	ADJ
ejpam-6638	133	46	squared	square	VERB
ejpam-6638	133	47	error	error	NOUN
ejpam-6638	133	48	(	(	PUNCT
ejpam-6638	133	49	mse	mse	NOUN
ejpam-6638	133	50	)	)	PUNCT
ejpam-6638	133	51	between	between	ADP
ejpam-6638	133	52	the	the	DET
ejpam-6638	133	53	network	network	NOUN
ejpam-6638	133	54	’s	’s	PART
ejpam-6638	133	55	output	output	NOUN
ejpam-6638	133	56	and	and	CCONJ
ejpam-6638	133	57	the	the	DET
ejpam-6638	133	58	corrupted	corrupted	ADJ
ejpam-6638	133	59	input	input	NOUN
ejpam-6638	133	60	.	.	PUNCT
ejpam-6638	134	1	multiscale	multiscale	ADJ
ejpam-6638	134	2	attention	attention	NOUN
ejpam-6638	134	3	and	and	CCONJ
ejpam-6638	134	4	svd	svd	PROPN
ejpam-6638	134	5	pooling	pool	VERB
ejpam-6638	134	6	enhance	enhance	NOUN
ejpam-6638	134	7	feature	feature	NOUN
ejpam-6638	134	8	aggregation	aggregation	NOUN
ejpam-6638	134	9	in	in	ADP
ejpam-6638	134	10	fig	fig	NOUN
ejpam-6638	134	11	.	.	PUNCT
ejpam-6638	135	1	1.(c	1.(c	NOUN
ejpam-6638	135	2	)	)	PUNCT
ejpam-6638	135	3	,	,	PUNCT
ejpam-6638	135	4	while	while	SCONJ
ejpam-6638	135	5	an	an	DET
ejpam-6638	135	6	encoder	encoder	NOUN
ejpam-6638	135	7	-	-	PUNCT
ejpam-6638	135	8	decoder	decoder	NOUN
ejpam-6638	135	9	cnn	cnn	PROPN
ejpam-6638	135	10	with	with	ADP
ejpam-6638	135	11	skip	skip	ADJ
ejpam-6638	135	12	connections	connection	NOUN
ejpam-6638	135	13	preserves	preserve	VERB
ejpam-6638	135	14	image	image	NOUN
ejpam-6638	135	15	structure	structure	NOUN
ejpam-6638	135	16	.	.	PUNCT
ejpam-6638	136	1	untrained	untrained	ADJ
ejpam-6638	136	2	convolutional	convolutional	ADJ
ejpam-6638	136	3	layers	layer	NOUN
ejpam-6638	136	4	,	,	PUNCT
ejpam-6638	136	5	like	like	ADP
ejpam-6638	136	6	3×3	3×3	NUM
ejpam-6638	136	7	and	and	CCONJ
ejpam-6638	136	8	1×1	1×1	NUM
ejpam-6638	136	9	filters	filter	NOUN
ejpam-6638	136	10	with	with	ADP
ejpam-6638	136	11	batch	batch	NOUN
ejpam-6638	136	12	normalization	normalization	NOUN
ejpam-6638	136	13	,	,	PUNCT
ejpam-6638	136	14	serve	serve	VERB
ejpam-6638	136	15	as	as	ADP
ejpam-6638	136	16	an	an	DET
ejpam-6638	136	17	implicit	implicit	ADJ
ejpam-6638	136	18	prior	prior	ADV
ejpam-6638	136	19	during	during	ADP
ejpam-6638	136	20	optimization	optimization	NOUN
ejpam-6638	136	21	.	.	PUNCT
ejpam-6638	137	1	this	this	DET
ejpam-6638	137	2	technique	technique	NOUN
ejpam-6638	137	3	allows	allow	VERB
ejpam-6638	137	4	for	for	ADP
ejpam-6638	137	5	data	data	NOUN
ejpam-6638	137	6	-	-	PUNCT
ejpam-6638	137	7	free	free	ADJ
ejpam-6638	137	8	reconstruction	reconstruction	NOUN
ejpam-6638	137	9	by	by	ADP
ejpam-6638	137	10	matching	match	VERB
ejpam-6638	137	11	the	the	DET
ejpam-6638	137	12	network	network	NOUN
ejpam-6638	137	13	output	output	NOUN
ejpam-6638	137	14	to	to	ADP
ejpam-6638	137	15	defective	defective	ADJ
ejpam-6638	137	16	inputs	input	NOUN
ejpam-6638	137	17	via	via	ADP
ejpam-6638	137	18	gradient	gradient	ADJ
ejpam-6638	137	19	descent	descent	NOUN
ejpam-6638	137	20	.	.	PUNCT
ejpam-6638	138	1	let	let	VERB
ejpam-6638	138	2	x	x	PRON
ejpam-6638	138	3	represent	represent	VERB
ejpam-6638	138	4	the	the	DET
ejpam-6638	138	5	clean	clean	ADJ
ejpam-6638	138	6	image	image	NOUN
ejpam-6638	138	7	,	,	PUNCT
ejpam-6638	138	8	y	y	PRON
ejpam-6638	138	9	signify	signify	VERB
ejpam-6638	138	10	the	the	DET
ejpam-6638	138	11	observed	observe	VERB
ejpam-6638	138	12	noisy	noisy	ADJ
ejpam-6638	138	13	or	or	CCONJ
ejpam-6638	138	14	incomplete	incomplete	ADJ
ejpam-6638	138	15	image	image	NOUN
ejpam-6638	138	16	,	,	PUNCT
ejpam-6638	138	17	and	and	CCONJ
ejpam-6638	138	18	fθ(z	fθ(z	X
ejpam-6638	138	19	)	)	PUNCT
ejpam-6638	138	20	be	be	VERB
ejpam-6638	138	21	the	the	DET
ejpam-6638	138	22	output	output	NOUN
ejpam-6638	138	23	from	from	ADP
ejpam-6638	138	24	the	the	DET
ejpam-6638	138	25	dip	dip	NOUN
ejpam-6638	138	26	network	network	NOUN
ejpam-6638	138	27	with	with	ADP
ejpam-6638	138	28	parameters	parameter	NOUN
ejpam-6638	138	29	θ	θ	PROPN
ejpam-6638	138	30	and	and	CCONJ
ejpam-6638	138	31	a	a	DET
ejpam-6638	138	32	fixed	fix	VERB
ejpam-6638	138	33	input	input	NOUN
ejpam-6638	138	34	z.	z.	PROPN
ejpam-6638	139	1	the	the	DET
ejpam-6638	139	2	objective	objective	ADJ
ejpam-6638	139	3	function	function	NOUN
ejpam-6638	139	4	is	be	AUX
ejpam-6638	139	5	expressed	express	VERB
ejpam-6638	139	6	as	as	ADP
ejpam-6638	139	7	:	:	PUNCT
ejpam-6638	139	8	lmse	lmse	NOUN
ejpam-6638	139	9	=	=	PUNCT
ejpam-6638	139	10	∥fθ(z)−	∥fθ(z)−	PUNCT
ejpam-6638	139	11	y∥22	y∥22	PROPN
ejpam-6638	139	12	.	.	PUNCT
ejpam-6638	140	1	(	(	PUNCT
ejpam-6638	140	2	3.1	3.1	NUM
ejpam-6638	140	3	)	)	PUNCT
ejpam-6638	140	4	even	even	ADV
ejpam-6638	140	5	with	with	ADP
ejpam-6638	140	6	its	its	PRON
ejpam-6638	140	7	straightforward	straightforward	ADJ
ejpam-6638	140	8	approach	approach	NOUN
ejpam-6638	140	9	,	,	PUNCT
ejpam-6638	140	10	dip	dip	NOUN
ejpam-6638	140	11	has	have	AUX
ejpam-6638	140	12	shown	show	VERB
ejpam-6638	140	13	remarkable	remarkable	ADJ
ejpam-6638	140	14	effectiveness	effectiveness	NOUN
ejpam-6638	140	15	in	in	ADP
ejpam-6638	140	16	tasks	task	NOUN
ejpam-6638	140	17	such	such	ADJ
ejpam-6638	140	18	as	as	ADP
ejpam-6638	140	19	denoising	denoise	VERB
ejpam-6638	140	20	,	,	PUNCT
ejpam-6638	140	21	inpainting	inpainting	ADJ
ejpam-6638	140	22	,	,	PUNCT
ejpam-6638	140	23	and	and	CCONJ
ejpam-6638	140	24	super	super	NOUN
ejpam-6638	140	25	-	-	NOUN
ejpam-6638	140	26	resolution	resolution	NOUN
ejpam-6638	140	27	.	.	PUNCT
ejpam-6638	141	1	3.2	3.2	NUM
ejpam-6638	141	2	.	.	PUNCT
ejpam-6638	142	1	svd	svd	PROPN
ejpam-6638	142	2	pooling	pool	VERB
ejpam-6638	142	3	mechanism	mechanism	NOUN
ejpam-6638	142	4	to	to	PART
ejpam-6638	142	5	enhance	enhance	VERB
ejpam-6638	142	6	the	the	DET
ejpam-6638	142	7	capacity	capacity	NOUN
ejpam-6638	142	8	of	of	ADP
ejpam-6638	142	9	dip	dip	NOUN
ejpam-6638	142	10	in	in	ADP
ejpam-6638	142	11	capturing	capture	VERB
ejpam-6638	142	12	global	global	ADJ
ejpam-6638	142	13	context	context	NOUN
ejpam-6638	142	14	and	and	CCONJ
ejpam-6638	142	15	structural	structural	ADJ
ejpam-6638	142	16	coherence	coherence	NOUN
ejpam-6638	142	17	,	,	PUNCT
ejpam-6638	142	18	we	we	PRON
ejpam-6638	142	19	propose	propose	VERB
ejpam-6638	142	20	to	to	PART
ejpam-6638	142	21	replace	replace	VERB
ejpam-6638	142	22	standard	standard	ADJ
ejpam-6638	142	23	max	max	PROPN
ejpam-6638	142	24	or	or	CCONJ
ejpam-6638	142	25	average	average	ADJ
ejpam-6638	142	26	pooling	pool	VERB
ejpam-6638	142	27	operations	operation	NOUN
ejpam-6638	142	28	with	with	ADP
ejpam-6638	142	29	svd	svd	PROPN
ejpam-6638	142	30	pooling	pool	VERB
ejpam-6638	142	31	blocks	block	NOUN
ejpam-6638	142	32	.	.	PUNCT
ejpam-6638	143	1	the	the	DET
ejpam-6638	143	2	svd	svd	PROPN
ejpam-6638	143	3	pooling	pool	VERB
ejpam-6638	143	4	module	module	NOUN
ejpam-6638	143	5	acts	act	NOUN
ejpam-6638	143	6	on	on	ADP
ejpam-6638	143	7	image	image	NOUN
ejpam-6638	143	8	feature	feature	NOUN
ejpam-6638	143	9	maps	map	NOUN
ejpam-6638	143	10	by	by	ADP
ejpam-6638	143	11	performing	perform	VERB
ejpam-6638	143	12	singular	singular	ADJ
ejpam-6638	143	13	value	value	NOUN
ejpam-6638	143	14	decomposition	decomposition	NOUN
ejpam-6638	143	15	on	on	ADP
ejpam-6638	143	16	localized	localized	ADJ
ejpam-6638	143	17	patches	patch	NOUN
ejpam-6638	143	18	or	or	CCONJ
ejpam-6638	143	19	the	the	DET
ejpam-6638	143	20	entire	entire	ADJ
ejpam-6638	143	21	feature	feature	NOUN
ejpam-6638	143	22	map	map	NOUN
ejpam-6638	143	23	.	.	PUNCT
ejpam-6638	144	1	to	to	PART
ejpam-6638	144	2	perform	perform	VERB
ejpam-6638	144	3	singular	singular	ADJ
ejpam-6638	144	4	value	value	NOUN
ejpam-6638	144	5	decomposition	decomposition	NOUN
ejpam-6638	144	6	(	(	PUNCT
ejpam-6638	144	7	svd	svd	PROPN
ejpam-6638	144	8	)	)	PUNCT
ejpam-6638	144	9	,	,	PUNCT
ejpam-6638	144	10	we	we	PRON
ejpam-6638	144	11	transform	transform	VERB
ejpam-6638	144	12	a	a	DET
ejpam-6638	144	13	feature	feature	NOUN
ejpam-6638	144	14	map	map	NOUN
ejpam-6638	144	15	tensor	tensor	NOUN
ejpam-6638	144	16	f	f	PROPN
ejpam-6638	144	17	∈	∈	PROPN
ejpam-6638	144	18	rc×h×w	rc×h×w	VERB
ejpam-6638	144	19	into	into	ADP
ejpam-6638	144	20	a	a	DET
ejpam-6638	144	21	matrix	matrix	NOUN
ejpam-6638	144	22	m	m	NOUN
ejpam-6638	144	23	∈	∈	NOUN
ejpam-6638	144	24	rc×(h·w	rc×(h·w	NOUN
ejpam-6638	144	25	)	)	PUNCT
ejpam-6638	144	26	.	.	PUNCT
ejpam-6638	145	1	in	in	ADP
ejpam-6638	145	2	this	this	DET
ejpam-6638	145	3	context	context	NOUN
ejpam-6638	145	4	,	,	PUNCT
ejpam-6638	145	5	u	u	NOUN
ejpam-6638	145	6	and	and	CCONJ
ejpam-6638	145	7	v	v	PROPN
ejpam-6638	145	8	represent	represent	ADJ
ejpam-6638	145	9	orthogonal	orthogonal	ADJ
ejpam-6638	145	10	matrices	matrix	NOUN
ejpam-6638	145	11	,	,	PUNCT
ejpam-6638	145	12	while	while	SCONJ
ejpam-6638	145	13	σ	σ	PROPN
ejpam-6638	145	14	is	be	AUX
ejpam-6638	145	15	a	a	DET
ejpam-6638	145	16	diagonal	diagonal	ADJ
ejpam-6638	145	17	matrix	matrix	NOUN
ejpam-6638	145	18	that	that	PRON
ejpam-6638	145	19	holds	hold	VERB
ejpam-6638	145	20	the	the	DET
ejpam-6638	145	21	singular	singular	ADJ
ejpam-6638	145	22	values	value	NOUN
ejpam-6638	145	23	:	:	PUNCT
ejpam-6638	145	24	m	m	VERB
ejpam-6638	145	25	=	=	ADJ
ejpam-6638	145	26	uσv	uσv	NUM
ejpam-6638	145	27	⊤	⊤	NOUN
ejpam-6638	145	28	(	(	PUNCT
ejpam-6638	145	29	3.2	3.2	NUM
ejpam-6638	145	30	)	)	PUNCT
ejpam-6638	145	31	.	.	PUNCT
ejpam-6638	146	1	one	one	NUM
ejpam-6638	146	2	way	way	NOUN
ejpam-6638	146	3	to	to	PART
ejpam-6638	146	4	apply	apply	VERB
ejpam-6638	146	5	svd	svd	PROPN
ejpam-6638	146	6	pooling	pool	VERB
ejpam-6638	146	7	is	be	AUX
ejpam-6638	146	8	by	by	ADP
ejpam-6638	146	9	projecting	project	VERB
ejpam-6638	146	10	onto	onto	ADP
ejpam-6638	146	11	the	the	DET
ejpam-6638	146	12	primary	primary	ADJ
ejpam-6638	146	13	singular	singular	ADJ
ejpam-6638	146	14	vectors	vector	NOUN
ejpam-6638	146	15	or	or	CCONJ
ejpam-6638	146	16	keeping	keep	VERB
ejpam-6638	146	17	the	the	DET
ejpam-6638	146	18	top	top	ADJ
ejpam-6638	146	19	-	-	PUNCT
ejpam-6638	146	20	k	k	NOUN
ejpam-6638	146	21	singular	singular	PROPN
ejpam-6638	146	22	values	value	NOUN
ejpam-6638	146	23	.	.	PUNCT
ejpam-6638	147	1	this	this	DET
ejpam-6638	147	2	method	method	NOUN
ejpam-6638	147	3	emphasizes	emphasize	VERB
ejpam-6638	147	4	the	the	DET
ejpam-6638	147	5	key	key	ADJ
ejpam-6638	147	6	structural	structural	ADJ
ejpam-6638	147	7	patterns	pattern	NOUN
ejpam-6638	147	8	and	and	CCONJ
ejpam-6638	147	9	acts	act	NOUN
ejpam-6638	147	10	as	as	ADP
ejpam-6638	147	11	a	a	DET
ejpam-6638	147	12	global	global	ADJ
ejpam-6638	147	13	pooling	pooling	NOUN
ejpam-6638	147	14	technique	technique	NOUN
ejpam-6638	147	15	,	,	PUNCT
ejpam-6638	147	16	reducing	reduce	VERB
ejpam-6638	147	17	noise	noise	NOUN
ejpam-6638	147	18	and	and	CCONJ
ejpam-6638	147	19	accentuating	accentuate	VERB
ejpam-6638	147	20	important	important	ADJ
ejpam-6638	147	21	features	feature	NOUN
ejpam-6638	147	22	.	.	PUNCT
ejpam-6638	148	1	as	as	SCONJ
ejpam-6638	148	2	shown	show	VERB
ejpam-6638	148	3	in	in	ADP
ejpam-6638	148	4	fig	fig	NOUN
ejpam-6638	148	5	.	.	PUNCT
ejpam-6638	149	1	1(b	1(b	NUM
ejpam-6638	149	2	)	)	PUNCT
ejpam-6638	149	3	,	,	PUNCT
ejpam-6638	149	4	svd	svd	PROPN
ejpam-6638	149	5	pooling	pool	VERB
ejpam-6638	149	6	condenses	condense	VERB
ejpam-6638	149	7	local	local	ADJ
ejpam-6638	149	8	patches	patch	NOUN
ejpam-6638	149	9	(	(	PUNCT
ejpam-6638	149	10	e.g.	e.g.	ADV
ejpam-6638	149	11	,	,	PUNCT
ejpam-6638	149	12	2×	2×	NUM
ejpam-6638	149	13	2	2	NUM
ejpam-6638	149	14	)	)	PUNCT
ejpam-6638	149	15	by	by	ADP
ejpam-6638	149	16	retaining	retain	VERB
ejpam-6638	149	17	dominant	dominant	ADJ
ejpam-6638	149	18	singular	singular	PROPN
ejpam-6638	149	19	m.	m.	NOUN
ejpam-6638	149	20	israr	israr	PROPN
ejpam-6638	149	21	et	et	PROPN
ejpam-6638	149	22	al	al	PROPN
ejpam-6638	149	23	.	.	PUNCT
ejpam-6638	149	24	/	/	SYM
ejpam-6638	149	25	eur	eur	PROPN
ejpam-6638	149	26	.	.	PUNCT
ejpam-6638	150	1	j.	j.	PROPN
ejpam-6638	150	2	pure	pure	PROPN
ejpam-6638	150	3	appl	appl	PROPN
ejpam-6638	150	4	.	.	PROPN
ejpam-6638	150	5	math	math	PROPN
ejpam-6638	150	6	,	,	PUNCT
ejpam-6638	150	7	18	18	NUM
ejpam-6638	150	8	(	(	PUNCT
ejpam-6638	150	9	4	4	NUM
ejpam-6638	150	10	)	)	PUNCT
ejpam-6638	150	11	(	(	PUNCT
ejpam-6638	150	12	2025	2025	NUM
ejpam-6638	150	13	)	)	PUNCT
ejpam-6638	150	14	,	,	PUNCT
ejpam-6638	150	15	6638	6638	NUM
ejpam-6638	150	16	6	6	NUM
ejpam-6638	150	17	of	of	ADP
ejpam-6638	150	18	22	22	NUM
ejpam-6638	150	19	values	value	NOUN
ejpam-6638	150	20	,	,	PUNCT
ejpam-6638	150	21	effectively	effectively	ADV
ejpam-6638	150	22	distilling	distil	VERB
ejpam-6638	150	23	structural	structural	ADJ
ejpam-6638	150	24	patterns	pattern	NOUN
ejpam-6638	150	25	while	while	SCONJ
ejpam-6638	150	26	suppressing	suppress	VERB
ejpam-6638	150	27	noise	noise	NOUN
ejpam-6638	150	28	.	.	PUNCT
ejpam-6638	151	1	this	this	DET
ejpam-6638	151	2	operation	operation	NOUN
ejpam-6638	151	3	enhances	enhance	VERB
ejpam-6638	151	4	feature	feature	NOUN
ejpam-6638	151	5	hierarchies	hierarchy	NOUN
ejpam-6638	151	6	by	by	ADP
ejpam-6638	151	7	preserving	preserve	VERB
ejpam-6638	151	8	global	global	ADJ
ejpam-6638	151	9	coherence	coherence	NOUN
ejpam-6638	151	10	,	,	PUNCT
ejpam-6638	151	11	as	as	SCONJ
ejpam-6638	151	12	evidenced	evidence	VERB
ejpam-6638	151	13	by	by	ADP
ejpam-6638	151	14	the	the	DET
ejpam-6638	151	15	refined	refined	ADJ
ejpam-6638	151	16	output	output	NOUN
ejpam-6638	151	17	matrix	matrix	NOUN
ejpam-6638	151	18	in	in	ADP
ejpam-6638	151	19	the	the	DET
ejpam-6638	151	20	illustration	illustration	NOUN
ejpam-6638	151	21	.	.	PUNCT
ejpam-6638	152	1	3.3	3.3	NUM
ejpam-6638	152	2	.	.	PUNCT
ejpam-6638	153	1	multiscale	multiscale	ADJ
ejpam-6638	153	2	attention	attention	NOUN
ejpam-6638	153	3	block	block	NOUN
ejpam-6638	153	4	we	we	PRON
ejpam-6638	153	5	use	use	VERB
ejpam-6638	153	6	a	a	DET
ejpam-6638	153	7	multiscale	multiscale	ADJ
ejpam-6638	153	8	attention	attention	NOUN
ejpam-6638	153	9	block	block	NOUN
ejpam-6638	153	10	(	(	PUNCT
ejpam-6638	153	11	mab	mab	NOUN
ejpam-6638	153	12	)	)	PUNCT
ejpam-6638	153	13	in	in	ADP
ejpam-6638	153	14	the	the	DET
ejpam-6638	153	15	encoder	encoder	NOUN
ejpam-6638	153	16	-	-	PUNCT
ejpam-6638	153	17	decoder	decoder	NOUN
ejpam-6638	153	18	structure	structure	NOUN
ejpam-6638	153	19	to	to	PART
ejpam-6638	153	20	further	far	ADV
ejpam-6638	153	21	increase	increase	VERB
ejpam-6638	153	22	the	the	DET
ejpam-6638	153	23	network	network	NOUN
ejpam-6638	153	24	’s	’s	PART
ejpam-6638	153	25	representational	representational	ADJ
ejpam-6638	153	26	power	power	NOUN
ejpam-6638	153	27	.	.	PUNCT
ejpam-6638	154	1	the	the	DET
ejpam-6638	154	2	goal	goal	NOUN
ejpam-6638	154	3	of	of	ADP
ejpam-6638	154	4	this	this	DET
ejpam-6638	154	5	component	component	NOUN
ejpam-6638	154	6	is	be	AUX
ejpam-6638	154	7	to	to	PART
ejpam-6638	154	8	record	record	VERB
ejpam-6638	154	9	contextual	contextual	ADJ
ejpam-6638	154	10	dependencies	dependency	NOUN
ejpam-6638	154	11	at	at	ADP
ejpam-6638	154	12	different	different	ADJ
ejpam-6638	154	13	scales	scale	NOUN
ejpam-6638	154	14	,	,	PUNCT
ejpam-6638	154	15	both	both	CCONJ
ejpam-6638	154	16	local	local	ADJ
ejpam-6638	154	17	and	and	CCONJ
ejpam-6638	154	18	global	global	ADJ
ejpam-6638	154	19	.	.	PUNCT
ejpam-6638	155	1	in	in	ADP
ejpam-6638	155	2	order	order	NOUN
ejpam-6638	155	3	to	to	PART
ejpam-6638	155	4	extract	extract	VERB
ejpam-6638	155	5	scale	scale	NOUN
ejpam-6638	155	6	-	-	PUNCT
ejpam-6638	155	7	conscious	conscious	ADJ
ejpam-6638	155	8	features	feature	NOUN
ejpam-6638	155	9	,	,	PUNCT
ejpam-6638	155	10	the	the	DET
ejpam-6638	155	11	input	input	NOUN
ejpam-6638	155	12	feature	feature	NOUN
ejpam-6638	155	13	map	map	NOUN
ejpam-6638	155	14	is	be	AUX
ejpam-6638	155	15	first	first	ADV
ejpam-6638	155	16	split	split	VERB
ejpam-6638	155	17	up	up	ADP
ejpam-6638	155	18	into	into	ADP
ejpam-6638	155	19	many	many	ADJ
ejpam-6638	155	20	branches	branch	NOUN
ejpam-6638	155	21	,	,	PUNCT
ejpam-6638	155	22	each	each	PRON
ejpam-6638	155	23	of	of	ADP
ejpam-6638	155	24	which	which	PRON
ejpam-6638	155	25	is	be	AUX
ejpam-6638	155	26	then	then	ADV
ejpam-6638	155	27	subjected	subject	VERB
ejpam-6638	155	28	to	to	ADP
ejpam-6638	155	29	convolutional	convolutional	ADJ
ejpam-6638	155	30	filters	filter	NOUN
ejpam-6638	155	31	with	with	ADP
ejpam-6638	155	32	varying	vary	VERB
ejpam-6638	155	33	receptive	receptive	ADJ
ejpam-6638	155	34	field	field	NOUN
ejpam-6638	155	35	widths	width	NOUN
ejpam-6638	155	36	(	(	PUNCT
ejpam-6638	155	37	3	3	NUM
ejpam-6638	155	38	×	×	NOUN
ejpam-6638	155	39	3	3	NUM
ejpam-6638	155	40	,	,	PUNCT
ejpam-6638	155	41	5	5	NUM
ejpam-6638	155	42	×	×	NOUN
ejpam-6638	155	43	5	5	NUM
ejpam-6638	155	44	,	,	PUNCT
ejpam-6638	155	45	and	and	CCONJ
ejpam-6638	155	46	7	7	NUM
ejpam-6638	155	47	×	×	NOUN
ejpam-6638	155	48	7	7	NUM
ejpam-6638	155	49	,	,	PUNCT
ejpam-6638	155	50	for	for	ADP
ejpam-6638	155	51	example	example	NOUN
ejpam-6638	155	52	)	)	PUNCT
ejpam-6638	155	53	.	.	PUNCT
ejpam-6638	156	1	these	these	DET
ejpam-6638	156	2	features	feature	NOUN
ejpam-6638	156	3	are	be	AUX
ejpam-6638	156	4	then	then	ADV
ejpam-6638	156	5	combined	combine	VERB
ejpam-6638	156	6	and	and	CCONJ
ejpam-6638	156	7	sent	send	VERB
ejpam-6638	156	8	through	through	ADP
ejpam-6638	156	9	a	a	DET
ejpam-6638	156	10	channel	channel	NOUN
ejpam-6638	156	11	attention	attention	NOUN
ejpam-6638	156	12	mechanism	mechanism	NOUN
ejpam-6638	156	13	,	,	PUNCT
ejpam-6638	156	14	which	which	PRON
ejpam-6638	156	15	calculates	calculate	VERB
ejpam-6638	156	16	adaptive	adaptive	ADJ
ejpam-6638	156	17	weights	weight	NOUN
ejpam-6638	156	18	for	for	ADP
ejpam-6638	156	19	every	every	DET
ejpam-6638	156	20	channel	channel	NOUN
ejpam-6638	156	21	using	use	VERB
ejpam-6638	156	22	svd	svd	PROPN
ejpam-6638	156	23	pooling	pool	VERB
ejpam-6638	156	24	,	,	PUNCT
ejpam-6638	156	25	followed	follow	VERB
ejpam-6638	156	26	by	by	ADP
ejpam-6638	156	27	a	a	DET
ejpam-6638	156	28	small	small	ADJ
ejpam-6638	156	29	fully	fully	ADV
ejpam-6638	156	30	connected	connect	VERB
ejpam-6638	156	31	network	network	NOUN
ejpam-6638	156	32	and	and	CCONJ
ejpam-6638	156	33	a	a	DET
ejpam-6638	156	34	sigmoid	sigmoid	NOUN
ejpam-6638	156	35	activation	activation	NOUN
ejpam-6638	156	36	function	function	NOUN
ejpam-6638	156	37	.	.	PUNCT
ejpam-6638	157	1	the	the	DET
ejpam-6638	157	2	adjusted	adjust	VERB
ejpam-6638	157	3	feature	feature	NOUN
ejpam-6638	157	4	maps	map	NOUN
ejpam-6638	157	5	are	be	AUX
ejpam-6638	157	6	integrated	integrate	VERB
ejpam-6638	157	7	to	to	PART
ejpam-6638	157	8	create	create	VERB
ejpam-6638	157	9	the	the	DET
ejpam-6638	157	10	final	final	ADJ
ejpam-6638	157	11	output	output	NOUN
ejpam-6638	157	12	of	of	ADP
ejpam-6638	157	13	the	the	DET
ejpam-6638	157	14	mab	mab	NOUN
ejpam-6638	157	15	.	.	PUNCT
ejpam-6638	158	1	by	by	ADP
ejpam-6638	158	2	merging	merge	VERB
ejpam-6638	158	3	multi	multi	ADJ
ejpam-6638	158	4	-	-	ADJ
ejpam-6638	158	5	resolution	resolution	NOUN
ejpam-6638	158	6	processing	processing	NOUN
ejpam-6638	158	7	with	with	ADP
ejpam-6638	158	8	channelwise	channelwise	NOUN
ejpam-6638	158	9	attention	attention	NOUN
ejpam-6638	158	10	,	,	PUNCT
ejpam-6638	158	11	the	the	DET
ejpam-6638	158	12	multiscale	multiscale	ADJ
ejpam-6638	158	13	attention	attention	NOUN
ejpam-6638	158	14	block	block	NOUN
ejpam-6638	158	15	considerably	considerably	ADV
ejpam-6638	158	16	enhances	enhance	VERB
ejpam-6638	158	17	the	the	DET
ejpam-6638	158	18	network	network	NOUN
ejpam-6638	158	19	’s	’s	PART
ejpam-6638	158	20	ability	ability	NOUN
ejpam-6638	158	21	to	to	PART
ejpam-6638	158	22	represent	represent	VERB
ejpam-6638	158	23	complex	complex	ADJ
ejpam-6638	158	24	structures	structure	NOUN
ejpam-6638	158	25	and	and	CCONJ
ejpam-6638	158	26	textures	texture	NOUN
ejpam-6638	158	27	in	in	ADP
ejpam-6638	158	28	natural	natural	ADJ
ejpam-6638	158	29	images	image	NOUN
ejpam-6638	158	30	,	,	PUNCT
ejpam-6638	158	31	thus	thus	ADV
ejpam-6638	158	32	improving	improve	VERB
ejpam-6638	158	33	restoration	restoration	NOUN
ejpam-6638	158	34	quality	quality	NOUN
ejpam-6638	158	35	in	in	ADP
ejpam-6638	158	36	tasks	task	NOUN
ejpam-6638	158	37	.	.	PUNCT
ejpam-6638	159	1	the	the	DET
ejpam-6638	159	2	mab	mab	NOUN
ejpam-6638	159	3	uses	use	VERB
ejpam-6638	159	4	parallel	parallel	ADJ
ejpam-6638	159	5	convolutional	convolutional	ADJ
ejpam-6638	159	6	branching	branching	NOUN
ejpam-6638	159	7	and	and	CCONJ
ejpam-6638	159	8	relu	relu	NOUN
ejpam-6638	159	9	activations	activation	NOUN
ejpam-6638	159	10	to	to	PART
ejpam-6638	159	11	process	process	VERB
ejpam-6638	159	12	multi	multi	ADJ
ejpam-6638	159	13	-	-	ADJ
ejpam-6638	159	14	scale	scale	ADJ
ejpam-6638	159	15	features	feature	NOUN
ejpam-6638	159	16	,	,	PUNCT
ejpam-6638	159	17	as	as	SCONJ
ejpam-6638	159	18	shown	show	VERB
ejpam-6638	159	19	in	in	ADP
ejpam-6638	159	20	fig	fig	NOUN
ejpam-6638	159	21	.	.	PUNCT
ejpam-6638	160	1	1(a	1(a	NUM
ejpam-6638	160	2	)	)	PUNCT
ejpam-6638	160	3	,	,	PUNCT
ejpam-6638	160	4	and	and	CCONJ
ejpam-6638	160	5	then	then	ADV
ejpam-6638	160	6	attention	attention	NOUN
ejpam-6638	160	7	map	map	NOUN
ejpam-6638	160	8	fusion	fusion	NOUN
ejpam-6638	160	9	.	.	PUNCT
ejpam-6638	161	1	the	the	DET
ejpam-6638	161	2	context	context	NOUN
ejpam-6638	161	3	-	-	PUNCT
ejpam-6638	161	4	aware	aware	ADJ
ejpam-6638	161	5	feature	feature	NOUN
ejpam-6638	161	6	refinement	refinement	NOUN
ejpam-6638	161	7	necessary	necessary	ADJ
ejpam-6638	161	8	for	for	SCONJ
ejpam-6638	161	9	image	image	NOUN
ejpam-6638	161	10	restoration	restoration	NOUN
ejpam-6638	161	11	is	be	AUX
ejpam-6638	161	12	made	make	VERB
ejpam-6638	161	13	possible	possible	ADJ
ejpam-6638	161	14	by	by	ADP
ejpam-6638	161	15	the	the	DET
ejpam-6638	161	16	softplus	softplus	ADJ
ejpam-6638	161	17	-	-	PUNCT
ejpam-6638	161	18	activated	activate	VERB
ejpam-6638	161	19	weights	weight	NOUN
ejpam-6638	161	20	,	,	PUNCT
ejpam-6638	161	21	which	which	PRON
ejpam-6638	161	22	adaptively	adaptively	ADV
ejpam-6638	161	23	emphasize	emphasize	VERB
ejpam-6638	161	24	important	important	ADJ
ejpam-6638	161	25	structures	structure	NOUN
ejpam-6638	161	26	across	across	ADP
ejpam-6638	161	27	scales	scale	NOUN
ejpam-6638	161	28	.	.	PUNCT
ejpam-6638	162	1	3.4	3.4	NUM
ejpam-6638	162	2	.	.	PUNCT
ejpam-6638	162	3	modified	modify	VERB
ejpam-6638	162	4	network	network	NOUN
ejpam-6638	162	5	architecture	architecture	NOUN
ejpam-6638	162	6	we	we	PRON
ejpam-6638	162	7	incorporate	incorporate	VERB
ejpam-6638	162	8	svd	svd	PROPN
ejpam-6638	162	9	pooling	pool	VERB
ejpam-6638	162	10	into	into	ADP
ejpam-6638	162	11	the	the	DET
ejpam-6638	162	12	dip	dip	NOUN
ejpam-6638	162	13	network	network	NOUN
ejpam-6638	162	14	by	by	ADP
ejpam-6638	162	15	replacing	replace	VERB
ejpam-6638	162	16	conventional	conventional	ADJ
ejpam-6638	162	17	pooling	pool	VERB
ejpam-6638	162	18	layers	layer	NOUN
ejpam-6638	162	19	at	at	ADP
ejpam-6638	162	20	multiple	multiple	ADJ
ejpam-6638	162	21	scales	scale	NOUN
ejpam-6638	162	22	in	in	ADP
ejpam-6638	162	23	the	the	DET
ejpam-6638	162	24	encoder	encoder	NOUN
ejpam-6638	162	25	-	-	PUNCT
ejpam-6638	162	26	decoder	decoder	NOUN
ejpam-6638	162	27	architecture	architecture	NOUN
ejpam-6638	162	28	.	.	PUNCT
ejpam-6638	163	1	the	the	DET
ejpam-6638	163	2	network	network	NOUN
ejpam-6638	163	3	follows	follow	VERB
ejpam-6638	163	4	a	a	DET
ejpam-6638	163	5	multi	multi	ADJ
ejpam-6638	163	6	-	-	ADJ
ejpam-6638	163	7	scale	scale	ADJ
ejpam-6638	163	8	unet	unet	NOUN
ejpam-6638	163	9	style	style	NOUN
ejpam-6638	163	10	configuration	configuration	NOUN
ejpam-6638	163	11	,	,	PUNCT
ejpam-6638	163	12	where	where	SCONJ
ejpam-6638	163	13	downsampling	downsampling	NOUN
ejpam-6638	163	14	is	be	AUX
ejpam-6638	163	15	performed	perform	VERB
ejpam-6638	163	16	via	via	ADP
ejpam-6638	163	17	strided	strided	ADJ
ejpam-6638	163	18	convolutions	convolution	NOUN
ejpam-6638	163	19	and	and	CCONJ
ejpam-6638	163	20	svd	svd	PROPN
ejpam-6638	163	21	pooling	pool	VERB
ejpam-6638	163	22	,	,	PUNCT
ejpam-6638	163	23	while	while	SCONJ
ejpam-6638	163	24	upsampling	upsample	VERB
ejpam-6638	163	25	uses	use	VERB
ejpam-6638	163	26	nearest	near	ADJ
ejpam-6638	163	27	-	-	PUNCT
ejpam-6638	163	28	neighbor	neighbor	NOUN
ejpam-6638	163	29	interpolation	interpolation	NOUN
ejpam-6638	163	30	and	and	CCONJ
ejpam-6638	163	31	convolutional	convolutional	ADJ
ejpam-6638	163	32	refinement	refinement	NOUN
ejpam-6638	163	33	.	.	PUNCT
ejpam-6638	164	1	furthermore	furthermore	ADV
ejpam-6638	164	2	,	,	PUNCT
ejpam-6638	164	3	skip	skip	ADJ
ejpam-6638	164	4	connections	connection	NOUN
ejpam-6638	164	5	maintain	maintain	VERB
ejpam-6638	164	6	spatial	spatial	ADJ
ejpam-6638	164	7	details	detail	NOUN
ejpam-6638	164	8	and	and	CCONJ
ejpam-6638	164	9	enhance	enhance	VERB
ejpam-6638	164	10	gradient	gradient	ADJ
ejpam-6638	164	11	propagation	propagation	NOUN
ejpam-6638	164	12	by	by	ADP
ejpam-6638	164	13	merging	merge	VERB
ejpam-6638	164	14	both	both	CCONJ
ejpam-6638	164	15	low	low	ADJ
ejpam-6638	164	16	-	-	PUNCT
ejpam-6638	164	17	level	level	NOUN
ejpam-6638	164	18	and	and	CCONJ
ejpam-6638	164	19	high	high	ADJ
ejpam-6638	164	20	-	-	PUNCT
ejpam-6638	164	21	level	level	NOUN
ejpam-6638	164	22	features	feature	NOUN
ejpam-6638	164	23	.	.	PUNCT
ejpam-6638	165	1	the	the	DET
ejpam-6638	165	2	design	design	NOUN
ejpam-6638	165	3	incorporates	incorporate	VERB
ejpam-6638	165	4	convolutional	convolutional	ADJ
ejpam-6638	165	5	layers	layer	NOUN
ejpam-6638	165	6	utilizing	utilize	VERB
ejpam-6638	165	7	3×	3×	NUM
ejpam-6638	165	8	3	3	NUM
ejpam-6638	165	9	kernels	kernel	NOUN
ejpam-6638	165	10	,	,	PUNCT
ejpam-6638	165	11	followed	follow	VERB
ejpam-6638	165	12	by	by	ADP
ejpam-6638	165	13	the	the	DET
ejpam-6638	165	14	softplus	softplus	ADJ
ejpam-6638	165	15	activation	activation	NOUN
ejpam-6638	165	16	function	function	NOUN
ejpam-6638	165	17	and	and	CCONJ
ejpam-6638	165	18	instance	instance	NOUN
ejpam-6638	165	19	normalization	normalization	NOUN
ejpam-6638	165	20	.	.	PUNCT
ejpam-6638	166	1	the	the	DET
ejpam-6638	166	2	equation	equation	NOUN
ejpam-6638	166	3	ssp(x	ssp(x	PROPN
ejpam-6638	166	4	)	)	PUNCT
ejpam-6638	166	5	=	=	PUNCT
ejpam-6638	166	6	log	log	NOUN
ejpam-6638	166	7	(	(	PUNCT
ejpam-6638	166	8	1	1	NUM
ejpam-6638	166	9	+	+	CCONJ
ejpam-6638	166	10	ex	ex	X
ejpam-6638	166	11	)	)	PUNCT
ejpam-6638	166	12	−	−	PROPN
ejpam-6638	166	13	log(2	log(2	NOUN
ejpam-6638	166	14	)	)	PUNCT
ejpam-6638	166	15	,	,	PUNCT
ejpam-6638	166	16	(	(	PUNCT
ejpam-6638	166	17	3.3	3.3	NUM
ejpam-6638	166	18	)	)	PUNCT
ejpam-6638	166	19	provides	provide	VERB
ejpam-6638	166	20	smooth	smooth	ADJ
ejpam-6638	166	21	gradients	gradient	NOUN
ejpam-6638	166	22	and	and	CCONJ
ejpam-6638	166	23	prevents	prevent	VERB
ejpam-6638	166	24	the	the	DET
ejpam-6638	166	25	occurrence	occurrence	NOUN
ejpam-6638	166	26	of	of	ADP
ejpam-6638	166	27	dead	dead	ADJ
ejpam-6638	166	28	neurons	neuron	NOUN
ejpam-6638	166	29	,	,	PUNCT
ejpam-6638	166	30	which	which	PRON
ejpam-6638	166	31	can	can	AUX
ejpam-6638	166	32	happen	happen	VERB
ejpam-6638	166	33	with	with	ADP
ejpam-6638	166	34	relu	relu	NOUN
ejpam-6638	166	35	.	.	PUNCT
ejpam-6638	167	1	to	to	PART
ejpam-6638	167	2	constrain	constrain	VERB
ejpam-6638	167	3	the	the	DET
ejpam-6638	167	4	output	output	NOUN
ejpam-6638	167	5	within	within	ADP
ejpam-6638	167	6	the	the	DET
ejpam-6638	167	7	range	range	NOUN
ejpam-6638	167	8	of	of	ADP
ejpam-6638	167	9	[	[	X
ejpam-6638	167	10	0	0	NUM
ejpam-6638	167	11	,	,	PUNCT
ejpam-6638	167	12	1	1	NUM
ejpam-6638	167	13	]	]	PUNCT
ejpam-6638	167	14	,	,	PUNCT
ejpam-6638	167	15	the	the	DET
ejpam-6638	167	16	output	output	NOUN
ejpam-6638	167	17	layer	layer	NOUN
ejpam-6638	167	18	employs	employ	VERB
ejpam-6638	167	19	a	a	DET
ejpam-6638	167	20	sigmoid	sigmoid	NOUN
ejpam-6638	167	21	activation	activation	NOUN
ejpam-6638	167	22	function	function	NOUN
ejpam-6638	167	23	.	.	PUNCT
ejpam-6638	168	1	3.5	3.5	NUM
ejpam-6638	168	2	.	.	PUNCT
ejpam-6638	169	1	optimization	optimization	NOUN
ejpam-6638	169	2	strategy	strategy	NOUN
ejpam-6638	169	3	a	a	DET
ejpam-6638	169	4	combination	combination	NOUN
ejpam-6638	169	5	of	of	ADP
ejpam-6638	169	6	adaptive	adaptive	ADJ
ejpam-6638	169	7	optimizers	optimizer	NOUN
ejpam-6638	169	8	and	and	CCONJ
ejpam-6638	169	9	preconditioned	preconditioned	ADJ
ejpam-6638	169	10	optimizers	optimizer	NOUN
ejpam-6638	169	11	is	be	AUX
ejpam-6638	169	12	employed	employ	VERB
ejpam-6638	169	13	to	to	PART
ejpam-6638	169	14	optimize	optimize	VERB
ejpam-6638	169	15	the	the	DET
ejpam-6638	169	16	network	network	NOUN
ejpam-6638	169	17	.	.	PUNCT
ejpam-6638	170	1	after	after	ADP
ejpam-6638	170	2	evaluating	evaluate	VERB
ejpam-6638	170	3	adam	adam	PROPN
ejpam-6638	170	4	,	,	PUNCT
ejpam-6638	170	5	apollo	apollo	PROPN
ejpam-6638	171	1	[	[	X
ejpam-6638	171	2	52	52	NUM
ejpam-6638	171	3	]	]	PUNCT
ejpam-6638	171	4	,	,	PUNCT
ejpam-6638	171	5	and	and	CCONJ
ejpam-6638	171	6	diffgrad	diffgrad	VERB
ejpam-6638	172	1	[	[	X
ejpam-6638	172	2	53	53	NUM
ejpam-6638	172	3	]	]	PUNCT
ejpam-6638	172	4	,	,	PUNCT
ejpam-6638	172	5	we	we	PRON
ejpam-6638	172	6	discovered	discover	VERB
ejpam-6638	172	7	that	that	SCONJ
ejpam-6638	172	8	preconditioned	preconditioned	ADJ
ejpam-6638	172	9	optimizers	optimizer	NOUN
ejpam-6638	172	10	improve	improve	VERB
ejpam-6638	172	11	restoration	restoration	NOUN
ejpam-6638	172	12	quality	quality	NOUN
ejpam-6638	172	13	and	and	CCONJ
ejpam-6638	172	14	accelerate	accelerate	VERB
ejpam-6638	172	15	convergence	convergence	NOUN
ejpam-6638	172	16	.	.	PUNCT
ejpam-6638	173	1	the	the	DET
ejpam-6638	173	2	loss	loss	NOUN
ejpam-6638	173	3	function	function	NOUN
ejpam-6638	173	4	is	be	AUX
ejpam-6638	173	5	defined	define	VERB
ejpam-6638	173	6	as	as	ADP
ejpam-6638	173	7	the	the	DET
ejpam-6638	173	8	mse	mse	NOUN
ejpam-6638	173	9	between	between	ADP
ejpam-6638	173	10	the	the	DET
ejpam-6638	173	11	target	target	NOUN
ejpam-6638	173	12	image	image	NOUN
ejpam-6638	173	13	and	and	CCONJ
ejpam-6638	173	14	the	the	DET
ejpam-6638	173	15	output	output	NOUN
ejpam-6638	173	16	of	of	ADP
ejpam-6638	173	17	the	the	DET
ejpam-6638	173	18	network	network	NOUN
ejpam-6638	173	19	:	:	PUNCT
ejpam-6638	173	20	ltotal	ltotal	ADJ
ejpam-6638	173	21	=	=	PUNCT
ejpam-6638	173	22	∥fθ(z)−	∥fθ(z)−	X
ejpam-6638	174	1	y∥22	y∥22	PROPN
ejpam-6638	174	2	.	.	PUNCT
ejpam-6638	175	1	(	(	PUNCT
ejpam-6638	175	2	3.4	3.4	NUM
ejpam-6638	175	3	)	)	PUNCT
ejpam-6638	175	4	to	to	PART
ejpam-6638	175	5	prevent	prevent	VERB
ejpam-6638	175	6	overfitting	overfitte	VERB
ejpam-6638	175	7	to	to	ADP
ejpam-6638	175	8	noise	noise	NOUN
ejpam-6638	175	9	,	,	PUNCT
ejpam-6638	175	10	a	a	DET
ejpam-6638	175	11	common	common	ADJ
ejpam-6638	175	12	issue	issue	NOUN
ejpam-6638	175	13	in	in	ADP
ejpam-6638	175	14	dip	dip	NOUN
ejpam-6638	175	15	-	-	PUNCT
ejpam-6638	175	16	based	base	VERB
ejpam-6638	175	17	models	model	NOUN
ejpam-6638	175	18	,	,	PUNCT
ejpam-6638	175	19	we	we	PRON
ejpam-6638	175	20	implement	implement	VERB
ejpam-6638	175	21	early	early	ADJ
ejpam-6638	175	22	stopping	stopping	NOUN
ejpam-6638	175	23	.	.	PUNCT
ejpam-6638	176	1	m.	m.	NOUN
ejpam-6638	176	2	israr	israr	PROPN
ejpam-6638	176	3	et	et	PROPN
ejpam-6638	176	4	al	al	PROPN
ejpam-6638	176	5	.	.	PUNCT
ejpam-6638	176	6	/	/	SYM
ejpam-6638	176	7	eur	eur	PROPN
ejpam-6638	176	8	.	.	PUNCT
ejpam-6638	177	1	j.	j.	PROPN
ejpam-6638	177	2	pure	pure	PROPN
ejpam-6638	177	3	appl	appl	PROPN
ejpam-6638	177	4	.	.	PROPN
ejpam-6638	177	5	math	math	PROPN
ejpam-6638	177	6	,	,	PUNCT
ejpam-6638	177	7	18	18	NUM
ejpam-6638	177	8	(	(	PUNCT
ejpam-6638	177	9	4	4	NUM
ejpam-6638	177	10	)	)	PUNCT
ejpam-6638	177	11	(	(	PUNCT
ejpam-6638	177	12	2025	2025	NUM
ejpam-6638	177	13	)	)	PUNCT
ejpam-6638	177	14	,	,	PUNCT
ejpam-6638	177	15	6638	6638	NUM
ejpam-6638	177	16	7	7	NUM
ejpam-6638	177	17	of	of	ADP
ejpam-6638	177	18	22	22	NUM
ejpam-6638	177	19	(	(	PUNCT
ejpam-6638	177	20	a	a	NOUN
ejpam-6638	177	21	)	)	PUNCT
ejpam-6638	177	22	(	(	PUNCT
ejpam-6638	177	23	b	b	X
ejpam-6638	177	24	)	)	PUNCT
ejpam-6638	177	25	(	(	PUNCT
ejpam-6638	177	26	c	c	X
ejpam-6638	177	27	)	)	PUNCT
ejpam-6638	177	28	figure	figure	NOUN
ejpam-6638	177	29	1	1	NUM
ejpam-6638	177	30	:	:	PUNCT
ejpam-6638	177	31	overview	overview	NOUN
ejpam-6638	177	32	of	of	ADP
ejpam-6638	177	33	the	the	DET
ejpam-6638	177	34	proposed	propose	VERB
ejpam-6638	177	35	svd	svd	PROPN
ejpam-6638	177	36	-	-	PUNCT
ejpam-6638	177	37	dip	dip	NOUN
ejpam-6638	177	38	framework	framework	NOUN
ejpam-6638	177	39	with	with	ADP
ejpam-6638	177	40	subfigure	subfigure	NOUN
ejpam-6638	177	41	labels	label	NOUN
ejpam-6638	177	42	:	:	PUNCT
ejpam-6638	177	43	(	(	PUNCT
ejpam-6638	177	44	a	a	X
ejpam-6638	177	45	)	)	PUNCT
ejpam-6638	177	46	multiscale	multiscale	ADJ
ejpam-6638	177	47	attention	attention	NOUN
ejpam-6638	177	48	block	block	NOUN
ejpam-6638	177	49	,	,	PUNCT
ejpam-6638	177	50	(	(	PUNCT
ejpam-6638	177	51	b	b	X
ejpam-6638	177	52	)	)	PUNCT
ejpam-6638	177	53	svd	svd	PROPN
ejpam-6638	177	54	-	-	PUNCT
ejpam-6638	177	55	based	base	VERB
ejpam-6638	177	56	pooling	pooling	NOUN
ejpam-6638	177	57	module	module	NOUN
ejpam-6638	177	58	,	,	PUNCT
ejpam-6638	177	59	and	and	CCONJ
ejpam-6638	177	60	(	(	PUNCT
ejpam-6638	177	61	c	c	NOUN
ejpam-6638	177	62	)	)	PUNCT
ejpam-6638	177	63	operational	operational	ADJ
ejpam-6638	177	64	workflow	workflow	NOUN
ejpam-6638	177	65	of	of	ADP
ejpam-6638	177	66	the	the	DET
ejpam-6638	177	67	svddip	svddip	ADJ
ejpam-6638	177	68	architecture	architecture	NOUN
ejpam-6638	177	69	.	.	PUNCT
ejpam-6638	178	1	properties	property	NOUN
ejpam-6638	178	2	of	of	ADP
ejpam-6638	178	3	diffgrad	diffgrad	PROPN
ejpam-6638	178	4	in	in	ADP
ejpam-6638	178	5	dip	dip	NOUN
ejpam-6638	178	6	-	-	PUNCT
ejpam-6638	178	7	svd	svd	PROPN
ejpam-6638	178	8	optimization	optimization	NOUN
ejpam-6638	178	9	.	.	PUNCT
ejpam-6638	179	1	the	the	DET
ejpam-6638	179	2	diffgrad	diffgrad	PROPN
ejpam-6638	179	3	optimizer	optimizer	NOUN
ejpam-6638	179	4	introduces	introduce	VERB
ejpam-6638	179	5	a	a	DET
ejpam-6638	179	6	gradient	gradient	NOUN
ejpam-6638	179	7	-	-	PUNCT
ejpam-6638	179	8	difference	difference	NOUN
ejpam-6638	179	9	–	–	PUNCT
ejpam-6638	179	10	based	base	VERB
ejpam-6638	179	11	modulation	modulation	NOUN
ejpam-6638	179	12	to	to	ADP
ejpam-6638	179	13	the	the	DET
ejpam-6638	179	14	learning	learning	NOUN
ejpam-6638	179	15	rate	rate	NOUN
ejpam-6638	179	16	,	,	PUNCT
ejpam-6638	179	17	which	which	PRON
ejpam-6638	179	18	reduces	reduce	VERB
ejpam-6638	179	19	updates	update	NOUN
ejpam-6638	179	20	when	when	SCONJ
ejpam-6638	179	21	gradients	gradient	NOUN
ejpam-6638	179	22	exhibit	exhibit	VERB
ejpam-6638	179	23	little	little	ADJ
ejpam-6638	179	24	variation	variation	NOUN
ejpam-6638	179	25	and	and	CCONJ
ejpam-6638	179	26	increases	increase	VERB
ejpam-6638	179	27	them	they	PRON
ejpam-6638	179	28	when	when	SCONJ
ejpam-6638	179	29	gradients	gradient	NOUN
ejpam-6638	179	30	fluctuate	fluctuate	VERB
ejpam-6638	179	31	.	.	PUNCT
ejpam-6638	180	1	this	this	DET
ejpam-6638	180	2	dynamic	dynamic	ADJ
ejpam-6638	180	3	adjustment	adjustment	NOUN
ejpam-6638	180	4	helps	help	VERB
ejpam-6638	180	5	the	the	DET
ejpam-6638	180	6	network	network	NOUN
ejpam-6638	180	7	avoid	avoid	VERB
ejpam-6638	180	8	stagnation	stagnation	NOUN
ejpam-6638	180	9	and	and	CCONJ
ejpam-6638	180	10	overshooting	overshoot	VERB
ejpam-6638	180	11	during	during	ADP
ejpam-6638	180	12	training	training	NOUN
ejpam-6638	180	13	.	.	PUNCT
ejpam-6638	181	1	in	in	ADP
ejpam-6638	181	2	the	the	DET
ejpam-6638	181	3	context	context	NOUN
ejpam-6638	181	4	of	of	ADP
ejpam-6638	181	5	dip	dip	NOUN
ejpam-6638	181	6	-	-	PUNCT
ejpam-6638	181	7	svd	svd	PROPN
ejpam-6638	181	8	,	,	PUNCT
ejpam-6638	181	9	where	where	SCONJ
ejpam-6638	181	10	the	the	DET
ejpam-6638	181	11	objective	objective	ADJ
ejpam-6638	181	12	landscape	landscape	NOUN
ejpam-6638	181	13	is	be	AUX
ejpam-6638	181	14	highly	highly	ADV
ejpam-6638	181	15	ill	ill	ADV
ejpam-6638	181	16	-	-	PUNCT
ejpam-6638	181	17	conditioned	condition	VERB
ejpam-6638	181	18	and	and	CCONJ
ejpam-6638	181	19	prone	prone	ADJ
ejpam-6638	181	20	to	to	ADP
ejpam-6638	181	21	overfitting	overfitte	VERB
ejpam-6638	181	22	,	,	PUNCT
ejpam-6638	181	23	diffm	diffm	NOUN
ejpam-6638	181	24	.	.	PUNCT
ejpam-6638	181	25	israr	israr	PROPN
ejpam-6638	182	1	et	et	PROPN
ejpam-6638	182	2	al	al	PROPN
ejpam-6638	182	3	.	.	PUNCT
ejpam-6638	182	4	/	/	SYM
ejpam-6638	182	5	eur	eur	PROPN
ejpam-6638	182	6	.	.	PUNCT
ejpam-6638	183	1	j.	j.	PROPN
ejpam-6638	183	2	pure	pure	PROPN
ejpam-6638	183	3	appl	appl	PROPN
ejpam-6638	183	4	.	.	PROPN
ejpam-6638	183	5	math	math	PROPN
ejpam-6638	183	6	,	,	PUNCT
ejpam-6638	183	7	18	18	NUM
ejpam-6638	183	8	(	(	PUNCT
ejpam-6638	183	9	4	4	NUM
ejpam-6638	183	10	)	)	PUNCT
ejpam-6638	183	11	(	(	PUNCT
ejpam-6638	183	12	2025	2025	NUM
ejpam-6638	183	13	)	)	PUNCT
ejpam-6638	183	14	,	,	PUNCT
ejpam-6638	183	15	6638	6638	NUM
ejpam-6638	183	16	8	8	NUM
ejpam-6638	183	17	of	of	ADP
ejpam-6638	183	18	22	22	NUM
ejpam-6638	183	19	grad	grad	NOUN
ejpam-6638	183	20	stabilizes	stabilize	VERB
ejpam-6638	183	21	convergence	convergence	NOUN
ejpam-6638	183	22	by	by	ADP
ejpam-6638	183	23	(	(	PUNCT
ejpam-6638	183	24	i	i	NOUN
ejpam-6638	183	25	)	)	PUNCT
ejpam-6638	183	26	adapting	adapt	VERB
ejpam-6638	183	27	step	step	NOUN
ejpam-6638	183	28	sizes	size	NOUN
ejpam-6638	183	29	based	base	VERB
ejpam-6638	183	30	on	on	ADP
ejpam-6638	183	31	gradient	gradient	ADJ
ejpam-6638	183	32	history	history	NOUN
ejpam-6638	183	33	,	,	PUNCT
ejpam-6638	183	34	(	(	PUNCT
ejpam-6638	183	35	ii	ii	NOUN
ejpam-6638	183	36	)	)	PUNCT
ejpam-6638	183	37	enhancing	enhance	VERB
ejpam-6638	183	38	robustness	robustness	NOUN
ejpam-6638	183	39	against	against	ADP
ejpam-6638	183	40	flat	flat	ADJ
ejpam-6638	183	41	or	or	CCONJ
ejpam-6638	183	42	noisy	noisy	ADJ
ejpam-6638	183	43	gradient	gradient	ADJ
ejpam-6638	183	44	regions	region	NOUN
ejpam-6638	183	45	,	,	PUNCT
ejpam-6638	183	46	and	and	CCONJ
ejpam-6638	183	47	(	(	PUNCT
ejpam-6638	183	48	iii	iii	NOUN
ejpam-6638	183	49	)	)	PUNCT
ejpam-6638	183	50	reducing	reduce	VERB
ejpam-6638	183	51	sensitivity	sensitivity	NOUN
ejpam-6638	183	52	to	to	ADP
ejpam-6638	183	53	the	the	DET
ejpam-6638	183	54	choice	choice	NOUN
ejpam-6638	183	55	of	of	ADP
ejpam-6638	183	56	initial	initial	ADJ
ejpam-6638	183	57	learning	learning	NOUN
ejpam-6638	183	58	rate	rate	NOUN
ejpam-6638	183	59	.	.	PUNCT
ejpam-6638	184	1	these	these	DET
ejpam-6638	184	2	properties	property	NOUN
ejpam-6638	184	3	collectively	collectively	ADV
ejpam-6638	184	4	lead	lead	VERB
ejpam-6638	184	5	to	to	ADP
ejpam-6638	184	6	more	more	ADV
ejpam-6638	184	7	consistent	consistent	ADJ
ejpam-6638	184	8	parameter	parameter	NOUN
ejpam-6638	184	9	updates	update	NOUN
ejpam-6638	184	10	and	and	CCONJ
ejpam-6638	184	11	better	well	ADJ
ejpam-6638	184	12	structural	structural	ADJ
ejpam-6638	184	13	preservation	preservation	NOUN
ejpam-6638	184	14	in	in	ADP
ejpam-6638	184	15	the	the	DET
ejpam-6638	184	16	reconstructed	reconstruct	VERB
ejpam-6638	184	17	images	image	NOUN
ejpam-6638	184	18	.	.	PUNCT
ejpam-6638	185	1	4	4	X
ejpam-6638	185	2	.	.	X
ejpam-6638	185	3	experiments	experiment	NOUN
ejpam-6638	185	4	we	we	PRON
ejpam-6638	185	5	apply	apply	VERB
ejpam-6638	185	6	our	our	PRON
ejpam-6638	185	7	method	method	NOUN
ejpam-6638	185	8	to	to	PART
ejpam-6638	185	9	image	image	VERB
ejpam-6638	185	10	denoising	denoising	NOUN
ejpam-6638	185	11	and	and	CCONJ
ejpam-6638	185	12	image	image	NOUN
ejpam-6638	185	13	inpainting	inpainte	VERB
ejpam-6638	185	14	tasks	task	NOUN
ejpam-6638	185	15	using	use	VERB
ejpam-6638	185	16	standard	standard	ADJ
ejpam-6638	185	17	datasets	dataset	NOUN
ejpam-6638	185	18	.	.	PUNCT
ejpam-6638	186	1	our	our	PRON
ejpam-6638	186	2	test	test	NOUN
ejpam-6638	186	3	dataset	dataset	VERB
ejpam-6638	186	4	consists	consist	NOUN
ejpam-6638	186	5	of	of	ADP
ejpam-6638	186	6	6	6	NUM
ejpam-6638	186	7	images	image	NOUN
ejpam-6638	186	8	for	for	ADP
ejpam-6638	186	9	denoising	denoising	NOUN
ejpam-6638	186	10	and	and	CCONJ
ejpam-6638	186	11	4	4	NUM
ejpam-6638	186	12	images	image	NOUN
ejpam-6638	186	13	for	for	ADP
ejpam-6638	186	14	inpainting	inpainte	VERB
ejpam-6638	186	15	of	of	ADP
ejpam-6638	186	16	different	different	ADJ
ejpam-6638	186	17	sizes	size	NOUN
ejpam-6638	186	18	.	.	PUNCT
ejpam-6638	187	1	the	the	DET
ejpam-6638	187	2	images	image	NOUN
ejpam-6638	187	3	for	for	ADP
ejpam-6638	187	4	denoising	denoising	NOUN
ejpam-6638	187	5	are	be	AUX
ejpam-6638	187	6	shown	show	VERB
ejpam-6638	187	7	in	in	ADP
ejpam-6638	187	8	fig	fig	NOUN
ejpam-6638	187	9	.	.	PUNCT
ejpam-6638	188	1	(	(	PUNCT
ejpam-6638	188	2	2	2	NUM
ejpam-6638	188	3	)	)	PUNCT
ejpam-6638	188	4	,	,	PUNCT
ejpam-6638	188	5	while	while	SCONJ
ejpam-6638	188	6	the	the	DET
ejpam-6638	188	7	images	image	NOUN
ejpam-6638	188	8	for	for	ADP
ejpam-6638	188	9	inpainting	inpainte	VERB
ejpam-6638	188	10	are	be	AUX
ejpam-6638	188	11	presented	present	VERB
ejpam-6638	188	12	in	in	ADP
ejpam-6638	188	13	fig	fig	NOUN
ejpam-6638	188	14	.	.	PUNCT
ejpam-6638	189	1	(	(	PUNCT
ejpam-6638	189	2	3	3	NUM
ejpam-6638	189	3	)	)	PUNCT
ejpam-6638	189	4	.	.	PUNCT
ejpam-6638	190	1	performance	performance	NOUN
ejpam-6638	190	2	is	be	AUX
ejpam-6638	190	3	evaluated	evaluate	VERB
ejpam-6638	190	4	using	use	VERB
ejpam-6638	190	5	psnr	psnr	NOUN
ejpam-6638	190	6	,	,	PUNCT
ejpam-6638	190	7	ssim	ssim	NOUN
ejpam-6638	190	8	,	,	PUNCT
ejpam-6638	190	9	and	and	CCONJ
ejpam-6638	190	10	visual	visual	ADJ
ejpam-6638	190	11	inspection	inspection	NOUN
ejpam-6638	190	12	.	.	PUNCT
ejpam-6638	191	1	the	the	DET
ejpam-6638	191	2	proposed	propose	VERB
ejpam-6638	191	3	svd	svd	PROPN
ejpam-6638	191	4	-	-	PUNCT
ejpam-6638	191	5	pooling	pool	VERB
ejpam-6638	191	6	dip	dip	NOUN
ejpam-6638	191	7	consistently	consistently	ADV
ejpam-6638	191	8	improves	improve	VERB
ejpam-6638	191	9	over	over	ADP
ejpam-6638	191	10	the	the	DET
ejpam-6638	191	11	original	original	ADJ
ejpam-6638	191	12	dip	dip	NOUN
ejpam-6638	191	13	in	in	ADP
ejpam-6638	191	14	both	both	CCONJ
ejpam-6638	191	15	quantitative	quantitative	ADJ
ejpam-6638	191	16	and	and	CCONJ
ejpam-6638	191	17	qualitative	qualitative	ADJ
ejpam-6638	191	18	metrics	metric	NOUN
ejpam-6638	191	19	.	.	PUNCT
ejpam-6638	192	1	all	all	DET
ejpam-6638	192	2	experiments	experiment	NOUN
ejpam-6638	192	3	were	be	AUX
ejpam-6638	192	4	carried	carry	VERB
ejpam-6638	192	5	out	out	ADP
ejpam-6638	192	6	using	use	VERB
ejpam-6638	192	7	the	the	DET
ejpam-6638	192	8	python	python	NOUN
ejpam-6638	192	9	programming	programming	NOUN
ejpam-6638	192	10	language	language	NOUN
ejpam-6638	192	11	.	.	PUNCT
ejpam-6638	193	1	the	the	DET
ejpam-6638	193	2	proposed	propose	VERB
ejpam-6638	193	3	model	model	NOUN
ejpam-6638	193	4	was	be	AUX
ejpam-6638	193	5	implemented	implement	VERB
ejpam-6638	193	6	in	in	ADP
ejpam-6638	193	7	the	the	DET
ejpam-6638	193	8	pytorch	pytorch	NOUN
ejpam-6638	193	9	deep	deep	ADJ
ejpam-6638	193	10	learning	learning	NOUN
ejpam-6638	193	11	framework	framework	NOUN
ejpam-6638	193	12	,	,	PUNCT
ejpam-6638	193	13	and	and	CCONJ
ejpam-6638	193	14	standard	standard	ADJ
ejpam-6638	193	15	scientific	scientific	ADJ
ejpam-6638	193	16	computing	computing	NOUN
ejpam-6638	193	17	libraries	library	NOUN
ejpam-6638	193	18	such	such	ADJ
ejpam-6638	193	19	as	as	ADP
ejpam-6638	193	20	numpy	numpy	NOUN
ejpam-6638	193	21	and	and	CCONJ
ejpam-6638	193	22	scipy	scipy	NOUN
ejpam-6638	193	23	were	be	AUX
ejpam-6638	193	24	used	use	VERB
ejpam-6638	193	25	for	for	ADP
ejpam-6638	193	26	tensor	tensor	NOUN
ejpam-6638	193	27	manipulation	manipulation	NOUN
ejpam-6638	193	28	and	and	CCONJ
ejpam-6638	193	29	numerical	numerical	ADJ
ejpam-6638	193	30	operations	operation	NOUN
ejpam-6638	193	31	.	.	PUNCT
ejpam-6638	194	1	all	all	DET
ejpam-6638	194	2	simulations	simulation	NOUN
ejpam-6638	194	3	and	and	CCONJ
ejpam-6638	194	4	training	training	NOUN
ejpam-6638	194	5	were	be	AUX
ejpam-6638	194	6	performed	perform	VERB
ejpam-6638	194	7	on	on	ADP
ejpam-6638	194	8	a	a	DET
ejpam-6638	194	9	cpu	cpu	NOUN
ejpam-6638	194	10	-	-	PUNCT
ejpam-6638	194	11	based	base	VERB
ejpam-6638	194	12	system	system	NOUN
ejpam-6638	194	13	,	,	PUNCT
ejpam-6638	194	14	without	without	ADP
ejpam-6638	194	15	the	the	DET
ejpam-6638	194	16	use	use	NOUN
ejpam-6638	194	17	of	of	ADP
ejpam-6638	194	18	gpu	gpu	NOUN
ejpam-6638	194	19	acceleration	acceleration	NOUN
ejpam-6638	194	20	.	.	PUNCT
ejpam-6638	195	1	the	the	DET
ejpam-6638	195	2	code	code	NOUN
ejpam-6638	195	3	was	be	AUX
ejpam-6638	195	4	executed	execute	VERB
ejpam-6638	195	5	in	in	ADP
ejpam-6638	195	6	a	a	DET
ejpam-6638	195	7	standard	standard	ADJ
ejpam-6638	195	8	python	python	NOUN
ejpam-6638	195	9	environment	environment	NOUN
ejpam-6638	195	10	to	to	PART
ejpam-6638	195	11	ensure	ensure	VERB
ejpam-6638	195	12	full	full	ADJ
ejpam-6638	195	13	reproducibility	reproducibility	NOUN
ejpam-6638	195	14	.	.	PUNCT
ejpam-6638	196	1	(	(	PUNCT
ejpam-6638	196	2	a	a	X
ejpam-6638	196	3	)	)	PUNCT
ejpam-6638	196	4	(	(	PUNCT
ejpam-6638	196	5	b	b	X
ejpam-6638	196	6	)	)	PUNCT
ejpam-6638	196	7	(	(	PUNCT
ejpam-6638	196	8	c	c	X
ejpam-6638	196	9	)	)	PUNCT
ejpam-6638	196	10	(	(	PUNCT
ejpam-6638	196	11	d	d	X
ejpam-6638	196	12	)	)	PUNCT
ejpam-6638	196	13	(	(	PUNCT
ejpam-6638	196	14	e	e	NOUN
ejpam-6638	196	15	)	)	PUNCT
ejpam-6638	196	16	(	(	PUNCT
ejpam-6638	196	17	f	f	X
ejpam-6638	196	18	)	)	PUNCT
ejpam-6638	196	19	figure	figure	NOUN
ejpam-6638	196	20	2	2	NUM
ejpam-6638	196	21	:	:	PUNCT
ejpam-6638	196	22	the	the	DET
ejpam-6638	196	23	set	set	NOUN
ejpam-6638	196	24	of	of	ADP
ejpam-6638	196	25	6	6	NUM
ejpam-6638	196	26	images	image	NOUN
ejpam-6638	196	27	used	use	VERB
ejpam-6638	196	28	in	in	ADP
ejpam-6638	196	29	experiments	experiment	NOUN
ejpam-6638	196	30	for	for	ADP
ejpam-6638	196	31	denoising	denoise	VERB
ejpam-6638	196	32	4.1	4.1	NUM
ejpam-6638	196	33	.	.	PUNCT
ejpam-6638	197	1	image	image	NOUN
ejpam-6638	197	2	denoising	denoise	VERB
ejpam-6638	197	3	in	in	ADP
ejpam-6638	197	4	this	this	DET
ejpam-6638	197	5	section	section	NOUN
ejpam-6638	197	6	,	,	PUNCT
ejpam-6638	197	7	we	we	PRON
ejpam-6638	197	8	evaluate	evaluate	VERB
ejpam-6638	197	9	the	the	DET
ejpam-6638	197	10	effectiveness	effectiveness	NOUN
ejpam-6638	197	11	of	of	ADP
ejpam-6638	197	12	the	the	DET
ejpam-6638	197	13	svd	svd	PROPN
ejpam-6638	197	14	-	-	PUNCT
ejpam-6638	197	15	dip	dip	NOUN
ejpam-6638	197	16	method	method	NOUN
ejpam-6638	197	17	for	for	ADP
ejpam-6638	197	18	tackling	tackle	VERB
ejpam-6638	197	19	imagedenoising	imagedenoise	VERB
ejpam-6638	197	20	challenges	challenge	NOUN
ejpam-6638	197	21	.	.	PUNCT
ejpam-6638	198	1	for	for	ADP
ejpam-6638	198	2	the	the	DET
ejpam-6638	198	3	images	image	NOUN
ejpam-6638	198	4	illustrated	illustrate	VERB
ejpam-6638	198	5	in	in	ADP
ejpam-6638	198	6	figure	figure	NOUN
ejpam-6638	198	7	(	(	PUNCT
ejpam-6638	198	8	2	2	NUM
ejpam-6638	198	9	)	)	PUNCT
ejpam-6638	198	10	,	,	PUNCT
ejpam-6638	198	11	we	we	PRON
ejpam-6638	198	12	utilize	utilize	VERB
ejpam-6638	198	13	the	the	DET
ejpam-6638	198	14	cnn	cnn	PROPN
ejpam-6638	198	15	architecture	architecture	NOUN
ejpam-6638	198	16	m.	m.	NOUN
ejpam-6638	198	17	israr	israr	PROPN
ejpam-6638	198	18	et	et	PROPN
ejpam-6638	198	19	al	al	PROPN
ejpam-6638	198	20	.	.	PUNCT
ejpam-6638	198	21	/	/	SYM
ejpam-6638	198	22	eur	eur	PROPN
ejpam-6638	198	23	.	.	PUNCT
ejpam-6638	199	1	j.	j.	PROPN
ejpam-6638	199	2	pure	pure	PROPN
ejpam-6638	199	3	appl	appl	PROPN
ejpam-6638	199	4	.	.	PROPN
ejpam-6638	199	5	math	math	PROPN
ejpam-6638	199	6	,	,	PUNCT
ejpam-6638	199	7	18	18	NUM
ejpam-6638	199	8	(	(	PUNCT
ejpam-6638	199	9	4	4	NUM
ejpam-6638	199	10	)	)	PUNCT
ejpam-6638	199	11	(	(	PUNCT
ejpam-6638	199	12	2025	2025	NUM
ejpam-6638	199	13	)	)	PUNCT
ejpam-6638	199	14	,	,	PUNCT
ejpam-6638	199	15	6638	6638	NUM
ejpam-6638	199	16	9	9	NUM
ejpam-6638	199	17	of	of	ADP
ejpam-6638	199	18	22	22	NUM
ejpam-6638	199	19	(	(	PUNCT
ejpam-6638	199	20	a	a	NOUN
ejpam-6638	199	21	)	)	PUNCT
ejpam-6638	199	22	(	(	PUNCT
ejpam-6638	199	23	b	b	X
ejpam-6638	199	24	)	)	PUNCT
ejpam-6638	199	25	(	(	PUNCT
ejpam-6638	199	26	c	c	X
ejpam-6638	199	27	)	)	PUNCT
ejpam-6638	199	28	(	(	PUNCT
ejpam-6638	199	29	d	d	X
ejpam-6638	199	30	)	)	PUNCT
ejpam-6638	199	31	(	(	PUNCT
ejpam-6638	199	32	e	e	NOUN
ejpam-6638	199	33	)	)	PUNCT
ejpam-6638	199	34	(	(	PUNCT
ejpam-6638	199	35	f	f	X
ejpam-6638	199	36	)	)	PUNCT
ejpam-6638	199	37	(	(	PUNCT
ejpam-6638	199	38	g	g	NOUN
ejpam-6638	199	39	)	)	PUNCT
ejpam-6638	199	40	(	(	PUNCT
ejpam-6638	199	41	h	h	NOUN
ejpam-6638	199	42	)	)	PUNCT
ejpam-6638	199	43	figure	figure	NOUN
ejpam-6638	199	44	3	3	NUM
ejpam-6638	199	45	:	:	PUNCT
ejpam-6638	199	46	set	set	VERB
ejpam-6638	199	47	of	of	ADP
ejpam-6638	199	48	8	8	NUM
ejpam-6638	199	49	images	image	NOUN
ejpam-6638	199	50	used	use	VERB
ejpam-6638	199	51	in	in	ADP
ejpam-6638	199	52	the	the	DET
ejpam-6638	199	53	inpainting	inpainting	ADJ
ejpam-6638	199	54	experiment	experiment	NOUN
ejpam-6638	199	55	:	:	PUNCT
ejpam-6638	199	56	(	(	PUNCT
ejpam-6638	199	57	a	a	X
ejpam-6638	199	58	–	–	PUNCT
ejpam-6638	199	59	d	d	NOUN
ejpam-6638	199	60	)	)	PUNCT
ejpam-6638	199	61	original	original	ADJ
ejpam-6638	199	62	images	image	NOUN
ejpam-6638	199	63	;	;	PUNCT
ejpam-6638	199	64	(	(	PUNCT
ejpam-6638	199	65	e	e	X
ejpam-6638	199	66	–	–	PUNCT
ejpam-6638	199	67	h	h	NOUN
ejpam-6638	199	68	)	)	PUNCT
ejpam-6638	199	69	corresponding	correspond	VERB
ejpam-6638	199	70	masks	mask	NOUN
ejpam-6638	199	71	.	.	PUNCT
ejpam-6638	200	1	depicted	depict	VERB
ejpam-6638	200	2	in	in	ADP
ejpam-6638	200	3	figure	figure	NOUN
ejpam-6638	200	4	(	(	PUNCT
ejpam-6638	200	5	4	4	NUM
ejpam-6638	200	6	)	)	PUNCT
ejpam-6638	200	7	,	,	PUNCT
ejpam-6638	200	8	with	with	ADP
ejpam-6638	200	9	each	each	DET
ejpam-6638	200	10	skip	skip	ADJ
ejpam-6638	200	11	layer	layer	NOUN
ejpam-6638	200	12	set	set	VERB
ejpam-6638	200	13	to	to	ADP
ejpam-6638	200	14	ns[i	ns[i	VERB
ejpam-6638	200	15	]	]	X
ejpam-6638	200	16	=	=	SYM
ejpam-6638	200	17	4	4	X
ejpam-6638	200	18	.	.	PUNCT
ejpam-6638	201	1	all	all	DET
ejpam-6638	201	2	hyperparameters	hyperparameter	NOUN
ejpam-6638	201	3	for	for	ADP
ejpam-6638	201	4	the	the	DET
ejpam-6638	201	5	algorithms	algorithm	NOUN
ejpam-6638	201	6	were	be	AUX
ejpam-6638	201	7	carefully	carefully	ADV
ejpam-6638	201	8	fine	fine	ADV
ejpam-6638	201	9	-	-	PUNCT
ejpam-6638	201	10	tuned	tune	VERB
ejpam-6638	201	11	in	in	ADP
ejpam-6638	201	12	every	every	DET
ejpam-6638	201	13	experiment	experiment	NOUN
ejpam-6638	201	14	to	to	PART
ejpam-6638	201	15	attain	attain	VERB
ejpam-6638	201	16	the	the	DET
ejpam-6638	201	17	highest	high	ADJ
ejpam-6638	201	18	possible	possible	ADJ
ejpam-6638	201	19	peaksignal	peaksignal	ADJ
ejpam-6638	201	20	-	-	PUNCT
ejpam-6638	201	21	to	to	ADP
ejpam-6638	201	22	-	-	PUNCT
ejpam-6638	201	23	noise	noise	NOUN
ejpam-6638	201	24	ratio	ratio	NOUN
ejpam-6638	201	25	(	(	PUNCT
ejpam-6638	201	26	psnr	psnr	NOUN
ejpam-6638	201	27	)	)	PUNCT
ejpam-6638	201	28	and	and	CCONJ
ejpam-6638	201	29	structural	structural	ADJ
ejpam-6638	201	30	similarity	similarity	NOUN
ejpam-6638	201	31	index	index	NOUN
ejpam-6638	201	32	measure	measure	NOUN
ejpam-6638	201	33	(	(	PUNCT
ejpam-6638	201	34	ssim	ssim	NOUN
ejpam-6638	201	35	)	)	PUNCT
ejpam-6638	201	36	in	in	ADP
ejpam-6638	201	37	comparison	comparison	NOUN
ejpam-6638	201	38	to	to	ADP
ejpam-6638	201	39	the	the	DET
ejpam-6638	201	40	ground	ground	NOUN
ejpam-6638	201	41	truth	truth	NOUN
ejpam-6638	201	42	test	test	NOUN
ejpam-6638	201	43	images	image	NOUN
ejpam-6638	201	44	.	.	PUNCT
ejpam-6638	202	1	we	we	PRON
ejpam-6638	202	2	provide	provide	VERB
ejpam-6638	202	3	the	the	DET
ejpam-6638	202	4	average	average	ADJ
ejpam-6638	202	5	psnr	psnr	NOUN
ejpam-6638	202	6	,	,	PUNCT
ejpam-6638	202	7	which	which	PRON
ejpam-6638	202	8	indicates	indicate	VERB
ejpam-6638	202	9	the	the	DET
ejpam-6638	202	10	psnr	psnr	NOUN
ejpam-6638	202	11	values	value	NOUN
ejpam-6638	202	12	averaged	average	VERB
ejpam-6638	202	13	over	over	ADP
ejpam-6638	202	14	the	the	DET
ejpam-6638	202	15	relevant	relevant	ADJ
ejpam-6638	202	16	set	set	NOUN
ejpam-6638	202	17	of	of	ADP
ejpam-6638	202	18	test	test	NOUN
ejpam-6638	202	19	images	image	NOUN
ejpam-6638	202	20	.	.	PUNCT
ejpam-6638	203	1	(	(	PUNCT
ejpam-6638	203	2	a	a	X
ejpam-6638	203	3	)	)	PUNCT
ejpam-6638	203	4	(	(	PUNCT
ejpam-6638	203	5	b	b	X
ejpam-6638	203	6	)	)	PUNCT
ejpam-6638	203	7	(	(	PUNCT
ejpam-6638	203	8	c	c	X
ejpam-6638	203	9	)	)	PUNCT
ejpam-6638	203	10	(	(	PUNCT
ejpam-6638	203	11	d	d	X
ejpam-6638	203	12	)	)	PUNCT
ejpam-6638	203	13	(	(	PUNCT
ejpam-6638	203	14	e	e	NOUN
ejpam-6638	203	15	)	)	PUNCT
ejpam-6638	203	16	(	(	PUNCT
ejpam-6638	203	17	f	f	X
ejpam-6638	203	18	)	)	PUNCT
ejpam-6638	203	19	(	(	PUNCT
ejpam-6638	203	20	g	g	NOUN
ejpam-6638	203	21	)	)	PUNCT
ejpam-6638	203	22	(	(	PUNCT
ejpam-6638	203	23	h	h	NOUN
ejpam-6638	203	24	)	)	PUNCT
ejpam-6638	203	25	(	(	PUNCT
ejpam-6638	203	26	i	i	NOUN
ejpam-6638	203	27	)	)	PUNCT
ejpam-6638	203	28	(	(	PUNCT
ejpam-6638	203	29	j	j	NOUN
ejpam-6638	203	30	)	)	PUNCT
ejpam-6638	203	31	(	(	PUNCT
ejpam-6638	203	32	k	k	NOUN
ejpam-6638	203	33	)	)	PUNCT
ejpam-6638	203	34	(	(	PUNCT
ejpam-6638	203	35	l	l	NOUN
ejpam-6638	203	36	)	)	PUNCT
ejpam-6638	203	37	figure	figure	NOUN
ejpam-6638	203	38	4	4	NUM
ejpam-6638	203	39	:	:	PUNCT
ejpam-6638	203	40	for	for	ADP
ejpam-6638	203	41	a	a	DET
ejpam-6638	203	42	visual	visual	ADJ
ejpam-6638	203	43	comparison	comparison	NOUN
ejpam-6638	203	44	of	of	ADP
ejpam-6638	203	45	denoising	denoise	VERB
ejpam-6638	203	46	with	with	ADP
ejpam-6638	203	47	our	our	PRON
ejpam-6638	203	48	svd	svd	PROPN
ejpam-6638	203	49	-	-	PUNCT
ejpam-6638	203	50	based	base	VERB
ejpam-6638	203	51	deep	deep	ADJ
ejpam-6638	203	52	image	image	NOUN
ejpam-6638	203	53	prior	prior	ADV
ejpam-6638	203	54	,	,	PUNCT
ejpam-6638	203	55	the	the	DET
ejpam-6638	203	56	first	first	ADJ
ejpam-6638	203	57	row	row	NOUN
ejpam-6638	203	58	shows	show	VERB
ejpam-6638	203	59	the	the	DET
ejpam-6638	203	60	noisy	noisy	ADJ
ejpam-6638	203	61	input	input	NOUN
ejpam-6638	203	62	images	image	NOUN
ejpam-6638	203	63	to	to	ADP
ejpam-6638	203	64	dip	dip	NOUN
ejpam-6638	203	65	-	-	PUNCT
ejpam-6638	203	66	svd	svd	PROPN
ejpam-6638	203	67	(	(	PUNCT
ejpam-6638	203	68	noise	noise	NOUN
ejpam-6638	203	69	ratio	ratio	NOUN
ejpam-6638	203	70	=	=	SYM
ejpam-6638	203	71	45	45	NUM
ejpam-6638	203	72	)	)	PUNCT
ejpam-6638	203	73	,	,	PUNCT
ejpam-6638	203	74	and	and	CCONJ
ejpam-6638	203	75	the	the	DET
ejpam-6638	203	76	second	second	ADJ
ejpam-6638	203	77	row	row	NOUN
ejpam-6638	203	78	shows	show	VERB
ejpam-6638	203	79	the	the	DET
ejpam-6638	203	80	corresponding	corresponding	ADJ
ejpam-6638	203	81	denoised	denoise	VERB
ejpam-6638	203	82	outputs	output	NOUN
ejpam-6638	203	83	produced	produce	VERB
ejpam-6638	203	84	by	by	ADP
ejpam-6638	203	85	dip	dip	NOUN
ejpam-6638	203	86	-	-	PUNCT
ejpam-6638	203	87	svd	svd	PROPN
ejpam-6638	203	88	.	.	PUNCT
ejpam-6638	204	1	visual	visual	ADJ
ejpam-6638	204	2	results	result	NOUN
ejpam-6638	204	3	of	of	ADP
ejpam-6638	204	4	dip	dip	NOUN
ejpam-6638	204	5	-	-	PUNCT
ejpam-6638	204	6	svd	svd	NOUN
ejpam-6638	204	7	figure	figure	VERB
ejpam-6638	204	8	4	4	NUM
ejpam-6638	204	9	displays	display	VERB
ejpam-6638	204	10	the	the	DET
ejpam-6638	204	11	qualitative	qualitative	ADJ
ejpam-6638	204	12	results	result	NOUN
ejpam-6638	204	13	of	of	ADP
ejpam-6638	204	14	the	the	DET
ejpam-6638	204	15	proposed	propose	VERB
ejpam-6638	204	16	deep	deep	ADJ
ejpam-6638	204	17	image	image	NOUN
ejpam-6638	204	18	prior	prior	ADV
ejpam-6638	204	19	with	with	ADP
ejpam-6638	204	20	svd	svd	PROPN
ejpam-6638	204	21	pooling	pool	VERB
ejpam-6638	204	22	(	(	PUNCT
ejpam-6638	204	23	dip	dip	NOUN
ejpam-6638	204	24	-	-	PUNCT
ejpam-6638	204	25	svd	svd	PROPN
ejpam-6638	204	26	)	)	PUNCT
ejpam-6638	204	27	on	on	ADP
ejpam-6638	204	28	a	a	DET
ejpam-6638	204	29	set	set	NOUN
ejpam-6638	204	30	of	of	ADP
ejpam-6638	204	31	test	test	NOUN
ejpam-6638	204	32	images	image	NOUN
ejpam-6638	204	33	contaminated	contaminate	VERB
ejpam-6638	204	34	with	with	ADP
ejpam-6638	204	35	gaussian	gaussian	ADJ
ejpam-6638	204	36	noise	noise	NOUN
ejpam-6638	204	37	(	(	PUNCT
ejpam-6638	204	38	noise	noise	NOUN
ejpam-6638	204	39	ratio	ratio	NOUN
ejpam-6638	204	40	=	=	SYM
ejpam-6638	204	41	45	45	NUM
ejpam-6638	204	42	)	)	PUNCT
ejpam-6638	204	43	.	.	PUNCT
ejpam-6638	205	1	the	the	DET
ejpam-6638	205	2	top	top	ADJ
ejpam-6638	205	3	row	row	NOUN
ejpam-6638	205	4	of	of	ADP
ejpam-6638	205	5	the	the	DET
ejpam-6638	205	6	noisy	noisy	ADJ
ejpam-6638	205	7	input	input	NOUN
ejpam-6638	205	8	photographs	photograph	NOUN
ejpam-6638	205	9	clearly	clearly	ADV
ejpam-6638	205	10	shows	show	VERB
ejpam-6638	205	11	how	how	SCONJ
ejpam-6638	205	12	much	much	ADJ
ejpam-6638	205	13	m.	m.	NOUN
ejpam-6638	205	14	israr	israr	NOUN
ejpam-6638	205	15	et	et	PROPN
ejpam-6638	205	16	al	al	PROPN
ejpam-6638	205	17	.	.	PUNCT
ejpam-6638	205	18	/	/	SYM
ejpam-6638	205	19	eur	eur	PROPN
ejpam-6638	205	20	.	.	PUNCT
ejpam-6638	206	1	j.	j.	PROPN
ejpam-6638	206	2	pure	pure	PROPN
ejpam-6638	206	3	appl	appl	PROPN
ejpam-6638	206	4	.	.	PROPN
ejpam-6638	206	5	math	math	PROPN
ejpam-6638	206	6	,	,	PUNCT
ejpam-6638	206	7	18	18	NUM
ejpam-6638	206	8	(	(	PUNCT
ejpam-6638	206	9	4	4	NUM
ejpam-6638	206	10	)	)	PUNCT
ejpam-6638	206	11	(	(	PUNCT
ejpam-6638	206	12	2025	2025	NUM
ejpam-6638	206	13	)	)	PUNCT
ejpam-6638	206	14	,	,	PUNCT
ejpam-6638	206	15	6638	6638	NUM
ejpam-6638	206	16	10	10	NUM
ejpam-6638	206	17	of	of	ADP
ejpam-6638	206	18	22	22	NUM
ejpam-6638	206	19	the	the	DET
ejpam-6638	206	20	noise	noise	NOUN
ejpam-6638	206	21	degrades	degrade	VERB
ejpam-6638	206	22	the	the	DET
ejpam-6638	206	23	image	image	NOUN
ejpam-6638	206	24	quality	quality	NOUN
ejpam-6638	206	25	.	.	PUNCT
ejpam-6638	207	1	the	the	DET
ejpam-6638	207	2	matching	matching	NOUN
ejpam-6638	207	3	denoised	denoised	AUX
ejpam-6638	207	4	outputs	output	NOUN
ejpam-6638	207	5	produced	produce	VERB
ejpam-6638	207	6	by	by	ADP
ejpam-6638	207	7	dip	dip	NOUN
ejpam-6638	207	8	-	-	PUNCT
ejpam-6638	207	9	svd	svd	PROPN
ejpam-6638	207	10	are	be	AUX
ejpam-6638	207	11	shown	show	VERB
ejpam-6638	207	12	in	in	ADP
ejpam-6638	207	13	the	the	DET
ejpam-6638	207	14	bottom	bottom	NOUN
ejpam-6638	207	15	row	row	NOUN
ejpam-6638	207	16	after	after	ADP
ejpam-6638	207	17	3,000	3,000	NUM
ejpam-6638	207	18	training	training	NOUN
ejpam-6638	207	19	epochs	epoch	NOUN
ejpam-6638	207	20	.	.	PUNCT
ejpam-6638	208	1	dip	dip	NOUN
ejpam-6638	208	2	-	-	PUNCT
ejpam-6638	208	3	svd	svd	PROPN
ejpam-6638	208	4	can	can	AUX
ejpam-6638	208	5	effectively	effectively	ADV
ejpam-6638	208	6	remove	remove	VERB
ejpam-6638	208	7	the	the	DET
ejpam-6638	208	8	majority	majority	NOUN
ejpam-6638	208	9	of	of	ADP
ejpam-6638	208	10	noise	noise	NOUN
ejpam-6638	208	11	artifacts	artifact	NOUN
ejpam-6638	208	12	while	while	SCONJ
ejpam-6638	208	13	maintaining	maintain	VERB
ejpam-6638	208	14	fine	fine	ADJ
ejpam-6638	208	15	structural	structural	ADJ
ejpam-6638	208	16	features	feature	NOUN
ejpam-6638	208	17	and	and	CCONJ
ejpam-6638	208	18	texturing	texturing	NOUN
ejpam-6638	208	19	.	.	PUNCT
ejpam-6638	209	1	edges	edge	NOUN
ejpam-6638	209	2	and	and	CCONJ
ejpam-6638	209	3	curves	curve	NOUN
ejpam-6638	209	4	appear	appear	VERB
ejpam-6638	209	5	sharp	sharp	ADJ
ejpam-6638	209	6	and	and	CCONJ
ejpam-6638	209	7	continuous	continuous	ADJ
ejpam-6638	209	8	,	,	PUNCT
ejpam-6638	209	9	while	while	SCONJ
ejpam-6638	209	10	homogeneous	homogeneous	ADJ
ejpam-6638	209	11	areas	area	NOUN
ejpam-6638	209	12	are	be	AUX
ejpam-6638	209	13	given	give	VERB
ejpam-6638	209	14	a	a	DET
ejpam-6638	209	15	smooth	smooth	ADJ
ejpam-6638	209	16	appearance	appearance	NOUN
ejpam-6638	209	17	without	without	ADP
ejpam-6638	209	18	oversmoothing	oversmoothe	VERB
ejpam-6638	209	19	important	important	ADJ
ejpam-6638	209	20	features	feature	NOUN
ejpam-6638	209	21	.	.	PUNCT
ejpam-6638	210	1	for	for	ADP
ejpam-6638	210	2	a	a	DET
ejpam-6638	210	3	range	range	NOUN
ejpam-6638	210	4	of	of	ADP
ejpam-6638	210	5	image	image	NOUN
ejpam-6638	210	6	content	content	NOUN
ejpam-6638	210	7	,	,	PUNCT
ejpam-6638	210	8	this	this	DET
ejpam-6638	210	9	figure	figure	NOUN
ejpam-6638	210	10	shows	show	VERB
ejpam-6638	210	11	how	how	SCONJ
ejpam-6638	210	12	well	well	ADV
ejpam-6638	210	13	the	the	DET
ejpam-6638	210	14	svd	svd	PROPN
ejpam-6638	210	15	-	-	PUNCT
ejpam-6638	210	16	enhanced	enhance	VERB
ejpam-6638	210	17	prior	prior	ADJ
ejpam-6638	210	18	balances	balance	NOUN
ejpam-6638	210	19	noise	noise	NOUN
ejpam-6638	210	20	reduction	reduction	NOUN
ejpam-6638	210	21	and	and	CCONJ
ejpam-6638	210	22	feature	feature	NOUN
ejpam-6638	210	23	retention	retention	NOUN
ejpam-6638	210	24	.	.	PUNCT
ejpam-6638	211	1	images	image	NOUN
ejpam-6638	211	2	psnr	psnr	NOUN
ejpam-6638	211	3	ssim	ssim	NOUN
ejpam-6638	211	4	1	1	NUM
ejpam-6638	211	5	25.07	25.07	NUM
ejpam-6638	211	6	0.87	0.87	NUM
ejpam-6638	211	7	2	2	NUM
ejpam-6638	211	8	32.63	32.63	NUM
ejpam-6638	211	9	0.97	0.97	NUM
ejpam-6638	211	10	3	3	NUM
ejpam-6638	211	11	24.72	24.72	NUM
ejpam-6638	211	12	0.90	0.90	NUM
ejpam-6638	211	13	4	4	NUM
ejpam-6638	211	14	26.66	26.66	NUM
ejpam-6638	211	15	0.87	0.87	NUM
ejpam-6638	211	16	5	5	NUM
ejpam-6638	211	17	31.59	31.59	NUM
ejpam-6638	211	18	0.90	0.90	NUM
ejpam-6638	211	19	6	6	NUM
ejpam-6638	211	20	24.45	24.45	NUM
ejpam-6638	211	21	0.92	0.92	NUM
ejpam-6638	211	22	table	table	NOUN
ejpam-6638	211	23	1	1	NUM
ejpam-6638	211	24	:	:	PUNCT
ejpam-6638	211	25	denoising	denoise	VERB
ejpam-6638	211	26	result	result	NOUN
ejpam-6638	211	27	i.e.	i.e.	X
ejpam-6638	211	28	psnr	psnr	NOUN
ejpam-6638	211	29	and	and	CCONJ
ejpam-6638	211	30	ssim	ssim	NOUN
ejpam-6638	211	31	of	of	ADP
ejpam-6638	211	32	6	6	NUM
ejpam-6638	211	33	images	image	NOUN
ejpam-6638	211	34	by	by	ADP
ejpam-6638	211	35	using	use	VERB
ejpam-6638	211	36	dip	dip	NOUN
ejpam-6638	211	37	-	-	PUNCT
ejpam-6638	211	38	svd	svd	PROPN
ejpam-6638	211	39	over	over	ADP
ejpam-6638	211	40	3000	3000	NUM
ejpam-6638	211	41	epochs	epoch	NOUN
ejpam-6638	211	42	denoising	denoise	VERB
ejpam-6638	211	43	performance	performance	NOUN
ejpam-6638	211	44	analysis	analysis	NOUN
ejpam-6638	211	45	the	the	DET
ejpam-6638	211	46	performance	performance	NOUN
ejpam-6638	211	47	of	of	ADP
ejpam-6638	211	48	dip	dip	NOUN
ejpam-6638	211	49	-	-	PUNCT
ejpam-6638	211	50	svd	svd	NOUN
ejpam-6638	211	51	in	in	ADP
ejpam-6638	211	52	reducing	reduce	VERB
ejpam-6638	211	53	noise	noise	NOUN
ejpam-6638	211	54	after	after	SCONJ
ejpam-6638	211	55	3,000	3,000	NUM
ejpam-6638	211	56	training	training	NOUN
ejpam-6638	211	57	epochs	epoch	NOUN
ejpam-6638	211	58	varies	vary	VERB
ejpam-6638	211	59	for	for	ADP
ejpam-6638	211	60	the	the	DET
ejpam-6638	211	61	six	six	NUM
ejpam-6638	211	62	test	test	NOUN
ejpam-6638	211	63	images	image	NOUN
ejpam-6638	211	64	,	,	PUNCT
ejpam-6638	211	65	as	as	SCONJ
ejpam-6638	211	66	shown	show	VERB
ejpam-6638	211	67	in	in	ADP
ejpam-6638	211	68	table	table	NOUN
ejpam-6638	211	69	1	1	NUM
ejpam-6638	211	70	.	.	PUNCT
ejpam-6638	212	1	while	while	SCONJ
ejpam-6638	212	2	some	some	DET
ejpam-6638	212	3	images	image	NOUN
ejpam-6638	212	4	gain	gain	VERB
ejpam-6638	212	5	more	more	ADV
ejpam-6638	212	6	from	from	ADP
ejpam-6638	212	7	the	the	DET
ejpam-6638	212	8	svd	svd	PROPN
ejpam-6638	212	9	-	-	PUNCT
ejpam-6638	212	10	based	base	VERB
ejpam-6638	212	11	prior	prior	ADV
ejpam-6638	212	12	than	than	ADP
ejpam-6638	212	13	others	other	NOUN
ejpam-6638	212	14	,	,	PUNCT
ejpam-6638	212	15	all	all	PRON
ejpam-6638	212	16	images	image	NOUN
ejpam-6638	212	17	experience	experience	VERB
ejpam-6638	212	18	notable	notable	ADJ
ejpam-6638	212	19	noise	noise	NOUN
ejpam-6638	212	20	reduction	reduction	NOUN
ejpam-6638	212	21	,	,	PUNCT
ejpam-6638	212	22	with	with	ADP
ejpam-6638	212	23	psnr	psnr	NOUN
ejpam-6638	212	24	values	value	NOUN
ejpam-6638	212	25	stretching	stretch	VERB
ejpam-6638	212	26	from	from	ADP
ejpam-6638	212	27	24.45	24.45	NUM
ejpam-6638	212	28	db	db	PROPN
ejpam-6638	212	29	(	(	PUNCT
ejpam-6638	212	30	image	image	NOUN
ejpam-6638	212	31	6	6	NUM
ejpam-6638	212	32	)	)	PUNCT
ejpam-6638	212	33	to	to	ADP
ejpam-6638	212	34	32.63	32.63	NUM
ejpam-6638	212	35	db	db	PROPN
ejpam-6638	212	36	(	(	PUNCT
ejpam-6638	212	37	image	image	NOUN
ejpam-6638	212	38	2	2	NUM
ejpam-6638	212	39	)	)	PUNCT
ejpam-6638	212	40	.	.	PUNCT
ejpam-6638	213	1	with	with	ADP
ejpam-6638	213	2	the	the	DET
ejpam-6638	213	3	highest	high	ADJ
ejpam-6638	213	4	ssim	ssim	NOUN
ejpam-6638	213	5	(	(	PUNCT
ejpam-6638	213	6	0.97	0.97	NUM
ejpam-6638	213	7	)	)	PUNCT
ejpam-6638	213	8	and	and	CCONJ
ejpam-6638	213	9	psnr	psnr	NOUN
ejpam-6638	213	10	(	(	PUNCT
ejpam-6638	213	11	32.63	32.63	NUM
ejpam-6638	213	12	db	db	NOUN
ejpam-6638	213	13	)	)	PUNCT
ejpam-6638	213	14	,	,	PUNCT
ejpam-6638	213	15	image	image	NOUN
ejpam-6638	213	16	2	2	NUM
ejpam-6638	213	17	in	in	ADP
ejpam-6638	213	18	particular	particular	ADJ
ejpam-6638	213	19	demonstrates	demonstrate	VERB
ejpam-6638	213	20	a	a	DET
ejpam-6638	213	21	substantial	substantial	ADJ
ejpam-6638	213	22	improvement	improvement	NOUN
ejpam-6638	213	23	in	in	ADP
ejpam-6638	213	24	the	the	DET
ejpam-6638	213	25	recovery	recovery	NOUN
ejpam-6638	213	26	of	of	ADP
ejpam-6638	213	27	perceptual	perceptual	ADJ
ejpam-6638	213	28	structure	structure	NOUN
ejpam-6638	213	29	and	and	CCONJ
ejpam-6638	213	30	pixel	pixel	ADJ
ejpam-6638	213	31	-	-	PUNCT
ejpam-6638	213	32	level	level	NOUN
ejpam-6638	213	33	accuracy	accuracy	NOUN
ejpam-6638	213	34	.	.	PUNCT
ejpam-6638	214	1	in	in	ADP
ejpam-6638	214	2	contrast	contrast	NOUN
ejpam-6638	214	3	,	,	PUNCT
ejpam-6638	214	4	image	image	NOUN
ejpam-6638	214	5	6	6	NUM
ejpam-6638	214	6	shows	show	VERB
ejpam-6638	214	7	that	that	SCONJ
ejpam-6638	214	8	dip	dip	NOUN
ejpam-6638	214	9	-	-	PUNCT
ejpam-6638	214	10	svd	svd	PROPN
ejpam-6638	214	11	can	can	AUX
ejpam-6638	214	12	keep	keep	VERB
ejpam-6638	214	13	structural	structural	ADJ
ejpam-6638	214	14	details	detail	NOUN
ejpam-6638	214	15	even	even	ADV
ejpam-6638	214	16	with	with	ADP
ejpam-6638	214	17	a	a	DET
ejpam-6638	214	18	higher	high	ADJ
ejpam-6638	214	19	absolute	absolute	ADJ
ejpam-6638	214	20	error	error	NOUN
ejpam-6638	214	21	,	,	PUNCT
ejpam-6638	214	22	achieving	achieve	VERB
ejpam-6638	214	23	a	a	DET
ejpam-6638	214	24	high	high	ADJ
ejpam-6638	214	25	ssim	ssim	NOUN
ejpam-6638	214	26	of	of	ADP
ejpam-6638	214	27	0.92	0.92	NUM
ejpam-6638	214	28	while	while	SCONJ
ejpam-6638	214	29	recording	record	VERB
ejpam-6638	214	30	the	the	DET
ejpam-6638	214	31	lowest	low	ADJ
ejpam-6638	214	32	psnr	psnr	NOUN
ejpam-6638	214	33	at	at	ADP
ejpam-6638	214	34	24.45	24.45	NUM
ejpam-6638	214	35	db	db	NOUN
ejpam-6638	214	36	.	.	PUNCT
ejpam-6638	215	1	all	all	DET
ejpam-6638	215	2	tested	test	VERB
ejpam-6638	215	3	photos	photo	NOUN
ejpam-6638	215	4	had	have	VERB
ejpam-6638	215	5	ssim	ssim	NOUN
ejpam-6638	215	6	scores	score	NOUN
ejpam-6638	215	7	between	between	ADP
ejpam-6638	215	8	0.87	0.87	NUM
ejpam-6638	215	9	and	and	CCONJ
ejpam-6638	215	10	0.97	0.97	NUM
ejpam-6638	215	11	,	,	PUNCT
ejpam-6638	215	12	indicating	indicate	VERB
ejpam-6638	215	13	that	that	SCONJ
ejpam-6638	215	14	the	the	DET
ejpam-6638	215	15	svd	svd	PROPN
ejpam-6638	215	16	-	-	PUNCT
ejpam-6638	215	17	augmented	augment	VERB
ejpam-6638	215	18	deep	deep	ADJ
ejpam-6638	215	19	image	image	NOUN
ejpam-6638	215	20	prior	prior	ADV
ejpam-6638	215	21	consistently	consistently	ADV
ejpam-6638	215	22	achieves	achieve	VERB
ejpam-6638	215	23	a	a	DET
ejpam-6638	215	24	significant	significant	ADJ
ejpam-6638	215	25	level	level	NOUN
ejpam-6638	215	26	of	of	ADP
ejpam-6638	215	27	structural	structural	ADJ
ejpam-6638	215	28	similarity	similarity	NOUN
ejpam-6638	215	29	with	with	ADP
ejpam-6638	215	30	the	the	DET
ejpam-6638	215	31	original	original	ADJ
ejpam-6638	215	32	image	image	NOUN
ejpam-6638	215	33	.	.	PUNCT
ejpam-6638	216	1	this	this	PRON
ejpam-6638	216	2	illustrates	illustrate	VERB
ejpam-6638	216	3	how	how	SCONJ
ejpam-6638	216	4	effectively	effectively	ADV
ejpam-6638	216	5	the	the	DET
ejpam-6638	216	6	method	method	NOUN
ejpam-6638	216	7	balances	balance	VERB
ejpam-6638	216	8	feature	feature	VERB
ejpam-6638	216	9	preservation	preservation	NOUN
ejpam-6638	216	10	with	with	ADP
ejpam-6638	216	11	noise	noise	NOUN
ejpam-6638	216	12	reduction	reduction	NOUN
ejpam-6638	216	13	.	.	PUNCT
ejpam-6638	217	1	(	(	PUNCT
ejpam-6638	217	2	a	a	X
ejpam-6638	217	3	)	)	PUNCT
ejpam-6638	217	4	(	(	PUNCT
ejpam-6638	217	5	b	b	X
ejpam-6638	217	6	)	)	PUNCT
ejpam-6638	217	7	(	(	PUNCT
ejpam-6638	217	8	c	c	X
ejpam-6638	217	9	)	)	PUNCT
ejpam-6638	217	10	(	(	PUNCT
ejpam-6638	217	11	d	d	X
ejpam-6638	217	12	)	)	PUNCT
ejpam-6638	217	13	(	(	PUNCT
ejpam-6638	217	14	e	e	NOUN
ejpam-6638	217	15	)	)	PUNCT
ejpam-6638	217	16	(	(	PUNCT
ejpam-6638	217	17	f	f	X
ejpam-6638	217	18	)	)	PUNCT
ejpam-6638	217	19	(	(	PUNCT
ejpam-6638	217	20	g	g	NOUN
ejpam-6638	217	21	)	)	PUNCT
ejpam-6638	217	22	(	(	PUNCT
ejpam-6638	217	23	h	h	NOUN
ejpam-6638	217	24	)	)	PUNCT
ejpam-6638	217	25	figure	figure	NOUN
ejpam-6638	217	26	5	5	NUM
ejpam-6638	217	27	:	:	PUNCT
ejpam-6638	217	28	best	good	ADJ
ejpam-6638	217	29	runs	run	NOUN
ejpam-6638	217	30	based	base	VERB
ejpam-6638	217	31	on	on	ADP
ejpam-6638	217	32	learned	learn	VERB
ejpam-6638	217	33	weights	weight	NOUN
ejpam-6638	217	34	for	for	ADP
ejpam-6638	217	35	each	each	DET
ejpam-6638	217	36	image	image	NOUN
ejpam-6638	217	37	.	.	PUNCT
ejpam-6638	218	1	the	the	DET
ejpam-6638	218	2	first	first	ADJ
ejpam-6638	218	3	row	row	NOUN
ejpam-6638	218	4	shows	show	VERB
ejpam-6638	218	5	denoising	denoising	NOUN
ejpam-6638	218	6	results	result	NOUN
ejpam-6638	218	7	,	,	PUNCT
ejpam-6638	218	8	and	and	CCONJ
ejpam-6638	218	9	the	the	DET
ejpam-6638	218	10	second	second	ADJ
ejpam-6638	218	11	row	row	NOUN
ejpam-6638	218	12	presents	present	VERB
ejpam-6638	218	13	inpainting	inpainte	VERB
ejpam-6638	218	14	results	result	NOUN
ejpam-6638	218	15	.	.	PUNCT
ejpam-6638	219	1	m.	m.	NOUN
ejpam-6638	219	2	israr	israr	PROPN
ejpam-6638	219	3	et	et	PROPN
ejpam-6638	219	4	al	al	PROPN
ejpam-6638	219	5	.	.	PUNCT
ejpam-6638	219	6	/	/	SYM
ejpam-6638	219	7	eur	eur	PROPN
ejpam-6638	219	8	.	.	PUNCT
ejpam-6638	220	1	j.	j.	PROPN
ejpam-6638	220	2	pure	pure	PROPN
ejpam-6638	220	3	appl	appl	PROPN
ejpam-6638	220	4	.	.	PROPN
ejpam-6638	220	5	math	math	PROPN
ejpam-6638	220	6	,	,	PUNCT
ejpam-6638	220	7	18	18	NUM
ejpam-6638	220	8	(	(	PUNCT
ejpam-6638	220	9	4	4	NUM
ejpam-6638	220	10	)	)	PUNCT
ejpam-6638	220	11	(	(	PUNCT
ejpam-6638	220	12	2025	2025	NUM
ejpam-6638	220	13	)	)	PUNCT
ejpam-6638	220	14	,	,	PUNCT
ejpam-6638	220	15	6638	6638	NUM
ejpam-6638	220	16	11	11	NUM
ejpam-6638	220	17	of	of	ADP
ejpam-6638	220	18	22	22	NUM
ejpam-6638	220	19	4.2	4.2	NUM
ejpam-6638	220	20	.	.	PUNCT
ejpam-6638	221	1	inpainting	inpainte	VERB
ejpam-6638	221	2	in	in	ADP
ejpam-6638	221	3	the	the	DET
ejpam-6638	221	4	image	image	NOUN
ejpam-6638	221	5	inpainting	inpainte	VERB
ejpam-6638	221	6	problem	problem	NOUN
ejpam-6638	221	7	,	,	PUNCT
ejpam-6638	221	8	we	we	PRON
ejpam-6638	221	9	are	be	AUX
ejpam-6638	221	10	given	give	VERB
ejpam-6638	221	11	a	a	DET
ejpam-6638	221	12	partially	partially	ADV
ejpam-6638	221	13	observed	observe	VERB
ejpam-6638	221	14	image	image	NOUN
ejpam-6638	221	15	x0	x0	PROPN
ejpam-6638	221	16	∈	∈	PROPN
ejpam-6638	221	17	rh×w	rh×w	PROPN
ejpam-6638	221	18	and	and	CCONJ
ejpam-6638	221	19	a	a	DET
ejpam-6638	221	20	binary	binary	PROPN
ejpam-6638	221	21	mask	mask	NOUN
ejpam-6638	221	22	m	m	PROPN
ejpam-6638	221	23	∈	∈	PROPN
ejpam-6638	221	24	{	{	PUNCT
ejpam-6638	221	25	0	0	NUM
ejpam-6638	221	26	,	,	PUNCT
ejpam-6638	221	27	1}h×w	1}h×w	NUM
ejpam-6638	221	28	indicating	indicate	VERB
ejpam-6638	221	29	the	the	DET
ejpam-6638	221	30	locations	location	NOUN
ejpam-6638	221	31	of	of	ADP
ejpam-6638	221	32	known	know	VERB
ejpam-6638	221	33	pixels	pixel	NOUN
ejpam-6638	221	34	.	.	PUNCT
ejpam-6638	222	1	the	the	DET
ejpam-6638	222	2	task	task	NOUN
ejpam-6638	222	3	is	be	AUX
ejpam-6638	222	4	to	to	PART
ejpam-6638	222	5	recover	recover	VERB
ejpam-6638	222	6	the	the	DET
ejpam-6638	222	7	missing	miss	VERB
ejpam-6638	222	8	entries	entry	NOUN
ejpam-6638	222	9	of	of	ADP
ejpam-6638	222	10	x0	x0	PROPN
ejpam-6638	222	11	.	.	PUNCT
ejpam-6638	223	1	a	a	DET
ejpam-6638	223	2	natural	natural	ADJ
ejpam-6638	223	3	data‐fidelity	data‐fidelity	NOUN
ejpam-6638	223	4	term	term	NOUN
ejpam-6638	223	5	for	for	ADP
ejpam-6638	223	6	this	this	DET
ejpam-6638	223	7	problem	problem	NOUN
ejpam-6638	223	8	is	be	AUX
ejpam-6638	223	9	e	e	NOUN
ejpam-6638	223	10	(	(	PUNCT
ejpam-6638	223	11	x;x0	x;x0	PROPN
ejpam-6638	223	12	)	)	PUNCT
ejpam-6638	223	13	=	=	PUNCT
ejpam-6638	223	14	∥∥	∥∥	X
ejpam-6638	223	15	(	(	PUNCT
ejpam-6638	223	16	x−	x−	PROPN
ejpam-6638	223	17	x0)⊙m	x0)⊙m	PROPN
ejpam-6638	223	18	∥∥2	∥∥2	PROPN
ejpam-6638	223	19	2	2	NUM
ejpam-6638	223	20	,	,	PUNCT
ejpam-6638	223	21	(	(	PUNCT
ejpam-6638	223	22	4.1	4.1	NUM
ejpam-6638	223	23	)	)	PUNCT
ejpam-6638	223	24	where	where	SCONJ
ejpam-6638	223	25	⊙	⊙	PROPN
ejpam-6638	223	26	denotes	denote	VERB
ejpam-6638	223	27	the	the	DET
ejpam-6638	223	28	hadamard	hadamard	ADJ
ejpam-6638	223	29	product	product	NOUN
ejpam-6638	223	30	.	.	PUNCT
ejpam-6638	224	1	since	since	SCONJ
ejpam-6638	224	2	e(x;x0	e(x;x0	PROPN
ejpam-6638	224	3	)	)	PUNCT
ejpam-6638	224	4	imposes	impose	VERB
ejpam-6638	224	5	no	no	DET
ejpam-6638	224	6	penalty	penalty	NOUN
ejpam-6638	224	7	on	on	ADP
ejpam-6638	224	8	the	the	DET
ejpam-6638	224	9	unobserved	unobserved	ADJ
ejpam-6638	224	10	pixels	pixel	NOUN
ejpam-6638	224	11	(	(	PUNCT
ejpam-6638	224	12	where	where	SCONJ
ejpam-6638	224	13	mij	mij	X
ejpam-6638	224	14	=	=	NOUN
ejpam-6638	224	15	0	0	NUM
ejpam-6638	224	16	)	)	PUNCT
ejpam-6638	224	17	,	,	PUNCT
ejpam-6638	224	18	direct	direct	ADJ
ejpam-6638	224	19	minimization	minimization	NOUN
ejpam-6638	224	20	over	over	ADP
ejpam-6638	224	21	the	the	DET
ejpam-6638	224	22	pixel	pixel	PROPN
ejpam-6638	224	23	values	value	NOUN
ejpam-6638	224	24	x	x	PRON
ejpam-6638	224	25	would	would	AUX
ejpam-6638	224	26	leave	leave	VERB
ejpam-6638	224	27	those	those	DET
ejpam-6638	224	28	entries	entry	NOUN
ejpam-6638	224	29	unchanged	unchanged	ADJ
ejpam-6638	224	30	from	from	ADP
ejpam-6638	224	31	their	their	PRON
ejpam-6638	224	32	initialization	initialization	NOUN
ejpam-6638	224	33	.	.	PUNCT
ejpam-6638	225	1	to	to	PART
ejpam-6638	225	2	remedy	remedy	VERB
ejpam-6638	225	3	this	this	PRON
ejpam-6638	225	4	,	,	PUNCT
ejpam-6638	225	5	we	we	PRON
ejpam-6638	225	6	introduce	introduce	VERB
ejpam-6638	225	7	a	a	DET
ejpam-6638	225	8	learned	learn	VERB
ejpam-6638	225	9	prior	prior	ADV
ejpam-6638	225	10	by	by	ADP
ejpam-6638	225	11	reparametrizing	reparametrize	VERB
ejpam-6638	225	12	x	x	PUNCT
ejpam-6638	225	13	as	as	ADP
ejpam-6638	225	14	in	in	ADP
ejpam-6638	225	15	(	(	PUNCT
ejpam-6638	225	16	2.3	2.3	NUM
ejpam-6638	225	17	)	)	PUNCT
ejpam-6638	225	18	and	and	CCONJ
ejpam-6638	225	19	perform	perform	VERB
ejpam-6638	225	20	optimization	optimization	NOUN
ejpam-6638	225	21	in	in	ADP
ejpam-6638	225	22	the	the	DET
ejpam-6638	225	23	network	network	NOUN
ejpam-6638	225	24	parameter	parameter	NOUN
ejpam-6638	225	25	space	space	NOUN
ejpam-6638	225	26	.	.	PUNCT
ejpam-6638	226	1	concretely	concretely	ADV
ejpam-6638	226	2	,	,	PUNCT
ejpam-6638	226	3	if	if	SCONJ
ejpam-6638	226	4	x	x	X
ejpam-6638	226	5	=	=	PUNCT
ejpam-6638	226	6	fθ(z	fθ(z	X
ejpam-6638	226	7	)	)	PUNCT
ejpam-6638	226	8	denotes	denote	VERB
ejpam-6638	226	9	the	the	DET
ejpam-6638	226	10	deep‐image‐prior	deep‐image‐prior	ADJ
ejpam-6638	226	11	mapping	mapping	NOUN
ejpam-6638	226	12	parameterized	parameterize	VERB
ejpam-6638	226	13	by	by	ADP
ejpam-6638	226	14	θ	θ	PROPN
ejpam-6638	226	15	and	and	CCONJ
ejpam-6638	226	16	driven	drive	VERB
ejpam-6638	226	17	by	by	ADP
ejpam-6638	226	18	a	a	DET
ejpam-6638	226	19	fixed	fix	VERB
ejpam-6638	226	20	input	input	NOUN
ejpam-6638	227	1	z	z	PROPN
ejpam-6638	227	2	,	,	PUNCT
ejpam-6638	227	3	then	then	ADV
ejpam-6638	227	4	we	we	PRON
ejpam-6638	227	5	minimize	minimize	VERB
ejpam-6638	227	6	min	min	PROPN
ejpam-6638	227	7	θ	θ	PROPN
ejpam-6638	227	8	e	e	X
ejpam-6638	227	9	(	(	PUNCT
ejpam-6638	227	10	fθ(z);x0	fθ(z);x0	NUM
ejpam-6638	227	11	)	)	PUNCT
ejpam-6638	227	12	(	(	PUNCT
ejpam-6638	227	13	4.2	4.2	NUM
ejpam-6638	227	14	)	)	PUNCT
ejpam-6638	227	15	so	so	SCONJ
ejpam-6638	227	16	that	that	SCONJ
ejpam-6638	227	17	the	the	DET
ejpam-6638	227	18	network	network	NOUN
ejpam-6638	227	19	naturally	naturally	ADV
ejpam-6638	227	20	infills	infill	VERB
ejpam-6638	227	21	the	the	DET
ejpam-6638	227	22	missing	miss	VERB
ejpam-6638	227	23	regions	region	NOUN
ejpam-6638	227	24	while	while	SCONJ
ejpam-6638	227	25	respecting	respect	VERB
ejpam-6638	227	26	the	the	DET
ejpam-6638	227	27	observed	observed	ADJ
ejpam-6638	227	28	data	datum	NOUN
ejpam-6638	227	29	.	.	PUNCT
ejpam-6638	228	1	image	image	NOUN
ejpam-6638	228	2	psnr	psnr	NOUN
ejpam-6638	228	3	(	(	PUNCT
ejpam-6638	228	4	db	db	NOUN
ejpam-6638	228	5	)	)	PUNCT
ejpam-6638	228	6	ssim	ssim	NOUN
ejpam-6638	228	7	butter	butter	NOUN
ejpam-6638	228	8	32.07	32.07	NUM
ejpam-6638	228	9	0.97	0.97	NUM
ejpam-6638	228	10	jet	jet	NOUN
ejpam-6638	228	11	35.88	35.88	NUM
ejpam-6638	228	12	0.96	0.96	NUM
ejpam-6638	228	13	library	library	NOUN
ejpam-6638	228	14	30.55	30.55	NUM
ejpam-6638	228	15	0.94	0.94	NUM
ejpam-6638	228	16	vase	vase	NOUN
ejpam-6638	228	17	34.47	34.47	NUM
ejpam-6638	228	18	0.99	0.99	NUM
ejpam-6638	228	19	table	table	NOUN
ejpam-6638	228	20	2	2	NUM
ejpam-6638	228	21	:	:	PUNCT
ejpam-6638	228	22	inpainting	inpainte	VERB
ejpam-6638	228	23	results	result	NOUN
ejpam-6638	228	24	using	use	VERB
ejpam-6638	228	25	dip	dip	NOUN
ejpam-6638	228	26	-	-	PUNCT
ejpam-6638	228	27	svd	svd	PROPN
ejpam-6638	228	28	over	over	ADP
ejpam-6638	228	29	3000	3000	NUM
ejpam-6638	228	30	epochs	epoch	NOUN
ejpam-6638	228	31	,	,	PUNCT
ejpam-6638	228	32	evaluated	evaluate	VERB
ejpam-6638	228	33	via	via	ADP
ejpam-6638	228	34	psnr	psnr	NOUN
ejpam-6638	228	35	and	and	CCONJ
ejpam-6638	228	36	ssim	ssim	NOUN
ejpam-6638	228	37	for	for	ADP
ejpam-6638	228	38	four	four	NUM
ejpam-6638	228	39	benchmark	benchmark	NOUN
ejpam-6638	228	40	images	image	NOUN
ejpam-6638	228	41	.	.	PUNCT
ejpam-6638	229	1	inpainting	inpainte	VERB
ejpam-6638	229	2	performance	performance	NOUN
ejpam-6638	229	3	analysis	analysis	NOUN
ejpam-6638	229	4	the	the	DET
ejpam-6638	229	5	quantitative	quantitative	ADJ
ejpam-6638	229	6	outcomes	outcome	NOUN
ejpam-6638	229	7	of	of	ADP
ejpam-6638	229	8	inpainting	inpainte	VERB
ejpam-6638	229	9	using	use	VERB
ejpam-6638	229	10	dipsvd	dipsvd	NOUN
ejpam-6638	229	11	after	after	ADP
ejpam-6638	229	12	training	train	VERB
ejpam-6638	229	13	for	for	ADP
ejpam-6638	229	14	3,000	3,000	NUM
ejpam-6638	229	15	epochs	epoch	NOUN
ejpam-6638	229	16	on	on	ADP
ejpam-6638	229	17	four	four	NUM
ejpam-6638	229	18	benchmark	benchmark	NOUN
ejpam-6638	229	19	images	image	NOUN
ejpam-6638	229	20	are	be	AUX
ejpam-6638	229	21	presented	present	VERB
ejpam-6638	229	22	in	in	ADP
ejpam-6638	229	23	table	table	NOUN
ejpam-6638	229	24	2	2	NUM
ejpam-6638	229	25	.	.	PUNCT
ejpam-6638	230	1	the	the	DET
ejpam-6638	230	2	peak	peak	NOUN
ejpam-6638	230	3	signal	signal	NOUN
ejpam-6638	230	4	-	-	PUNCT
ejpam-6638	230	5	to	to	ADP
ejpam-6638	230	6	-	-	PUNCT
ejpam-6638	230	7	noise	noise	NOUN
ejpam-6638	230	8	ratio	ratio	NOUN
ejpam-6638	230	9	(	(	PUNCT
ejpam-6638	230	10	psnr	psnr	NOUN
ejpam-6638	230	11	)	)	PUNCT
ejpam-6638	230	12	varies	vary	VERB
ejpam-6638	230	13	from	from	ADP
ejpam-6638	230	14	30.55	30.55	NUM
ejpam-6638	230	15	db	db	PROPN
ejpam-6638	230	16	(	(	PUNCT
ejpam-6638	230	17	library	library	NOUN
ejpam-6638	230	18	)	)	PUNCT
ejpam-6638	230	19	to	to	ADP
ejpam-6638	230	20	35.88	35.88	NUM
ejpam-6638	230	21	db	db	PROPN
ejpam-6638	230	22	(	(	PUNCT
ejpam-6638	230	23	jet	jet	NOUN
ejpam-6638	230	24	)	)	PUNCT
ejpam-6638	230	25	,	,	PUNCT
ejpam-6638	230	26	indicating	indicate	VERB
ejpam-6638	230	27	notable	notable	ADJ
ejpam-6638	230	28	noise	noise	NOUN
ejpam-6638	230	29	reduction	reduction	NOUN
ejpam-6638	230	30	and	and	CCONJ
ejpam-6638	230	31	precision	precision	NOUN
ejpam-6638	230	32	at	at	ADP
ejpam-6638	230	33	the	the	DET
ejpam-6638	230	34	pixel	pixel	PROPN
ejpam-6638	230	35	level	level	NOUN
ejpam-6638	230	36	across	across	ADP
ejpam-6638	230	37	different	different	ADJ
ejpam-6638	230	38	types	type	NOUN
ejpam-6638	230	39	of	of	ADP
ejpam-6638	230	40	content	content	NOUN
ejpam-6638	230	41	.	.	PUNCT
ejpam-6638	231	1	the	the	DET
ejpam-6638	231	2	library	library	ADJ
ejpam-6638	231	3	scene	scene	NOUN
ejpam-6638	231	4	is	be	AUX
ejpam-6638	231	5	the	the	DET
ejpam-6638	231	6	hardest	hard	ADJ
ejpam-6638	231	7	to	to	PART
ejpam-6638	231	8	record	record	VERB
ejpam-6638	231	9	at	at	ADP
ejpam-6638	231	10	30.55	30.55	NUM
ejpam-6638	231	11	db	db	NOUN
ejpam-6638	231	12	,	,	PUNCT
ejpam-6638	231	13	while	while	SCONJ
ejpam-6638	231	14	the	the	DET
ejpam-6638	231	15	jet	jet	NOUN
ejpam-6638	231	16	image	image	NOUN
ejpam-6638	231	17	has	have	VERB
ejpam-6638	231	18	the	the	DET
ejpam-6638	231	19	greatest	great	ADJ
ejpam-6638	231	20	psnr	psnr	NOUN
ejpam-6638	231	21	at	at	ADP
ejpam-6638	231	22	35.88	35.88	NUM
ejpam-6638	231	23	db	db	NOUN
ejpam-6638	231	24	.	.	PUNCT
ejpam-6638	232	1	the	the	DET
ejpam-6638	232	2	structural	structural	ADJ
ejpam-6638	232	3	similarity	similarity	NOUN
ejpam-6638	232	4	index	index	NOUN
ejpam-6638	232	5	(	(	PUNCT
ejpam-6638	232	6	ssim	ssim	NOUN
ejpam-6638	232	7	)	)	PUNCT
ejpam-6638	232	8	values	value	NOUN
ejpam-6638	232	9	consistently	consistently	ADV
ejpam-6638	232	10	remain	remain	VERB
ejpam-6638	232	11	high	high	ADJ
ejpam-6638	232	12	,	,	PUNCT
ejpam-6638	232	13	ranging	range	VERB
ejpam-6638	232	14	from	from	ADP
ejpam-6638	232	15	0.94	0.94	NUM
ejpam-6638	232	16	(	(	PUNCT
ejpam-6638	232	17	library	library	NOUN
ejpam-6638	232	18	)	)	PUNCT
ejpam-6638	232	19	to	to	ADP
ejpam-6638	232	20	0.99	0.99	NUM
ejpam-6638	232	21	(	(	PUNCT
ejpam-6638	232	22	vase	vase	NOUN
ejpam-6638	232	23	)	)	PUNCT
ejpam-6638	232	24	,	,	PUNCT
ejpam-6638	232	25	suggesting	suggest	VERB
ejpam-6638	232	26	that	that	SCONJ
ejpam-6638	232	27	dip	dip	NOUN
ejpam-6638	232	28	-	-	PUNCT
ejpam-6638	232	29	svd	svd	NOUN
ejpam-6638	232	30	effectively	effectively	ADV
ejpam-6638	232	31	maintains	maintain	VERB
ejpam-6638	232	32	the	the	DET
ejpam-6638	232	33	perceptual	perceptual	ADJ
ejpam-6638	232	34	structure	structure	NOUN
ejpam-6638	232	35	of	of	ADP
ejpam-6638	232	36	the	the	DET
ejpam-6638	232	37	images	image	NOUN
ejpam-6638	232	38	.	.	PUNCT
ejpam-6638	233	1	the	the	DET
ejpam-6638	233	2	vase	vase	NOUN
ejpam-6638	233	3	image	image	NOUN
ejpam-6638	233	4	shows	show	VERB
ejpam-6638	233	5	almost	almost	ADV
ejpam-6638	233	6	perfect	perfect	ADJ
ejpam-6638	233	7	structural	structural	ADJ
ejpam-6638	233	8	recovery	recovery	NOUN
ejpam-6638	233	9	with	with	ADP
ejpam-6638	233	10	a	a	DET
ejpam-6638	233	11	peak	peak	NOUN
ejpam-6638	233	12	ssim	ssim	NOUN
ejpam-6638	233	13	of	of	ADP
ejpam-6638	233	14	0.99	0.99	NUM
ejpam-6638	233	15	.	.	PUNCT
ejpam-6638	234	1	all	all	PRON
ejpam-6638	234	2	of	of	ADP
ejpam-6638	234	3	these	these	DET
ejpam-6638	234	4	metrics	metric	NOUN
ejpam-6638	234	5	together	together	ADV
ejpam-6638	234	6	confirm	confirm	VERB
ejpam-6638	234	7	that	that	SCONJ
ejpam-6638	234	8	the	the	DET
ejpam-6638	234	9	svd	svd	PROPN
ejpam-6638	234	10	-	-	PUNCT
ejpam-6638	234	11	enhanced	enhance	VERB
ejpam-6638	234	12	deep	deep	ADJ
ejpam-6638	234	13	image	image	NOUN
ejpam-6638	234	14	prior	prior	ADV
ejpam-6638	234	15	is	be	AUX
ejpam-6638	234	16	successful	successful	ADJ
ejpam-6638	234	17	in	in	ADP
ejpam-6638	234	18	bringing	bring	VERB
ejpam-6638	234	19	back	back	ADP
ejpam-6638	234	20	the	the	DET
ejpam-6638	234	21	fine	fine	ADJ
ejpam-6638	234	22	details	detail	NOUN
ejpam-6638	234	23	and	and	CCONJ
ejpam-6638	234	24	overall	overall	ADJ
ejpam-6638	234	25	beauty	beauty	NOUN
ejpam-6638	234	26	in	in	ADP
ejpam-6638	234	27	the	the	DET
ejpam-6638	234	28	areas	area	NOUN
ejpam-6638	234	29	that	that	PRON
ejpam-6638	234	30	are	be	AUX
ejpam-6638	234	31	absent	absent	ADJ
ejpam-6638	234	32	.	.	PUNCT
ejpam-6638	235	1	superiority	superiority	NOUN
ejpam-6638	235	2	of	of	ADP
ejpam-6638	235	3	dip	dip	NOUN
ejpam-6638	235	4	-	-	PUNCT
ejpam-6638	235	5	svd	svd	PROPN
ejpam-6638	235	6	for	for	ADP
ejpam-6638	235	7	inpainting	inpainte	VERB
ejpam-6638	235	8	according	accord	VERB
ejpam-6638	235	9	to	to	ADP
ejpam-6638	235	10	the	the	DET
ejpam-6638	235	11	results	result	NOUN
ejpam-6638	235	12	,	,	PUNCT
ejpam-6638	235	13	the	the	DET
ejpam-6638	235	14	proposed	propose	VERB
ejpam-6638	235	15	dip	dip	NOUN
ejpam-6638	235	16	-	-	PUNCT
ejpam-6638	235	17	svd	svd	PROPN
ejpam-6638	235	18	architecture	architecture	NOUN
ejpam-6638	235	19	outperforms	outperform	VERB
ejpam-6638	235	20	conventional	conventional	ADJ
ejpam-6638	235	21	deep	deep	ADJ
ejpam-6638	235	22	-	-	PUNCT
ejpam-6638	235	23	image	image	NOUN
ejpam-6638	235	24	-	-	PUNCT
ejpam-6638	235	25	prior	prior	ADJ
ejpam-6638	235	26	methods	method	NOUN
ejpam-6638	235	27	in	in	ADP
ejpam-6638	235	28	image	image	NOUN
ejpam-6638	235	29	inpainting	inpainting	NOUN
ejpam-6638	235	30	.	.	PUNCT
ejpam-6638	236	1	the	the	DET
ejpam-6638	236	2	incorporation	incorporation	NOUN
ejpam-6638	236	3	of	of	ADP
ejpam-6638	236	4	svd	svd	PROPN
ejpam-6638	236	5	-	-	PUNCT
ejpam-6638	236	6	based	base	VERB
ejpam-6638	236	7	pooling	pooling	NOUN
ejpam-6638	236	8	into	into	ADP
ejpam-6638	236	9	the	the	DET
ejpam-6638	236	10	design	design	NOUN
ejpam-6638	236	11	of	of	ADP
ejpam-6638	236	12	dip	dip	NOUN
ejpam-6638	236	13	-	-	PUNCT
ejpam-6638	236	14	svd	svd	PROPN
ejpam-6638	236	15	improves	improve	VERB
ejpam-6638	236	16	the	the	DET
ejpam-6638	236	17	transfer	transfer	NOUN
ejpam-6638	236	18	of	of	ADP
ejpam-6638	236	19	contextual	contextual	ADJ
ejpam-6638	236	20	information	information	NOUN
ejpam-6638	236	21	from	from	ADP
ejpam-6638	236	22	the	the	DET
ejpam-6638	236	23	known	know	VERB
ejpam-6638	236	24	regions	region	NOUN
ejpam-6638	236	25	to	to	ADP
ejpam-6638	236	26	the	the	DET
ejpam-6638	236	27	missing	miss	VERB
ejpam-6638	236	28	areas	area	NOUN
ejpam-6638	236	29	,	,	PUNCT
ejpam-6638	236	30	leading	lead	VERB
ejpam-6638	236	31	to	to	ADP
ejpam-6638	236	32	more	more	ADV
ejpam-6638	236	33	accurate	accurate	ADJ
ejpam-6638	236	34	and	and	CCONJ
ejpam-6638	236	35	less	less	ADV
ejpam-6638	236	36	artifactable	artifactable	ADJ
ejpam-6638	236	37	reconstructions	reconstruction	NOUN
ejpam-6638	236	38	.	.	PUNCT
ejpam-6638	237	1	dip	dip	NOUN
ejpam-6638	237	2	-	-	PUNCT
ejpam-6638	237	3	svd	svd	PROPN
ejpam-6638	237	4	consistently	consistently	ADV
ejpam-6638	237	5	achieves	achieve	VERB
ejpam-6638	237	6	better	well	ADJ
ejpam-6638	237	7	quantitative	quantitative	ADJ
ejpam-6638	237	8	psnr	psnr	NOUN
ejpam-6638	237	9	and	and	CCONJ
ejpam-6638	237	10	ssim	ssim	NOUN
ejpam-6638	237	11	results	result	NOUN
ejpam-6638	237	12	across	across	ADP
ejpam-6638	237	13	a	a	DET
ejpam-6638	237	14	variety	variety	NOUN
ejpam-6638	237	15	of	of	ADP
ejpam-6638	237	16	test	test	NOUN
ejpam-6638	237	17	images	image	NOUN
ejpam-6638	237	18	(	(	PUNCT
ejpam-6638	237	19	refer	refer	VERB
ejpam-6638	237	20	to	to	ADP
ejpam-6638	237	21	table	table	NOUN
ejpam-6638	237	22	2	2	NUM
ejpam-6638	237	23	)	)	PUNCT
ejpam-6638	237	24	,	,	PUNCT
ejpam-6638	237	25	while	while	SCONJ
ejpam-6638	237	26	also	also	ADV
ejpam-6638	237	27	producing	produce	VERB
ejpam-6638	237	28	smoother	smooth	ADJ
ejpam-6638	237	29	edges	edge	NOUN
ejpam-6638	237	30	,	,	PUNCT
ejpam-6638	237	31	more	more	ADV
ejpam-6638	237	32	cohesive	cohesive	ADJ
ejpam-6638	237	33	textures	texture	NOUN
ejpam-6638	237	34	,	,	PUNCT
ejpam-6638	237	35	and	and	CCONJ
ejpam-6638	237	36	less	less	ADV
ejpam-6638	237	37	visible	visible	ADJ
ejpam-6638	237	38	seams	seam	NOUN
ejpam-6638	237	39	in	in	ADP
ejpam-6638	237	40	the	the	DET
ejpam-6638	237	41	inpainted	inpainte	VERB
ejpam-6638	237	42	areas	area	NOUN
ejpam-6638	237	43	(	(	PUNCT
ejpam-6638	237	44	see	see	VERB
ejpam-6638	237	45	fig	fig	NOUN
ejpam-6638	237	46	.	.	PUNCT
ejpam-6638	238	1	7	7	NUM
ejpam-6638	238	2	)	)	PUNCT
ejpam-6638	238	3	.	.	PUNCT
ejpam-6638	239	1	these	these	DET
ejpam-6638	239	2	enhancements	enhancement	NOUN
ejpam-6638	239	3	suggest	suggest	VERB
ejpam-6638	239	4	that	that	SCONJ
ejpam-6638	239	5	the	the	DET
ejpam-6638	239	6	svd	svd	PROPN
ejpam-6638	239	7	-	-	PUNCT
ejpam-6638	239	8	augmented	augment	VERB
ejpam-6638	239	9	prior	prior	ADV
ejpam-6638	239	10	is	be	AUX
ejpam-6638	239	11	particularly	particularly	ADV
ejpam-6638	239	12	effective	effective	ADJ
ejpam-6638	239	13	for	for	ADP
ejpam-6638	239	14	addressing	address	VERB
ejpam-6638	239	15	large	large	ADJ
ejpam-6638	239	16	or	or	CCONJ
ejpam-6638	239	17	complex	complex	ADJ
ejpam-6638	239	18	missing	missing	ADJ
ejpam-6638	239	19	regions	region	NOUN
ejpam-6638	239	20	,	,	PUNCT
ejpam-6638	239	21	making	make	VERB
ejpam-6638	239	22	dip	dip	NOUN
ejpam-6638	239	23	-	-	PUNCT
ejpam-6638	239	24	svd	svd	PROPN
ejpam-6638	239	25	a	a	DET
ejpam-6638	239	26	preferable	preferable	ADJ
ejpam-6638	239	27	option	option	NOUN
ejpam-6638	239	28	for	for	ADP
ejpam-6638	239	29	image	image	NOUN
ejpam-6638	239	30	inpainting	inpainte	VERB
ejpam-6638	239	31	tasks	task	NOUN
ejpam-6638	239	32	.	.	PUNCT
ejpam-6638	240	1	•	•	NOUN
ejpam-6638	240	2	denoising	denoise	VERB
ejpam-6638	240	3	using	use	VERB
ejpam-6638	240	4	dip	dip	NOUN
ejpam-6638	240	5	-	-	PUNCT
ejpam-6638	240	6	svd	svd	PROPN
ejpam-6638	240	7	:	:	PUNCT
ejpam-6638	240	8	in	in	ADP
ejpam-6638	240	9	dip	dip	NOUN
ejpam-6638	240	10	-	-	PUNCT
ejpam-6638	240	11	svd	svd	PROPN
ejpam-6638	240	12	,	,	PUNCT
ejpam-6638	240	13	the	the	DET
ejpam-6638	240	14	weights	weight	NOUN
ejpam-6638	240	15	taken	take	VERB
ejpam-6638	240	16	from	from	ADP
ejpam-6638	240	17	the	the	DET
ejpam-6638	240	18	convolutional	convolutional	ADJ
ejpam-6638	240	19	layers	layer	NOUN
ejpam-6638	240	20	(	(	PUNCT
ejpam-6638	240	21	fig	fig	NOUN
ejpam-6638	240	22	.	.	PUNCT
ejpam-6638	241	1	5	5	NUM
ejpam-6638	241	2	,	,	PUNCT
ejpam-6638	241	3	top	top	ADJ
ejpam-6638	241	4	row	row	NOUN
ejpam-6638	241	5	)	)	PUNCT
ejpam-6638	241	6	show	show	VERB
ejpam-6638	241	7	three	three	NUM
ejpam-6638	241	8	important	important	ADJ
ejpam-6638	241	9	features	feature	NOUN
ejpam-6638	241	10	:	:	PUNCT
ejpam-6638	241	11	(	(	PUNCT
ejpam-6638	241	12	i	i	NOUN
ejpam-6638	241	13	)	)	PUNCT
ejpam-6638	241	14	near	near	ADP
ejpam-6638	241	15	-	-	PUNCT
ejpam-6638	241	16	zero	zero	NUM
ejpam-6638	241	17	mean	mean	NOUN
ejpam-6638	241	18	values	value	NOUN
ejpam-6638	241	19	(	(	PUNCT
ejpam-6638	241	20	µw	µw	PROPN
ejpam-6638	241	21	≈	≈	PROPN
ejpam-6638	241	22	0	0	NUM
ejpam-6638	241	23	)	)	PUNCT
ejpam-6638	241	24	showing	show	VERB
ejpam-6638	241	25	balanced	balanced	ADJ
ejpam-6638	241	26	noise	noise	NOUN
ejpam-6638	241	27	suppression	suppression	NOUN
ejpam-6638	241	28	m.	m.	NOUN
ejpam-6638	241	29	israr	israr	PROPN
ejpam-6638	241	30	et	et	PROPN
ejpam-6638	241	31	al	al	PROPN
ejpam-6638	241	32	.	.	PUNCT
ejpam-6638	241	33	/	/	SYM
ejpam-6638	241	34	eur	eur	PROPN
ejpam-6638	241	35	.	.	PUNCT
ejpam-6638	242	1	j.	j.	PROPN
ejpam-6638	242	2	pure	pure	PROPN
ejpam-6638	242	3	appl	appl	PROPN
ejpam-6638	242	4	.	.	PROPN
ejpam-6638	242	5	math	math	PROPN
ejpam-6638	242	6	,	,	PUNCT
ejpam-6638	242	7	18	18	NUM
ejpam-6638	242	8	(	(	PUNCT
ejpam-6638	242	9	4	4	NUM
ejpam-6638	242	10	)	)	PUNCT
ejpam-6638	242	11	(	(	PUNCT
ejpam-6638	242	12	2025	2025	NUM
ejpam-6638	242	13	)	)	PUNCT
ejpam-6638	242	14	,	,	PUNCT
ejpam-6638	242	15	6638	6638	NUM
ejpam-6638	242	16	12	12	NUM
ejpam-6638	242	17	of	of	ADP
ejpam-6638	242	18	22	22	NUM
ejpam-6638	242	19	(	(	PUNCT
ejpam-6638	242	20	a	a	NOUN
ejpam-6638	242	21	)	)	PUNCT
ejpam-6638	242	22	(	(	PUNCT
ejpam-6638	242	23	b	b	X
ejpam-6638	242	24	)	)	PUNCT
ejpam-6638	242	25	(	(	PUNCT
ejpam-6638	242	26	c	c	X
ejpam-6638	242	27	)	)	PUNCT
ejpam-6638	242	28	(	(	PUNCT
ejpam-6638	242	29	d	d	X
ejpam-6638	242	30	)	)	PUNCT
ejpam-6638	242	31	(	(	PUNCT
ejpam-6638	242	32	e	e	NOUN
ejpam-6638	242	33	)	)	PUNCT
ejpam-6638	242	34	(	(	PUNCT
ejpam-6638	242	35	f	f	X
ejpam-6638	242	36	)	)	PUNCT
ejpam-6638	242	37	(	(	PUNCT
ejpam-6638	242	38	g	g	NOUN
ejpam-6638	242	39	)	)	PUNCT
ejpam-6638	242	40	(	(	PUNCT
ejpam-6638	242	41	h	h	NOUN
ejpam-6638	242	42	)	)	PUNCT
ejpam-6638	242	43	figure	figure	NOUN
ejpam-6638	242	44	6	6	NUM
ejpam-6638	242	45	:	:	PUNCT
ejpam-6638	242	46	visual	visual	ADJ
ejpam-6638	242	47	comparison	comparison	NOUN
ejpam-6638	242	48	on	on	ADP
ejpam-6638	242	49	eight	eight	NUM
ejpam-6638	242	50	test	test	NOUN
ejpam-6638	242	51	images	image	NOUN
ejpam-6638	242	52	:	:	PUNCT
ejpam-6638	242	53	the	the	DET
ejpam-6638	242	54	first	first	ADJ
ejpam-6638	242	55	row	row	NOUN
ejpam-6638	242	56	shows	show	VERB
ejpam-6638	242	57	the	the	DET
ejpam-6638	242	58	masked	mask	VERB
ejpam-6638	242	59	inputs	input	NOUN
ejpam-6638	242	60	to	to	ADP
ejpam-6638	242	61	dip	dip	NOUN
ejpam-6638	242	62	-	-	PUNCT
ejpam-6638	242	63	svd	svd	X
ejpam-6638	242	64	(	(	PUNCT
ejpam-6638	242	65	each	each	DET
ejpam-6638	242	66	image	image	NOUN
ejpam-6638	242	67	overlaid	overlay	VERB
ejpam-6638	242	68	with	with	ADP
ejpam-6638	242	69	its	its	PRON
ejpam-6638	242	70	corresponding	corresponding	ADJ
ejpam-6638	242	71	mask	mask	NOUN
ejpam-6638	242	72	)	)	PUNCT
ejpam-6638	242	73	,	,	PUNCT
ejpam-6638	242	74	and	and	CCONJ
ejpam-6638	242	75	the	the	DET
ejpam-6638	242	76	second	second	ADJ
ejpam-6638	242	77	row	row	NOUN
ejpam-6638	242	78	presents	present	VERB
ejpam-6638	242	79	the	the	DET
ejpam-6638	242	80	restored	restore	VERB
ejpam-6638	242	81	,	,	PUNCT
ejpam-6638	242	82	clear	clear	ADJ
ejpam-6638	242	83	outputs	output	NOUN
ejpam-6638	242	84	produced	produce	VERB
ejpam-6638	242	85	by	by	ADP
ejpam-6638	242	86	the	the	DET
ejpam-6638	242	87	proposed	propose	VERB
ejpam-6638	242	88	method	method	NOUN
ejpam-6638	242	89	.	.	PUNCT
ejpam-6638	243	1	(	(	PUNCT
ejpam-6638	243	2	ii	ii	NOUN
ejpam-6638	243	3	)	)	PUNCT
ejpam-6638	243	4	low	low	ADJ
ejpam-6638	243	5	standard	standard	ADJ
ejpam-6638	243	6	deviation	deviation	NOUN
ejpam-6638	243	7	(	(	PUNCT
ejpam-6638	243	8	σw	σw	X
ejpam-6638	243	9	<	<	X
ejpam-6638	243	10	0.1	0.1	NUM
ejpam-6638	243	11	)	)	PUNCT
ejpam-6638	243	12	for	for	ADP
ejpam-6638	243	13	stable	stable	ADJ
ejpam-6638	243	14	feature	feature	NOUN
ejpam-6638	243	15	preservation	preservation	NOUN
ejpam-6638	243	16	(	(	PUNCT
ejpam-6638	243	17	iii	iii	NOUN
ejpam-6638	243	18	)	)	PUNCT
ejpam-6638	243	19	edge	edge	NOUN
ejpam-6638	243	20	-	-	PUNCT
ejpam-6638	243	21	selective	selective	ADJ
ejpam-6638	243	22	filters	filter	NOUN
ejpam-6638	243	23	formed	form	VERB
ejpam-6638	243	24	by	by	ADP
ejpam-6638	243	25	negative	negative	ADJ
ejpam-6638	243	26	-	-	ADJ
ejpam-6638	243	27	positive	positive	ADJ
ejpam-6638	243	28	weight	weight	NOUN
ejpam-6638	243	29	pairs	pair	NOUN
ejpam-6638	243	30	these	these	DET
ejpam-6638	243	31	characteristics	characteristic	NOUN
ejpam-6638	243	32	arise	arise	VERB
ejpam-6638	243	33	because	because	SCONJ
ejpam-6638	243	34	the	the	DET
ejpam-6638	243	35	convolutional	convolutional	ADJ
ejpam-6638	243	36	weights	weight	NOUN
ejpam-6638	243	37	are	be	AUX
ejpam-6638	243	38	constrained	constrain	VERB
ejpam-6638	243	39	by	by	ADP
ejpam-6638	243	40	svd	svd	PROPN
ejpam-6638	243	41	pooling	pool	VERB
ejpam-6638	243	42	to	to	PART
ejpam-6638	243	43	function	function	VERB
ejpam-6638	243	44	as	as	ADP
ejpam-6638	243	45	adaptive	adaptive	ADJ
ejpam-6638	243	46	band	band	NOUN
ejpam-6638	243	47	-	-	PUNCT
ejpam-6638	243	48	pass	pass	NOUN
ejpam-6638	243	49	filters	filter	NOUN
ejpam-6638	243	50	.	.	PUNCT
ejpam-6638	244	1	specifically	specifically	ADV
ejpam-6638	244	2	,	,	PUNCT
ejpam-6638	244	3	the	the	DET
ejpam-6638	244	4	svd	svd	PROPN
ejpam-6638	244	5	decomposition	decomposition	NOUN
ejpam-6638	244	6	of	of	ADP
ejpam-6638	244	7	the	the	DET
ejpam-6638	244	8	weight	weight	NOUN
ejpam-6638	244	9	matrix	matrix	NOUN
ejpam-6638	244	10	is	be	AUX
ejpam-6638	244	11	:	:	PUNCT
ejpam-6638	244	12	wdenoise	wdenoise	NOUN
ejpam-6638	244	13	=	=	PUNCT
ejpam-6638	244	14	uσv	uσv	PRON
ejpam-6638	244	15	⊤	⊤	NOUN
ejpam-6638	244	16	,	,	PUNCT
ejpam-6638	244	17	(	(	PUNCT
ejpam-6638	244	18	4.3	4.3	NUM
ejpam-6638	244	19	)	)	PUNCT
ejpam-6638	244	20	where	where	SCONJ
ejpam-6638	244	21	:	:	PUNCT
ejpam-6638	244	22	–	–	PUNCT
ejpam-6638	244	23	wdenoise	wdenoise	NOUN
ejpam-6638	244	24	∈	∈	NOUN
ejpam-6638	244	25	rm×n	rm×n	NOUN
ejpam-6638	244	26	is	be	AUX
ejpam-6638	244	27	the	the	DET
ejpam-6638	244	28	weight	weight	NOUN
ejpam-6638	244	29	matrix	matrix	NOUN
ejpam-6638	244	30	of	of	ADP
ejpam-6638	244	31	a	a	DET
ejpam-6638	244	32	convolutional	convolutional	ADJ
ejpam-6638	244	33	filter	filter	NOUN
ejpam-6638	244	34	,	,	PUNCT
ejpam-6638	244	35	–	–	PUNCT
ejpam-6638	244	36	u	u	PROPN
ejpam-6638	244	37	∈	∈	PROPN
ejpam-6638	244	38	rm×m	rm×m	NOUN
ejpam-6638	244	39	contains	contain	VERB
ejpam-6638	244	40	the	the	DET
ejpam-6638	244	41	left	left	ADJ
ejpam-6638	244	42	singular	singular	ADJ
ejpam-6638	244	43	vectors	vector	NOUN
ejpam-6638	244	44	,	,	PUNCT
ejpam-6638	244	45	–	–	PUNCT
ejpam-6638	244	46	σ	σ	PROPN
ejpam-6638	244	47	∈	∈	PROPN
ejpam-6638	244	48	rm×n	rm×n	NOUN
ejpam-6638	244	49	is	be	AUX
ejpam-6638	244	50	a	a	DET
ejpam-6638	244	51	diagonal	diagonal	ADJ
ejpam-6638	244	52	matrix	matrix	NOUN
ejpam-6638	244	53	of	of	ADP
ejpam-6638	244	54	singular	singular	ADJ
ejpam-6638	244	55	values	value	NOUN
ejpam-6638	244	56	,	,	PUNCT
ejpam-6638	244	57	here	here	ADV
ejpam-6638	244	58	truncated	truncate	VERB
ejpam-6638	244	59	as	as	ADP
ejpam-6638	244	60	σ	σ	PROPN
ejpam-6638	244	61	=	=	PUNCT
ejpam-6638	244	62	diag(σ1	diag(σ1	NOUN
ejpam-6638	244	63	,	,	PUNCT
ejpam-6638	244	64	σ2	σ2	NOUN
ejpam-6638	244	65	,	,	PUNCT
ejpam-6638	244	66	0	0	NUM
ejpam-6638	244	67	,	,	PUNCT
ejpam-6638	244	68	.	.	PUNCT
ejpam-6638	244	69	.	.	PUNCT
ejpam-6638	244	70	.	.	PUNCT
ejpam-6638	245	1	,	,	PUNCT
ejpam-6638	245	2	0	0	NUM
ejpam-6638	245	3	)	)	PUNCT
ejpam-6638	245	4	,	,	PUNCT
ejpam-6638	245	5	–	–	PUNCT
ejpam-6638	245	6	v	v	X
ejpam-6638	245	7	∈	∈	NOUN
ejpam-6638	245	8	rn×n	rn×n	NOUN
ejpam-6638	245	9	contains	contain	VERB
ejpam-6638	245	10	the	the	DET
ejpam-6638	245	11	right	right	ADJ
ejpam-6638	245	12	singular	singular	ADJ
ejpam-6638	245	13	vectors	vector	NOUN
ejpam-6638	245	14	,	,	PUNCT
ejpam-6638	245	15	so	so	SCONJ
ejpam-6638	245	16	that	that	SCONJ
ejpam-6638	245	17	v	v	ADP
ejpam-6638	245	18	⊤	⊤	PROPN
ejpam-6638	245	19	∈	∈	PROPN
ejpam-6638	245	20	rn×n	rn×n	NOUN
ejpam-6638	245	21	.	.	PUNCT
ejpam-6638	246	1	the	the	DET
ejpam-6638	246	2	truncation	truncation	NOUN
ejpam-6638	246	3	effectively	effectively	ADV
ejpam-6638	246	4	keeps	keep	VERB
ejpam-6638	246	5	only	only	ADV
ejpam-6638	246	6	the	the	DET
ejpam-6638	246	7	first	first	ADJ
ejpam-6638	246	8	two	two	NUM
ejpam-6638	246	9	singular	singular	ADJ
ejpam-6638	246	10	components	component	NOUN
ejpam-6638	246	11	,	,	PUNCT
ejpam-6638	246	12	which	which	PRON
ejpam-6638	246	13	capture	capture	VERB
ejpam-6638	246	14	the	the	DET
ejpam-6638	246	15	dominant	dominant	ADJ
ejpam-6638	246	16	structures	structure	NOUN
ejpam-6638	246	17	while	while	SCONJ
ejpam-6638	246	18	suppressing	suppress	VERB
ejpam-6638	246	19	noise	noise	NOUN
ejpam-6638	246	20	.	.	PUNCT
ejpam-6638	247	1	the	the	DET
ejpam-6638	247	2	truncated	truncate	VERB
ejpam-6638	247	3	svd	svd	PROPN
ejpam-6638	247	4	automatically	automatically	ADV
ejpam-6638	247	5	eliminates	eliminate	VERB
ejpam-6638	247	6	noise	noise	NOUN
ejpam-6638	247	7	-	-	PUNCT
ejpam-6638	247	8	corrupted	corrupt	VERB
ejpam-6638	247	9	singular	singular	ADJ
ejpam-6638	247	10	values	value	NOUN
ejpam-6638	247	11	,	,	PUNCT
ejpam-6638	247	12	leaving	leave	VERB
ejpam-6638	247	13	only	only	ADV
ejpam-6638	247	14	σ1	σ1	PROPN
ejpam-6638	247	15	,	,	PUNCT
ejpam-6638	247	16	σ2	σ2	PROPN
ejpam-6638	247	17	.	.	PUNCT
ejpam-6638	247	18	•	•	NOUN
ejpam-6638	247	19	inpainting	inpainte	VERB
ejpam-6638	247	20	using	use	VERB
ejpam-6638	247	21	dip	dip	NOUN
ejpam-6638	247	22	-	-	PUNCT
ejpam-6638	247	23	svd	svd	PROPN
ejpam-6638	247	24	:	:	PUNCT
ejpam-6638	247	25	inpainting	inpainte	VERB
ejpam-6638	247	26	using	use	VERB
ejpam-6638	247	27	convolutional	convolutional	ADJ
ejpam-6638	247	28	weights	weight	NOUN
ejpam-6638	247	29	(	(	PUNCT
ejpam-6638	247	30	fig	fig	NOUN
ejpam-6638	247	31	.	.	PUNCT
ejpam-6638	248	1	5	5	NUM
ejpam-6638	248	2	,	,	PUNCT
ejpam-6638	248	3	bottom	bottom	ADJ
ejpam-6638	248	4	row	row	NOUN
ejpam-6638	248	5	)	)	PUNCT
ejpam-6638	249	1	displays	display	NOUN
ejpam-6638	249	2	:	:	PUNCT
ejpam-6638	249	3	(	(	PUNCT
ejpam-6638	249	4	i	i	NOUN
ejpam-6638	249	5	)	)	PUNCT
ejpam-6638	249	6	positive	positive	ADJ
ejpam-6638	249	7	mean	mean	NOUN
ejpam-6638	249	8	weights	weight	NOUN
ejpam-6638	249	9	(	(	PUNCT
ejpam-6638	249	10	µw	µw	X
ejpam-6638	249	11	>	>	X
ejpam-6638	249	12	0.2	0.2	NUM
ejpam-6638	249	13	)	)	PUNCT
ejpam-6638	249	14	for	for	ADP
ejpam-6638	249	15	content	content	NOUN
ejpam-6638	249	16	generation	generation	NOUN
ejpam-6638	249	17	(	(	PUNCT
ejpam-6638	249	18	ii	ii	NOUN
ejpam-6638	249	19	)	)	PUNCT
ejpam-6638	249	20	higher	high	ADJ
ejpam-6638	249	21	variance	variance	NOUN
ejpam-6638	249	22	(	(	PUNCT
ejpam-6638	249	23	σw	σw	X
ejpam-6638	249	24	∈	∈	PROPN
ejpam-6638	250	1	[	[	X
ejpam-6638	250	2	0.1	0.1	NUM
ejpam-6638	250	3	,	,	PUNCT
ejpam-6638	250	4	0.3	0.3	NUM
ejpam-6638	250	5	]	]	PUNCT
ejpam-6638	250	6	)	)	PUNCT
ejpam-6638	250	7	allowing	allow	VERB
ejpam-6638	250	8	feature	feature	NOUN
ejpam-6638	250	9	synthesis	synthesis	NOUN
ejpam-6638	250	10	(	(	PUNCT
ejpam-6638	250	11	iii	iii	NOUN
ejpam-6638	250	12	)	)	PUNCT
ejpam-6638	250	13	strong	strong	ADJ
ejpam-6638	250	14	bias	bias	NOUN
ejpam-6638	250	15	terms	term	NOUN
ejpam-6638	250	16	(	(	PUNCT
ejpam-6638	250	17	|b|	|b|	X
ejpam-6638	250	18	>	>	X
ejpam-6638	250	19	0.1	0.1	NUM
ejpam-6638	250	20	)	)	PUNCT
ejpam-6638	250	21	triggering	trigger	VERB
ejpam-6638	250	22	reconstruction	reconstruction	NOUN
ejpam-6638	250	23	m.	m.	NOUN
ejpam-6638	250	24	israr	israr	PROPN
ejpam-6638	250	25	et	et	PROPN
ejpam-6638	250	26	al	al	PROPN
ejpam-6638	250	27	.	.	PUNCT
ejpam-6638	250	28	/	/	SYM
ejpam-6638	250	29	eur	eur	PROPN
ejpam-6638	250	30	.	.	PUNCT
ejpam-6638	251	1	j.	j.	PROPN
ejpam-6638	251	2	pure	pure	PROPN
ejpam-6638	251	3	appl	appl	PROPN
ejpam-6638	251	4	.	.	PROPN
ejpam-6638	251	5	math	math	PROPN
ejpam-6638	251	6	,	,	PUNCT
ejpam-6638	251	7	18	18	NUM
ejpam-6638	251	8	(	(	PUNCT
ejpam-6638	251	9	4	4	NUM
ejpam-6638	251	10	)	)	PUNCT
ejpam-6638	251	11	(	(	PUNCT
ejpam-6638	251	12	2025	2025	NUM
ejpam-6638	251	13	)	)	PUNCT
ejpam-6638	251	14	,	,	PUNCT
ejpam-6638	251	15	6638	6638	NUM
ejpam-6638	251	16	13	13	NUM
ejpam-6638	251	17	of	of	ADP
ejpam-6638	251	18	22	22	NUM
ejpam-6638	251	19	(	(	PUNCT
ejpam-6638	251	20	a	a	NOUN
ejpam-6638	251	21	)	)	PUNCT
ejpam-6638	251	22	(	(	PUNCT
ejpam-6638	251	23	b	b	X
ejpam-6638	251	24	)	)	PUNCT
ejpam-6638	251	25	(	(	PUNCT
ejpam-6638	251	26	c	c	X
ejpam-6638	251	27	)	)	PUNCT
ejpam-6638	251	28	(	(	PUNCT
ejpam-6638	251	29	e	e	NOUN
ejpam-6638	251	30	)	)	PUNCT
ejpam-6638	251	31	(	(	PUNCT
ejpam-6638	251	32	f	f	X
ejpam-6638	251	33	)	)	PUNCT
ejpam-6638	251	34	figure	figure	NOUN
ejpam-6638	251	35	7	7	NUM
ejpam-6638	251	36	:	:	PUNCT
ejpam-6638	251	37	the	the	DET
ejpam-6638	251	38	left	left	ADJ
ejpam-6638	251	39	column	column	NOUN
ejpam-6638	251	40	presents	present	VERB
ejpam-6638	251	41	the	the	DET
ejpam-6638	251	42	training	training	NOUN
ejpam-6638	251	43	graphs	graph	NOUN
ejpam-6638	251	44	for	for	ADP
ejpam-6638	251	45	the	the	DET
ejpam-6638	251	46	denoising	denoising	NOUN
ejpam-6638	251	47	task	task	NOUN
ejpam-6638	251	48	,	,	PUNCT
ejpam-6638	251	49	showing	show	VERB
ejpam-6638	251	50	the	the	DET
ejpam-6638	251	51	relationships	relationship	NOUN
ejpam-6638	251	52	between	between	ADP
ejpam-6638	251	53	epochs	epoch	NOUN
ejpam-6638	251	54	and	and	CCONJ
ejpam-6638	251	55	loss	loss	NOUN
ejpam-6638	251	56	,	,	PUNCT
ejpam-6638	251	57	psnr	psnr	NOUN
ejpam-6638	251	58	,	,	PUNCT
ejpam-6638	251	59	and	and	CCONJ
ejpam-6638	251	60	ssim	ssim	NOUN
ejpam-6638	251	61	.	.	PUNCT
ejpam-6638	252	1	the	the	DET
ejpam-6638	252	2	right	right	ADJ
ejpam-6638	252	3	column	column	NOUN
ejpam-6638	252	4	displays	display	VERB
ejpam-6638	252	5	the	the	DET
ejpam-6638	252	6	corresponding	correspond	VERB
ejpam-6638	252	7	plots	plot	NOUN
ejpam-6638	252	8	for	for	ADP
ejpam-6638	252	9	the	the	DET
ejpam-6638	252	10	inpainting	inpainting	ADJ
ejpam-6638	252	11	task	task	NOUN
ejpam-6638	252	12	,	,	PUNCT
ejpam-6638	252	13	illustrating	illustrate	VERB
ejpam-6638	252	14	how	how	SCONJ
ejpam-6638	252	15	the	the	DET
ejpam-6638	252	16	loss	loss	NOUN
ejpam-6638	252	17	,	,	PUNCT
ejpam-6638	252	18	psnr	psnr	NOUN
ejpam-6638	252	19	,	,	PUNCT
ejpam-6638	252	20	and	and	CCONJ
ejpam-6638	252	21	ssim	ssim	NOUN
ejpam-6638	252	22	evolve	evolve	VERB
ejpam-6638	252	23	over	over	ADP
ejpam-6638	252	24	training	training	NOUN
ejpam-6638	252	25	epochs	epoch	NOUN
ejpam-6638	252	26	.	.	PUNCT
ejpam-6638	253	1	structure	structure	NOUN
ejpam-6638	253	2	-	-	PUNCT
ejpam-6638	253	3	aware	aware	ADJ
ejpam-6638	253	4	completion	completion	NOUN
ejpam-6638	253	5	is	be	AUX
ejpam-6638	253	6	carried	carry	VERB
ejpam-6638	253	7	out	out	ADP
ejpam-6638	253	8	via	via	ADP
ejpam-6638	253	9	the	the	DET
ejpam-6638	253	10	svd	svd	PROPN
ejpam-6638	253	11	-	-	PUNCT
ejpam-6638	253	12	modulated	modulate	VERB
ejpam-6638	253	13	weights	weight	NOUN
ejpam-6638	253	14	using	use	VERB
ejpam-6638	253	15	:	:	PUNCT
ejpam-6638	253	16	winpaint	winpaint	NOUN
ejpam-6638	254	1	=	=	SYM
ejpam-6638	254	2	k∑	k∑	PROPN
ejpam-6638	254	3	k=1	k=1	X
ejpam-6638	255	1	σkukv	σkukv	PROPN
ejpam-6638	255	2	⊤	⊤	PROPN
ejpam-6638	255	3	k	k	PROPN
ejpam-6638	255	4	(	(	PUNCT
ejpam-6638	255	5	k	k	NOUN
ejpam-6638	255	6	=	=	SYM
ejpam-6638	255	7	3	3	NUM
ejpam-6638	255	8	)	)	PUNCT
ejpam-6638	255	9	(	(	PUNCT
ejpam-6638	255	10	4.4	4.4	NUM
ejpam-6638	255	11	)	)	PUNCT
ejpam-6638	255	12	m.	m.	NOUN
ejpam-6638	255	13	israr	israr	PROPN
ejpam-6638	255	14	et	et	PROPN
ejpam-6638	255	15	al	al	PROPN
ejpam-6638	255	16	.	.	PUNCT
ejpam-6638	255	17	/	/	SYM
ejpam-6638	255	18	eur	eur	PROPN
ejpam-6638	255	19	.	.	PUNCT
ejpam-6638	256	1	j.	j.	PROPN
ejpam-6638	256	2	pure	pure	PROPN
ejpam-6638	256	3	appl	appl	PROPN
ejpam-6638	256	4	.	.	PROPN
ejpam-6638	256	5	math	math	PROPN
ejpam-6638	256	6	,	,	PUNCT
ejpam-6638	256	7	18	18	NUM
ejpam-6638	256	8	(	(	PUNCT
ejpam-6638	256	9	4	4	NUM
ejpam-6638	256	10	)	)	PUNCT
ejpam-6638	256	11	(	(	PUNCT
ejpam-6638	256	12	2025	2025	NUM
ejpam-6638	256	13	)	)	PUNCT
ejpam-6638	256	14	,	,	PUNCT
ejpam-6638	256	15	6638	6638	NUM
ejpam-6638	256	16	14	14	NUM
ejpam-6638	256	17	of	of	ADP
ejpam-6638	256	18	22	22	NUM
ejpam-6638	256	19	where	where	SCONJ
ejpam-6638	256	20	the	the	DET
ejpam-6638	256	21	top	top	ADJ
ejpam-6638	256	22	three	three	NUM
ejpam-6638	256	23	singular	singular	ADJ
ejpam-6638	256	24	components	component	NOUN
ejpam-6638	256	25	capture	capture	VERB
ejpam-6638	256	26	multi	multi	ADJ
ejpam-6638	256	27	-	-	ADJ
ejpam-6638	256	28	scale	scale	ADJ
ejpam-6638	256	29	context	context	NOUN
ejpam-6638	256	30	for	for	ADP
ejpam-6638	256	31	believable	believable	ADJ
ejpam-6638	256	32	infilling	infilling	NOUN
ejpam-6638	256	33	.	.	PUNCT
ejpam-6638	257	1	here	here	ADV
ejpam-6638	257	2	,	,	PUNCT
ejpam-6638	257	3	k	k	PROPN
ejpam-6638	257	4	=	=	SYM
ejpam-6638	257	5	3	3	NUM
ejpam-6638	257	6	is	be	AUX
ejpam-6638	257	7	selected	select	VERB
ejpam-6638	257	8	based	base	VERB
ejpam-6638	257	9	on	on	ADP
ejpam-6638	257	10	empirical	empirical	ADJ
ejpam-6638	257	11	evaluation	evaluation	NOUN
ejpam-6638	257	12	,	,	PUNCT
ejpam-6638	257	13	which	which	PRON
ejpam-6638	257	14	shows	show	VERB
ejpam-6638	257	15	that	that	SCONJ
ejpam-6638	257	16	the	the	DET
ejpam-6638	257	17	first	first	ADJ
ejpam-6638	257	18	three	three	NUM
ejpam-6638	257	19	singular	singular	ADJ
ejpam-6638	257	20	values	value	NOUN
ejpam-6638	257	21	are	be	AUX
ejpam-6638	257	22	sufficient	sufficient	ADJ
ejpam-6638	257	23	to	to	PART
ejpam-6638	257	24	capture	capture	VERB
ejpam-6638	257	25	the	the	DET
ejpam-6638	257	26	dominant	dominant	ADJ
ejpam-6638	257	27	image	image	NOUN
ejpam-6638	257	28	structures	structure	NOUN
ejpam-6638	257	29	while	while	SCONJ
ejpam-6638	257	30	suppressing	suppress	VERB
ejpam-6638	257	31	noise	noise	NOUN
ejpam-6638	257	32	.	.	PUNCT
ejpam-6638	258	1	•	•	NUM
ejpam-6638	258	2	activation	activation	NOUN
ejpam-6638	258	3	function	function	NOUN
ejpam-6638	258	4	(	(	PUNCT
ejpam-6638	258	5	ssp	ssp	NOUN
ejpam-6638	258	6	):	):	PUNCT
ejpam-6638	258	7	all	all	DET
ejpam-6638	258	8	convolutional	convolutional	ADJ
ejpam-6638	258	9	activations	activation	NOUN
ejpam-6638	258	10	in	in	ADP
ejpam-6638	258	11	dip	dip	NOUN
ejpam-6638	258	12	-	-	PUNCT
ejpam-6638	258	13	svd	svd	PROPN
ejpam-6638	258	14	use	use	VERB
ejpam-6638	258	15	the	the	DET
ejpam-6638	258	16	shifted	shift	VERB
ejpam-6638	258	17	softplus	softplus	ADJ
ejpam-6638	258	18	(	(	PUNCT
ejpam-6638	258	19	ssp	ssp	NOUN
ejpam-6638	258	20	)	)	PUNCT
ejpam-6638	258	21	function	function	NOUN
ejpam-6638	258	22	:	:	PUNCT
ejpam-6638	258	23	ssp(x	ssp(x	PROPN
ejpam-6638	258	24	)	)	PUNCT
ejpam-6638	258	25	=	=	SYM
ejpam-6638	259	1	ln(1	ln(1	NOUN
ejpam-6638	259	2	+	+	NUM
ejpam-6638	259	3	ex)−	ex)−	NOUN
ejpam-6638	259	4	ln(2	ln(2	NOUN
ejpam-6638	259	5	)	)	PUNCT
ejpam-6638	259	6	(	(	PUNCT
ejpam-6638	259	7	4.5	4.5	NUM
ejpam-6638	259	8	)	)	PUNCT
ejpam-6638	259	9	which	which	PRON
ejpam-6638	259	10	ensures	ensure	VERB
ejpam-6638	259	11	smooth	smooth	ADJ
ejpam-6638	259	12	,	,	PUNCT
ejpam-6638	259	13	positive	positive	ADJ
ejpam-6638	259	14	activations	activation	NOUN
ejpam-6638	259	15	and	and	CCONJ
ejpam-6638	259	16	improves	improve	VERB
ejpam-6638	259	17	convergence	convergence	NOUN
ejpam-6638	259	18	during	during	ADP
ejpam-6638	259	19	inpainting	inpainte	VERB
ejpam-6638	259	20	.	.	PUNCT
ejpam-6638	260	1	the	the	DET
ejpam-6638	260	2	ssp	ssp	ADJ
ejpam-6638	260	3	term	term	NOUN
ejpam-6638	260	4	is	be	AUX
ejpam-6638	260	5	defined	define	VERB
ejpam-6638	260	6	earlier	early	ADV
ejpam-6638	260	7	in	in	ADP
ejpam-6638	260	8	the	the	DET
ejpam-6638	260	9	manuscript	manuscript	NOUN
ejpam-6638	260	10	for	for	ADP
ejpam-6638	260	11	clarity	clarity	NOUN
ejpam-6638	260	12	.	.	PUNCT
ejpam-6638	261	1	weight	weight	NOUN
ejpam-6638	261	2	extraction	extraction	NOUN
ejpam-6638	261	3	methodology	methodology	NOUN
ejpam-6638	261	4	:	:	PUNCT
ejpam-6638	261	5	all	all	PRON
ejpam-6638	261	6	of	of	ADP
ejpam-6638	261	7	the	the	DET
ejpam-6638	261	8	weights	weight	NOUN
ejpam-6638	261	9	from	from	ADP
ejpam-6638	261	10	the	the	DET
ejpam-6638	261	11	penultimate	penultimate	NOUN
ejpam-6638	261	12	convolutional	convolutional	ADJ
ejpam-6638	261	13	layer	layer	NOUN
ejpam-6638	261	14	were	be	AUX
ejpam-6638	261	15	extracted	extract	VERB
ejpam-6638	261	16	using	use	VERB
ejpam-6638	261	17	:	:	PUNCT
ejpam-6638	261	18	(	(	PUNCT
ejpam-6638	261	19	i	i	NOUN
ejpam-6638	261	20	)	)	PUNCT
ejpam-6638	261	21	forward	forward	ADV
ejpam-6638	261	22	pass	pass	VERB
ejpam-6638	261	23	of	of	ADP
ejpam-6638	261	24	noisy	noisy	ADJ
ejpam-6638	261	25	or	or	CCONJ
ejpam-6638	261	26	masked	masked	ADJ
ejpam-6638	261	27	images	image	NOUN
ejpam-6638	261	28	(	(	PUNCT
ejpam-6638	261	29	ii	ii	NOUN
ejpam-6638	261	30	)	)	PUNCT
ejpam-6638	261	31	neural	neural	ADJ
ejpam-6638	261	32	sorting	sorting	NOUN
ejpam-6638	261	33	by	by	ADP
ejpam-6638	261	34	score	score	NOUN
ejpam-6638	261	35	=	=	PUNCT
ejpam-6638	261	36	|µw	|µw	PRON
ejpam-6638	261	37	|+	|+	NOUN
ejpam-6638	261	38	|b|+	|b|+	NOUN
ejpam-6638	261	39	σw	σw	PROPN
ejpam-6638	261	40	(	(	PUNCT
ejpam-6638	261	41	iii	iii	NOUN
ejpam-6638	261	42	)	)	PUNCT
ejpam-6638	261	43	selecting	select	VERB
ejpam-6638	261	44	the	the	DET
ejpam-6638	261	45	top	top	ADJ
ejpam-6638	261	46	15	15	NUM
ejpam-6638	261	47	neurons	neuron	NOUN
ejpam-6638	261	48	for	for	ADP
ejpam-6638	261	49	each	each	DET
ejpam-6638	261	50	task	task	NOUN
ejpam-6638	261	51	4.3	4.3	NUM
ejpam-6638	261	52	.	.	PUNCT
ejpam-6638	262	1	ablation	ablation	NOUN
ejpam-6638	262	2	study	study	NOUN
ejpam-6638	262	3	an	an	DET
ejpam-6638	262	4	ablation	ablation	NOUN
ejpam-6638	262	5	study	study	NOUN
ejpam-6638	262	6	systematically	systematically	ADV
ejpam-6638	262	7	evaluates	evaluate	VERB
ejpam-6638	262	8	the	the	DET
ejpam-6638	262	9	contribution	contribution	NOUN
ejpam-6638	262	10	of	of	ADP
ejpam-6638	262	11	individual	individual	ADJ
ejpam-6638	262	12	components	component	NOUN
ejpam-6638	262	13	or	or	CCONJ
ejpam-6638	262	14	design	design	NOUN
ejpam-6638	262	15	choices	choice	NOUN
ejpam-6638	262	16	to	to	ADP
ejpam-6638	262	17	overall	overall	ADJ
ejpam-6638	262	18	performance	performance	NOUN
ejpam-6638	262	19	.	.	PUNCT
ejpam-6638	263	1	in	in	ADP
ejpam-6638	263	2	this	this	DET
ejpam-6638	263	3	work	work	NOUN
ejpam-6638	263	4	,	,	PUNCT
ejpam-6638	263	5	we	we	PRON
ejpam-6638	263	6	conduct	conduct	VERB
ejpam-6638	263	7	an	an	DET
ejpam-6638	263	8	ablation	ablation	NOUN
ejpam-6638	263	9	study	study	NOUN
ejpam-6638	263	10	with	with	ADP
ejpam-6638	263	11	respect	respect	NOUN
ejpam-6638	263	12	to	to	ADP
ejpam-6638	263	13	the	the	DET
ejpam-6638	263	14	optimizer	optimizer	NOUN
ejpam-6638	263	15	only	only	ADV
ejpam-6638	263	16	,	,	PUNCT
ejpam-6638	263	17	isolating	isolate	VERB
ejpam-6638	263	18	its	its	PRON
ejpam-6638	263	19	effect	effect	NOUN
ejpam-6638	263	20	on	on	ADP
ejpam-6638	263	21	convergence	convergence	NOUN
ejpam-6638	263	22	speed	speed	NOUN
ejpam-6638	263	23	and	and	CCONJ
ejpam-6638	263	24	denoising	denoising	NOUN
ejpam-6638	263	25	/	/	SYM
ejpam-6638	263	26	inpainting	inpainte	VERB
ejpam-6638	263	27	quality	quality	NOUN
ejpam-6638	263	28	in	in	ADP
ejpam-6638	263	29	the	the	DET
ejpam-6638	263	30	dip	dip	NOUN
ejpam-6638	263	31	-	-	PUNCT
ejpam-6638	263	32	svd	svd	PROPN
ejpam-6638	263	33	framework	framework	NOUN
ejpam-6638	263	34	.	.	PUNCT
ejpam-6638	264	1	optimizer	optimizer	NOUN
ejpam-6638	264	2	/	/	SYM
ejpam-6638	264	3	image	image	NOUN
ejpam-6638	264	4	1	1	NUM
ejpam-6638	264	5	2	2	NUM
ejpam-6638	264	6	3	3	NUM
ejpam-6638	264	7	4	4	NUM
ejpam-6638	264	8	5	5	NUM
ejpam-6638	264	9	6	6	NUM
ejpam-6638	264	10	psnr	psnr	NOUN
ejpam-6638	264	11	(	(	PUNCT
ejpam-6638	264	12	db	db	PROPN
ejpam-6638	264	13	)	)	PUNCT
ejpam-6638	264	14	adam	adam	PROPN
ejpam-6638	264	15	25.35	25.35	NUM
ejpam-6638	264	16	25.23	25.23	NUM
ejpam-6638	264	17	24.50	24.50	NUM
ejpam-6638	264	18	24.34	24.34	NUM
ejpam-6638	264	19	24.76	24.76	NUM
ejpam-6638	264	20	24.30	24.30	NUM
ejpam-6638	264	21	apollo	apollo	PROPN
ejpam-6638	264	22	21.67	21.67	NUM
ejpam-6638	264	23	23.54	23.54	NUM
ejpam-6638	264	24	23.70	23.70	NUM
ejpam-6638	264	25	21.92	21.92	NUM
ejpam-6638	264	26	26.79	26.79	NUM
ejpam-6638	264	27	22.64	22.64	NUM
ejpam-6638	264	28	diff	diff	NOUN
ejpam-6638	264	29	-	-	PUNCT
ejpam-6638	264	30	grad	grad	NOUN
ejpam-6638	264	31	25.07	25.07	NUM
ejpam-6638	264	32	32.63	32.63	NUM
ejpam-6638	264	33	24.72	24.72	NUM
ejpam-6638	264	34	26.66	26.66	NUM
ejpam-6638	264	35	31.59	31.59	NUM
ejpam-6638	264	36	24.45	24.45	NUM
ejpam-6638	264	37	ssim	ssim	NOUN
ejpam-6638	264	38	adam	adam	PROPN
ejpam-6638	264	39	0.87	0.87	NUM
ejpam-6638	264	40	0.86	0.86	NUM
ejpam-6638	264	41	0.90	0.90	NUM
ejpam-6638	264	42	0.81	0.81	NUM
ejpam-6638	264	43	0.68	0.68	NUM
ejpam-6638	264	44	0.92	0.92	NUM
ejpam-6638	264	45	apollo	apollo	PROPN
ejpam-6638	264	46	0.81	0.81	NUM
ejpam-6638	264	47	0.83	0.83	NUM
ejpam-6638	264	48	0.91	0.91	NUM
ejpam-6638	264	49	0.76	0.76	NUM
ejpam-6638	264	50	0.80	0.80	NUM
ejpam-6638	264	51	0.89	0.89	NUM
ejpam-6638	264	52	diff	diff	NOUN
ejpam-6638	264	53	-	-	PUNCT
ejpam-6638	264	54	grad	grad	NOUN
ejpam-6638	264	55	0.87	0.87	NUM
ejpam-6638	264	56	0.97	0.97	NUM
ejpam-6638	264	57	0.90	0.90	NUM
ejpam-6638	264	58	0.87	0.87	NUM
ejpam-6638	264	59	0.90	0.90	NUM
ejpam-6638	264	60	0.82	0.82	NUM
ejpam-6638	264	61	table	table	NOUN
ejpam-6638	264	62	3	3	NUM
ejpam-6638	264	63	:	:	PUNCT
ejpam-6638	264	64	ablation	ablation	NOUN
ejpam-6638	264	65	study	study	NOUN
ejpam-6638	264	66	:	:	PUNCT
ejpam-6638	264	67	denoising	denoise	VERB
ejpam-6638	264	68	performance	performance	NOUN
ejpam-6638	264	69	(	(	PUNCT
ejpam-6638	264	70	psnr	psnr	NOUN
ejpam-6638	264	71	and	and	CCONJ
ejpam-6638	264	72	ssim	ssim	NOUN
ejpam-6638	264	73	)	)	PUNCT
ejpam-6638	264	74	of	of	ADP
ejpam-6638	264	75	dip	dip	NOUN
ejpam-6638	264	76	-	-	PUNCT
ejpam-6638	264	77	svd	svd	PROPN
ejpam-6638	264	78	using	use	VERB
ejpam-6638	264	79	different	different	ADJ
ejpam-6638	264	80	optimizers	optimizer	NOUN
ejpam-6638	264	81	across	across	ADP
ejpam-6638	264	82	six	six	NUM
ejpam-6638	264	83	sample	sample	NOUN
ejpam-6638	264	84	images	image	NOUN
ejpam-6638	264	85	.	.	PUNCT
ejpam-6638	265	1	4.3.1	4.3.1	X
ejpam-6638	265	2	.	.	PUNCT
ejpam-6638	266	1	denoising	denoise	VERB
ejpam-6638	266	2	using	use	VERB
ejpam-6638	266	3	six	six	NUM
ejpam-6638	266	4	test	test	NOUN
ejpam-6638	266	5	images	image	NOUN
ejpam-6638	266	6	,	,	PUNCT
ejpam-6638	266	7	we	we	PRON
ejpam-6638	266	8	performed	perform	VERB
ejpam-6638	266	9	ablation	ablation	NOUN
ejpam-6638	266	10	research	research	NOUN
ejpam-6638	266	11	to	to	PART
ejpam-6638	266	12	examine	examine	VERB
ejpam-6638	266	13	three	three	NUM
ejpam-6638	266	14	optimization	optimization	NOUN
ejpam-6638	266	15	algorithms	algorithm	NOUN
ejpam-6638	266	16	:	:	PUNCT
ejpam-6638	266	17	adam	adam	PROPN
ejpam-6638	266	18	,	,	PUNCT
ejpam-6638	266	19	apollo	apollo	PROPN
ejpam-6638	266	20	,	,	PUNCT
ejpam-6638	266	21	and	and	CCONJ
ejpam-6638	266	22	diff	diff	NOUN
ejpam-6638	266	23	-	-	PUNCT
ejpam-6638	266	24	grad	grad	NOUN
ejpam-6638	266	25	.	.	PUNCT
ejpam-6638	267	1	this	this	PRON
ejpam-6638	267	2	allowed	allow	VERB
ejpam-6638	267	3	us	we	PRON
ejpam-6638	267	4	to	to	PART
ejpam-6638	267	5	assess	assess	VERB
ejpam-6638	267	6	how	how	SCONJ
ejpam-6638	267	7	the	the	DET
ejpam-6638	267	8	optimizer	optimizer	NOUN
ejpam-6638	267	9	affected	affect	VERB
ejpam-6638	267	10	the	the	DET
ejpam-6638	267	11	dipsvd	dipsvd	ADJ
ejpam-6638	267	12	architecture	architecture	NOUN
ejpam-6638	267	13	’s	’s	PART
ejpam-6638	267	14	ability	ability	NOUN
ejpam-6638	267	15	to	to	PART
ejpam-6638	267	16	reduce	reduce	VERB
ejpam-6638	267	17	noise	noise	NOUN
ejpam-6638	267	18	.	.	PUNCT
ejpam-6638	268	1	table	table	NOUN
ejpam-6638	268	2	3	3	NUM
ejpam-6638	268	3	shows	show	VERB
ejpam-6638	268	4	the	the	DET
ejpam-6638	268	5	structural	structural	ADJ
ejpam-6638	268	6	similarity	similarity	NOUN
ejpam-6638	268	7	index	index	NOUN
ejpam-6638	268	8	(	(	PUNCT
ejpam-6638	268	9	ssim	ssim	NOUN
ejpam-6638	268	10	)	)	PUNCT
ejpam-6638	268	11	scores	score	NOUN
ejpam-6638	268	12	in	in	ADP
ejpam-6638	268	13	the	the	DET
ejpam-6638	268	14	lower	low	ADJ
ejpam-6638	268	15	block	block	NOUN
ejpam-6638	268	16	and	and	CCONJ
ejpam-6638	268	17	the	the	DET
ejpam-6638	268	18	peak	peak	NOUN
ejpam-6638	268	19	signal	signal	NOUN
ejpam-6638	268	20	-	-	PUNCT
ejpam-6638	268	21	to	to	ADP
ejpam-6638	268	22	-	-	PUNCT
ejpam-6638	268	23	noise	noise	NOUN
ejpam-6638	268	24	ratio	ratio	NOUN
ejpam-6638	268	25	(	(	PUNCT
ejpam-6638	268	26	psnr	psnr	NOUN
ejpam-6638	268	27	)	)	PUNCT
ejpam-6638	268	28	values	value	NOUN
ejpam-6638	268	29	in	in	ADP
ejpam-6638	268	30	the	the	DET
ejpam-6638	268	31	upper	upper	ADJ
ejpam-6638	268	32	block	block	NOUN
ejpam-6638	268	33	.	.	PUNCT
ejpam-6638	269	1	to	to	PART
ejpam-6638	269	2	focus	focus	VERB
ejpam-6638	269	3	solely	solely	ADV
ejpam-6638	269	4	on	on	ADP
ejpam-6638	269	5	the	the	DET
ejpam-6638	269	6	effects	effect	NOUN
ejpam-6638	269	7	of	of	ADP
ejpam-6638	269	8	the	the	DET
ejpam-6638	269	9	optimization	optimization	NOUN
ejpam-6638	269	10	method	method	NOUN
ejpam-6638	269	11	,	,	PUNCT
ejpam-6638	269	12	each	each	DET
ejpam-6638	269	13	optimizer	optimizer	NOUN
ejpam-6638	269	14	was	be	AUX
ejpam-6638	269	15	trained	train	VERB
ejpam-6638	269	16	for	for	ADP
ejpam-6638	269	17	3,000	3,000	NUM
ejpam-6638	269	18	epochs	epoch	NOUN
ejpam-6638	269	19	under	under	ADP
ejpam-6638	269	20	identical	identical	ADJ
ejpam-6638	269	21	network	network	NOUN
ejpam-6638	269	22	and	and	CCONJ
ejpam-6638	269	23	training	training	NOUN
ejpam-6638	269	24	conditions	condition	NOUN
ejpam-6638	269	25	.	.	PUNCT
ejpam-6638	270	1	diff	diff	NOUN
ejpam-6638	270	2	-	-	PUNCT
ejpam-6638	270	3	grad	grad	NOUN
ejpam-6638	270	4	consistently	consistently	ADV
ejpam-6638	270	5	achieved	achieve	VERB
ejpam-6638	270	6	the	the	DET
ejpam-6638	270	7	best	good	ADJ
ejpam-6638	270	8	psnr	psnr	NOUN
ejpam-6638	270	9	and	and	CCONJ
ejpam-6638	270	10	ssim	ssim	NOUN
ejpam-6638	270	11	scores	score	NOUN
ejpam-6638	270	12	on	on	ADP
ejpam-6638	270	13	the	the	DET
ejpam-6638	270	14	majority	majority	NOUN
ejpam-6638	270	15	of	of	ADP
ejpam-6638	270	16	the	the	DET
ejpam-6638	270	17	images	image	NOUN
ejpam-6638	270	18	,	,	PUNCT
ejpam-6638	270	19	highlighting	highlight	VERB
ejpam-6638	270	20	its	its	PRON
ejpam-6638	270	21	enhanced	enhanced	ADJ
ejpam-6638	270	22	ability	ability	NOUN
ejpam-6638	270	23	to	to	PART
ejpam-6638	270	24	navigate	navigate	VERB
ejpam-6638	270	25	the	the	DET
ejpam-6638	270	26	intricate	intricate	ADJ
ejpam-6638	270	27	loss	loss	NOUN
ejpam-6638	270	28	landscape	landscape	NOUN
ejpam-6638	270	29	created	create	VERB
ejpam-6638	270	30	by	by	ADP
ejpam-6638	270	31	the	the	DET
ejpam-6638	270	32	svd	svd	PROPN
ejpam-6638	270	33	-	-	PUNCT
ejpam-6638	270	34	based	base	VERB
ejpam-6638	270	35	prior	prior	ADV
ejpam-6638	270	36	.	.	PUNCT
ejpam-6638	271	1	in	in	ADP
ejpam-6638	271	2	particular	particular	ADJ
ejpam-6638	271	3	,	,	PUNCT
ejpam-6638	271	4	diff	diff	NOUN
ejpam-6638	271	5	-	-	PUNCT
ejpam-6638	271	6	grad	grad	NOUN
ejpam-6638	271	7	outperformed	outperform	VERB
ejpam-6638	271	8	both	both	CCONJ
ejpam-6638	271	9	adam	adam	PROPN
ejpam-6638	271	10	and	and	CCONJ
ejpam-6638	271	11	apollo	apollo	PROPN
ejpam-6638	271	12	by	by	ADP
ejpam-6638	271	13	achieving	achieve	VERB
ejpam-6638	271	14	m.	m.	NOUN
ejpam-6638	271	15	israr	israr	PROPN
ejpam-6638	271	16	et	et	PROPN
ejpam-6638	271	17	al	al	PROPN
ejpam-6638	271	18	.	.	PUNCT
ejpam-6638	271	19	/	/	SYM
ejpam-6638	271	20	eur	eur	PROPN
ejpam-6638	271	21	.	.	PUNCT
ejpam-6638	272	1	j.	j.	PROPN
ejpam-6638	272	2	pure	pure	PROPN
ejpam-6638	272	3	appl	appl	PROPN
ejpam-6638	272	4	.	.	PROPN
ejpam-6638	272	5	math	math	PROPN
ejpam-6638	272	6	,	,	PUNCT
ejpam-6638	272	7	18	18	NUM
ejpam-6638	272	8	(	(	PUNCT
ejpam-6638	272	9	4	4	NUM
ejpam-6638	272	10	)	)	PUNCT
ejpam-6638	272	11	(	(	PUNCT
ejpam-6638	272	12	2025	2025	NUM
ejpam-6638	272	13	)	)	PUNCT
ejpam-6638	272	14	,	,	PUNCT
ejpam-6638	272	15	6638	6638	NUM
ejpam-6638	272	16	15	15	NUM
ejpam-6638	272	17	of	of	ADP
ejpam-6638	272	18	22	22	NUM
ejpam-6638	272	19	(	(	PUNCT
ejpam-6638	272	20	a	a	NOUN
ejpam-6638	272	21	)	)	PUNCT
ejpam-6638	272	22	(	(	PUNCT
ejpam-6638	272	23	b	b	X
ejpam-6638	272	24	)	)	PUNCT
ejpam-6638	272	25	(	(	PUNCT
ejpam-6638	272	26	c	c	X
ejpam-6638	272	27	)	)	PUNCT
ejpam-6638	272	28	(	(	PUNCT
ejpam-6638	272	29	d	d	X
ejpam-6638	272	30	)	)	PUNCT
ejpam-6638	272	31	(	(	PUNCT
ejpam-6638	272	32	e	e	NOUN
ejpam-6638	272	33	)	)	PUNCT
ejpam-6638	272	34	(	(	PUNCT
ejpam-6638	272	35	f	f	X
ejpam-6638	272	36	)	)	PUNCT
ejpam-6638	272	37	(	(	PUNCT
ejpam-6638	272	38	g	g	NOUN
ejpam-6638	272	39	)	)	PUNCT
ejpam-6638	272	40	(	(	PUNCT
ejpam-6638	272	41	h	h	NOUN
ejpam-6638	272	42	)	)	PUNCT
ejpam-6638	272	43	(	(	PUNCT
ejpam-6638	272	44	i	i	NOUN
ejpam-6638	272	45	)	)	PUNCT
ejpam-6638	272	46	figure	figure	VERB
ejpam-6638	272	47	8	8	NUM
ejpam-6638	272	48	:	:	PUNCT
ejpam-6638	272	49	training	training	NOUN
ejpam-6638	272	50	loss	loss	NOUN
ejpam-6638	272	51	,	,	PUNCT
ejpam-6638	272	52	psnr	psnr	NOUN
ejpam-6638	272	53	,	,	PUNCT
ejpam-6638	272	54	and	and	CCONJ
ejpam-6638	272	55	ssim	ssim	NOUN
ejpam-6638	272	56	across	across	ADP
ejpam-6638	272	57	3000	3000	NUM
ejpam-6638	272	58	epochs	epoch	NOUN
ejpam-6638	272	59	for	for	ADP
ejpam-6638	272	60	each	each	DET
ejpam-6638	272	61	optimizer	optimizer	NOUN
ejpam-6638	272	62	on	on	ADP
ejpam-6638	272	63	four	four	NUM
ejpam-6638	272	64	selected	select	VERB
ejpam-6638	272	65	images	image	NOUN
ejpam-6638	272	66	in	in	ADP
ejpam-6638	272	67	case	case	NOUN
ejpam-6638	272	68	of	of	ADP
ejpam-6638	272	69	denoising	denoising	NOUN
ejpam-6638	272	70	.	.	PUNCT
ejpam-6638	273	1	the	the	DET
ejpam-6638	273	2	first	first	ADJ
ejpam-6638	273	3	row	row	NOUN
ejpam-6638	273	4	shows	show	VERB
ejpam-6638	273	5	the	the	DET
ejpam-6638	273	6	training	training	NOUN
ejpam-6638	273	7	loss	loss	NOUN
ejpam-6638	273	8	,	,	PUNCT
ejpam-6638	273	9	the	the	DET
ejpam-6638	273	10	second	second	ADJ
ejpam-6638	273	11	row	row	NOUN
ejpam-6638	273	12	displays	display	VERB
ejpam-6638	273	13	psnr	psnr	NOUN
ejpam-6638	273	14	,	,	PUNCT
ejpam-6638	273	15	and	and	CCONJ
ejpam-6638	273	16	the	the	DET
ejpam-6638	273	17	third	third	ADJ
ejpam-6638	273	18	row	row	NOUN
ejpam-6638	273	19	presents	present	NOUN
ejpam-6638	273	20	ssim	ssim	NOUN
ejpam-6638	273	21	.	.	PUNCT
ejpam-6638	274	1	a	a	DET
ejpam-6638	274	2	psnr	psnr	NOUN
ejpam-6638	274	3	of	of	ADP
ejpam-6638	274	4	32.63	32.63	NUM
ejpam-6638	274	5	db	db	NOUN
ejpam-6638	274	6	and	and	CCONJ
ejpam-6638	274	7	an	an	DET
ejpam-6638	274	8	ssim	ssim	NOUN
ejpam-6638	274	9	of	of	ADP
ejpam-6638	274	10	0.97	0.97	NUM
ejpam-6638	274	11	for	for	ADP
ejpam-6638	274	12	image	image	NOUN
ejpam-6638	274	13	2	2	NUM
ejpam-6638	274	14	.	.	PUNCT
ejpam-6638	275	1	apollo	apollo	PROPN
ejpam-6638	275	2	prioritizes	prioritize	VERB
ejpam-6638	275	3	preserving	preserve	VERB
ejpam-6638	275	4	structural	structural	ADJ
ejpam-6638	275	5	similarity	similarity	NOUN
ejpam-6638	275	6	over	over	ADP
ejpam-6638	275	7	pixel	pixel	ADJ
ejpam-6638	275	8	-	-	PUNCT
ejpam-6638	275	9	level	level	NOUN
ejpam-6638	275	10	accuracy	accuracy	NOUN
ejpam-6638	275	11	,	,	PUNCT
ejpam-6638	275	12	as	as	SCONJ
ejpam-6638	275	13	evidenced	evidence	VERB
ejpam-6638	275	14	by	by	ADP
ejpam-6638	275	15	its	its	PRON
ejpam-6638	275	16	competitive	competitive	ADJ
ejpam-6638	275	17	ssim	ssim	NOUN
ejpam-6638	275	18	performance	performance	NOUN
ejpam-6638	275	19	on	on	ADP
ejpam-6638	275	20	some	some	DET
ejpam-6638	275	21	images	image	NOUN
ejpam-6638	275	22	but	but	CCONJ
ejpam-6638	275	23	its	its	PRON
ejpam-6638	275	24	typically	typically	ADV
ejpam-6638	275	25	poor	poor	ADJ
ejpam-6638	275	26	psnr	psnr	NOUN
ejpam-6638	275	27	.	.	PUNCT
ejpam-6638	276	1	adam	adam	PROPN
ejpam-6638	276	2	,	,	PUNCT
ejpam-6638	276	3	however	however	ADV
ejpam-6638	276	4	,	,	PUNCT
ejpam-6638	276	5	delivers	deliver	VERB
ejpam-6638	276	6	consistent	consistent	ADJ
ejpam-6638	276	7	but	but	CCONJ
ejpam-6638	276	8	mediocre	mediocre	ADJ
ejpam-6638	276	9	outcomes	outcome	NOUN
ejpam-6638	276	10	in	in	ADP
ejpam-6638	276	11	every	every	DET
ejpam-6638	276	12	parameter	parameter	NOUN
ejpam-6638	276	13	.	.	PUNCT
ejpam-6638	277	1	these	these	DET
ejpam-6638	277	2	results	result	NOUN
ejpam-6638	277	3	show	show	VERB
ejpam-6638	277	4	that	that	SCONJ
ejpam-6638	277	5	the	the	DET
ejpam-6638	277	6	optimizer	optimizer	NOUN
ejpam-6638	277	7	choice	choice	NOUN
ejpam-6638	277	8	has	have	VERB
ejpam-6638	277	9	a	a	DET
ejpam-6638	277	10	considerable	considerable	ADJ
ejpam-6638	277	11	impact	impact	NOUN
ejpam-6638	277	12	on	on	ADP
ejpam-6638	277	13	both	both	CCONJ
ejpam-6638	277	14	the	the	DET
ejpam-6638	277	15	numeric	numeric	ADJ
ejpam-6638	277	16	fidelity	fidelity	NOUN
ejpam-6638	277	17	and	and	CCONJ
ejpam-6638	277	18	the	the	DET
ejpam-6638	277	19	perceived	perceive	VERB
ejpam-6638	277	20	quality	quality	NOUN
ejpam-6638	277	21	in	in	ADP
ejpam-6638	277	22	dip	dip	NOUN
ejpam-6638	277	23	-	-	PUNCT
ejpam-6638	277	24	svd	svd	PROPN
ejpam-6638	277	25	,	,	PUNCT
ejpam-6638	277	26	and	and	CCONJ
ejpam-6638	277	27	that	that	SCONJ
ejpam-6638	277	28	diff	diff	NOUN
ejpam-6638	277	29	-	-	PUNCT
ejpam-6638	277	30	grad	grad	NOUN
ejpam-6638	277	31	is	be	AUX
ejpam-6638	277	32	the	the	DET
ejpam-6638	277	33	best	good	ADJ
ejpam-6638	277	34	optimizer	optimizer	NOUN
ejpam-6638	277	35	for	for	ADP
ejpam-6638	277	36	our	our	PRON
ejpam-6638	277	37	denoising	denoising	NOUN
ejpam-6638	277	38	goals	goal	NOUN
ejpam-6638	277	39	.	.	PUNCT
ejpam-6638	278	1	•	•	NOUN
ejpam-6638	278	2	training	training	NOUN
ejpam-6638	278	3	loss	loss	NOUN
ejpam-6638	278	4	analysis	analysis	NOUN
ejpam-6638	278	5	:	:	PUNCT
ejpam-6638	278	6	the	the	DET
ejpam-6638	278	7	training	training	NOUN
ejpam-6638	278	8	loss	loss	NOUN
ejpam-6638	278	9	histories	history	NOUN
ejpam-6638	278	10	of	of	ADP
ejpam-6638	278	11	three	three	NUM
ejpam-6638	278	12	optimizers	optimizer	NOUN
ejpam-6638	278	13	(	(	PUNCT
ejpam-6638	278	14	apollo	apollo	PROPN
ejpam-6638	278	15	,	,	PUNCT
ejpam-6638	278	16	adam	adam	PROPN
ejpam-6638	278	17	,	,	PUNCT
ejpam-6638	278	18	and	and	CCONJ
ejpam-6638	278	19	diffgrad	diffgrad	PROPN
ejpam-6638	278	20	)	)	PUNCT
ejpam-6638	278	21	across	across	ADP
ejpam-6638	278	22	3000	3000	NUM
ejpam-6638	278	23	epochs	epoch	NOUN
ejpam-6638	278	24	are	be	AUX
ejpam-6638	278	25	compared	compare	VERB
ejpam-6638	278	26	in	in	ADP
ejpam-6638	278	27	the	the	DET
ejpam-6638	278	28	first	first	ADJ
ejpam-6638	278	29	row	row	NOUN
ejpam-6638	278	30	of	of	ADP
ejpam-6638	278	31	fig	fig	NOUN
ejpam-6638	278	32	.	.	PUNCT
ejpam-6638	279	1	8	8	NUM
ejpam-6638	279	2	.	.	PUNCT
ejpam-6638	280	1	the	the	DET
ejpam-6638	280	2	asymptotic	asymptotic	ADJ
ejpam-6638	280	3	behavior	behavior	NOUN
ejpam-6638	280	4	of	of	ADP
ejpam-6638	280	5	all	all	DET
ejpam-6638	280	6	optimizers	optimizer	NOUN
ejpam-6638	280	7	is	be	AUX
ejpam-6638	280	8	comparable	comparable	ADJ
ejpam-6638	280	9	,	,	PUNCT
ejpam-6638	280	10	but	but	CCONJ
ejpam-6638	280	11	adam	adam	PROPN
ejpam-6638	280	12	’s	’s	PART
ejpam-6638	280	13	variance	variance	NOUN
ejpam-6638	280	14	in	in	ADP
ejpam-6638	280	15	later	later	ADJ
ejpam-6638	280	16	epochs	epoch	NOUN
ejpam-6638	280	17	is	be	AUX
ejpam-6638	280	18	slightly	slightly	ADV
ejpam-6638	280	19	greater	great	ADJ
ejpam-6638	280	20	—	—	PUNCT
ejpam-6638	280	21	this	this	PRON
ejpam-6638	280	22	is	be	AUX
ejpam-6638	280	23	especially	especially	ADV
ejpam-6638	280	24	evident	evident	ADJ
ejpam-6638	280	25	in	in	ADP
ejpam-6638	280	26	the	the	DET
ejpam-6638	280	27	subplot	subplot	NOUN
ejpam-6638	280	28	on	on	ADP
ejpam-6638	280	29	the	the	DET
ejpam-6638	280	30	right	right	NOUN
ejpam-6638	280	31	.	.	PUNCT
ejpam-6638	281	1	these	these	DET
ejpam-6638	281	2	charts	chart	NOUN
ejpam-6638	281	3	show	show	VERB
ejpam-6638	281	4	how	how	SCONJ
ejpam-6638	281	5	the	the	DET
ejpam-6638	281	6	trade	trade	NOUN
ejpam-6638	281	7	-	-	PUNCT
ejpam-6638	281	8	off	off	NOUN
ejpam-6638	281	9	between	between	ADP
ejpam-6638	281	10	stability	stability	NOUN
ejpam-6638	281	11	and	and	CCONJ
ejpam-6638	281	12	convergence	convergence	NOUN
ejpam-6638	281	13	speed	speed	NOUN
ejpam-6638	281	14	during	during	ADP
ejpam-6638	281	15	dip	dip	NOUN
ejpam-6638	281	16	-	-	PUNCT
ejpam-6638	281	17	svd	svd	PROPN
ejpam-6638	281	18	training	training	NOUN
ejpam-6638	281	19	is	be	AUX
ejpam-6638	281	20	influenced	influence	VERB
ejpam-6638	281	21	by	by	ADP
ejpam-6638	281	22	optimizer	optimizer	NOUN
ejpam-6638	281	23	selection	selection	NOUN
ejpam-6638	281	24	.	.	PUNCT
ejpam-6638	282	1	•	•	NUM
ejpam-6638	282	2	psnr	psnr	NOUN
ejpam-6638	282	3	performance	performance	NOUN
ejpam-6638	282	4	:	:	PUNCT
ejpam-6638	282	5	the	the	DET
ejpam-6638	282	6	second	second	ADJ
ejpam-6638	282	7	row	row	NOUN
ejpam-6638	282	8	of	of	ADP
ejpam-6638	282	9	fig	fig	NOUN
ejpam-6638	282	10	.	.	PUNCT
ejpam-6638	283	1	8	8	NUM
ejpam-6638	283	2	demonstrates	demonstrate	VERB
ejpam-6638	283	3	our	our	PRON
ejpam-6638	283	4	method	method	NOUN
ejpam-6638	283	5	’s	’s	PART
ejpam-6638	283	6	robust	robust	ADJ
ejpam-6638	283	7	psnr	psnr	NOUN
ejpam-6638	283	8	improvement	improvement	NOUN
ejpam-6638	283	9	across	across	ADP
ejpam-6638	283	10	all	all	DET
ejpam-6638	283	11	optimizers	optimizer	NOUN
ejpam-6638	283	12	(	(	PUNCT
ejpam-6638	283	13	apollo	apollo	PROPN
ejpam-6638	283	14	,	,	PUNCT
ejpam-6638	283	15	adam	adam	PROPN
ejpam-6638	283	16	,	,	PUNCT
ejpam-6638	283	17	diffgrad	diffgrad	PROPN
ejpam-6638	283	18	)	)	PUNCT
ejpam-6638	283	19	,	,	PUNCT
ejpam-6638	283	20	with	with	ADP
ejpam-6638	283	21	consistent	consistent	ADJ
ejpam-6638	283	22	upward	upward	ADJ
ejpam-6638	283	23	trends	trend	NOUN
ejpam-6638	283	24	stabilizing	stabilize	VERB
ejpam-6638	283	25	after	after	ADP
ejpam-6638	283	26	3000	3000	NUM
ejpam-6638	283	27	epochs	epoch	NOUN
ejpam-6638	283	28	.	.	PUNCT
ejpam-6638	284	1	while	while	SCONJ
ejpam-6638	284	2	convergence	convergence	NOUN
ejpam-6638	284	3	rates	rate	NOUN
ejpam-6638	284	4	differ	differ	VERB
ejpam-6638	284	5	,	,	PUNCT
ejpam-6638	284	6	each	each	DET
ejpam-6638	284	7	optimizer	optimizer	NOUN
ejpam-6638	284	8	achieves	achieve	VERB
ejpam-6638	284	9	comparable	comparable	ADJ
ejpam-6638	284	10	final	final	ADJ
ejpam-6638	284	11	psnr	psnr	NOUN
ejpam-6638	284	12	values	value	NOUN
ejpam-6638	284	13	,	,	PUNCT
ejpam-6638	284	14	confirming	confirm	VERB
ejpam-6638	284	15	the	the	DET
ejpam-6638	284	16	architecture	architecture	NOUN
ejpam-6638	284	17	’s	’s	PART
ejpam-6638	284	18	optimizer	optimizer	NOUN
ejpam-6638	284	19	-	-	PUNCT
ejpam-6638	284	20	agnostic	agnostic	ADJ
ejpam-6638	284	21	stability	stability	NOUN
ejpam-6638	284	22	.	.	PUNCT
ejpam-6638	285	1	this	this	DET
ejpam-6638	285	2	pattern	pattern	NOUN
ejpam-6638	285	3	holds	hold	VERB
ejpam-6638	285	4	uniformly	uniformly	ADV
ejpam-6638	285	5	across	across	ADP
ejpam-6638	285	6	all	all	DET
ejpam-6638	285	7	four	four	NUM
ejpam-6638	285	8	test	test	NOUN
ejpam-6638	285	9	images	image	NOUN
ejpam-6638	285	10	,	,	PUNCT
ejpam-6638	285	11	evidenced	evidence	VERB
ejpam-6638	285	12	by	by	ADP
ejpam-6638	285	13	the	the	DET
ejpam-6638	285	14	parallel	parallel	ADJ
ejpam-6638	285	15	curves	curve	NOUN
ejpam-6638	285	16	in	in	ADP
ejpam-6638	285	17	each	each	DET
ejpam-6638	285	18	subplot	subplot	NOUN
ejpam-6638	285	19	.	.	PUNCT
ejpam-6638	286	1	•	•	NUM
ejpam-6638	286	2	psnr	psnr	NOUN
ejpam-6638	286	3	performance	performance	NOUN
ejpam-6638	286	4	:	:	PUNCT
ejpam-6638	286	5	fig	fig	NOUN
ejpam-6638	286	6	.	.	PUNCT
ejpam-6638	287	1	8	8	NUM
ejpam-6638	287	2	’s	’s	PART
ejpam-6638	287	3	ssim	ssim	NOUN
ejpam-6638	287	4	results	result	NOUN
ejpam-6638	287	5	(	(	PUNCT
ejpam-6638	287	6	third	third	ADJ
ejpam-6638	287	7	row	row	NOUN
ejpam-6638	287	8	)	)	PUNCT
ejpam-6638	287	9	demonstrate	demonstrate	VERB
ejpam-6638	287	10	our	our	PRON
ejpam-6638	287	11	method	method	NOUN
ejpam-6638	287	12	’s	’s	PART
ejpam-6638	287	13	structural	structural	ADJ
ejpam-6638	287	14	preservation	preservation	NOUN
ejpam-6638	287	15	capability	capability	NOUN
ejpam-6638	287	16	,	,	PUNCT
ejpam-6638	287	17	with	with	ADP
ejpam-6638	287	18	all	all	DET
ejpam-6638	287	19	optimizers	optimizer	NOUN
ejpam-6638	287	20	achieving	achieve	VERB
ejpam-6638	287	21	>	>	X
ejpam-6638	287	22	0.90	0.90	NUM
ejpam-6638	287	23	ssim	ssim	NOUN
ejpam-6638	287	24	by	by	ADP
ejpam-6638	287	25	epoch	epoch	PROPN
ejpam-6638	287	26	3000	3000	NUM
ejpam-6638	287	27	.	.	PUNCT
ejpam-6638	288	1	the	the	DET
ejpam-6638	288	2	curves	curve	NOUN
ejpam-6638	288	3	’	'	PUNCT
ejpam-6638	288	4	asymptotic	asymptotic	ADJ
ejpam-6638	288	5	convergence	convergence	NOUN
ejpam-6638	288	6	indicates	indicate	VERB
ejpam-6638	288	7	robust	robust	ADJ
ejpam-6638	288	8	learning	learning	NOUN
ejpam-6638	288	9	of	of	ADP
ejpam-6638	288	10	perceptual	perceptual	ADJ
ejpam-6638	288	11	features	feature	NOUN
ejpam-6638	288	12	.	.	PUNCT
ejpam-6638	289	1	•	•	NUM
ejpam-6638	289	2	denoising	denoising	NOUN
ejpam-6638	289	3	performance	performance	NOUN
ejpam-6638	289	4	(	(	PUNCT
ejpam-6638	289	5	first	first	ADJ
ejpam-6638	289	6	row	row	NOUN
ejpam-6638	289	7	):	):	PUNCT
ejpam-6638	289	8	the	the	DET
ejpam-6638	289	9	top	top	ADJ
ejpam-6638	289	10	row	row	NOUN
ejpam-6638	289	11	of	of	ADP
ejpam-6638	289	12	figure	figure	NOUN
ejpam-6638	289	13	9	9	NUM
ejpam-6638	289	14	illustrates	illustrate	VERB
ejpam-6638	289	15	the	the	DET
ejpam-6638	289	16	perform	perform	NOUN
ejpam-6638	289	17	.	.	PUNCT
ejpam-6638	290	1	israr	israr	NOUN
ejpam-6638	291	1	et	et	PROPN
ejpam-6638	291	2	al	al	PROPN
ejpam-6638	291	3	.	.	PUNCT
ejpam-6638	291	4	/	/	SYM
ejpam-6638	291	5	eur	eur	PROPN
ejpam-6638	291	6	.	.	PUNCT
ejpam-6638	292	1	j.	j.	PROPN
ejpam-6638	292	2	pure	pure	PROPN
ejpam-6638	292	3	appl	appl	PROPN
ejpam-6638	292	4	.	.	PROPN
ejpam-6638	292	5	math	math	PROPN
ejpam-6638	292	6	,	,	PUNCT
ejpam-6638	292	7	18	18	NUM
ejpam-6638	292	8	(	(	PUNCT
ejpam-6638	292	9	4	4	NUM
ejpam-6638	292	10	)	)	PUNCT
ejpam-6638	292	11	(	(	PUNCT
ejpam-6638	292	12	2025	2025	NUM
ejpam-6638	292	13	)	)	PUNCT
ejpam-6638	292	14	,	,	PUNCT
ejpam-6638	292	15	6638	6638	NUM
ejpam-6638	292	16	16	16	NUM
ejpam-6638	292	17	of	of	ADP
ejpam-6638	292	18	22	22	NUM
ejpam-6638	292	19	mance	mance	NOUN
ejpam-6638	292	20	of	of	ADP
ejpam-6638	292	21	each	each	DET
ejpam-6638	292	22	optimizer	optimizer	NOUN
ejpam-6638	292	23	on	on	ADP
ejpam-6638	292	24	denoising	denoise	VERB
ejpam-6638	292	25	tasks	task	NOUN
ejpam-6638	292	26	across	across	ADP
ejpam-6638	292	27	four	four	NUM
ejpam-6638	292	28	test	test	NOUN
ejpam-6638	292	29	images	image	NOUN
ejpam-6638	292	30	.	.	PUNCT
ejpam-6638	293	1	diffgrad	diffgrad	PROPN
ejpam-6638	293	2	consistently	consistently	ADV
ejpam-6638	293	3	delivers	deliver	VERB
ejpam-6638	293	4	superior	superior	ADJ
ejpam-6638	293	5	results	result	NOUN
ejpam-6638	293	6	,	,	PUNCT
ejpam-6638	293	7	achieving	achieve	VERB
ejpam-6638	293	8	psnr	psnr	NOUN
ejpam-6638	293	9	values	value	NOUN
ejpam-6638	293	10	around	around	ADP
ejpam-6638	293	11	50	50	NUM
ejpam-6638	293	12	and	and	CCONJ
ejpam-6638	293	13	ssim	ssim	NOUN
ejpam-6638	293	14	scores	score	NOUN
ejpam-6638	293	15	(	(	PUNCT
ejpam-6638	293	16	scaled	scale	VERB
ejpam-6638	293	17	by	by	ADP
ejpam-6638	293	18	100	100	NUM
ejpam-6638	293	19	)	)	PUNCT
ejpam-6638	293	20	near	near	ADP
ejpam-6638	293	21	95	95	NUM
ejpam-6638	293	22	,	,	PUNCT
ejpam-6638	293	23	with	with	ADP
ejpam-6638	293	24	moderate	moderate	ADJ
ejpam-6638	293	25	update	update	NOUN
ejpam-6638	293	26	costs	cost	NOUN
ejpam-6638	293	27	.	.	PUNCT
ejpam-6638	294	1	the	the	DET
ejpam-6638	294	2	3d	3d	PROPN
ejpam-6638	294	3	plots	plot	NOUN
ejpam-6638	294	4	further	far	ADV
ejpam-6638	294	5	highlight	highlight	VERB
ejpam-6638	294	6	apollo	apollo	PROPN
ejpam-6638	294	7	’s	’s	PART
ejpam-6638	294	8	computational	computational	ADJ
ejpam-6638	294	9	efficiency	efficiency	NOUN
ejpam-6638	294	10	,	,	PUNCT
ejpam-6638	294	11	demonstrating	demonstrate	VERB
ejpam-6638	294	12	the	the	DET
ejpam-6638	294	13	lowest	low	ADJ
ejpam-6638	294	14	update	update	NOUN
ejpam-6638	294	15	costs	cost	NOUN
ejpam-6638	294	16	,	,	PUNCT
ejpam-6638	294	17	though	though	SCONJ
ejpam-6638	294	18	it	it	PRON
ejpam-6638	294	19	lags	lag	VERB
ejpam-6638	294	20	behind	behind	ADP
ejpam-6638	294	21	diffgrad	diffgrad	NOUN
ejpam-6638	294	22	by	by	ADP
ejpam-6638	294	23	approximately	approximately	ADV
ejpam-6638	294	24	5−	5−	NUM
ejpam-6638	294	25	8	8	NUM
ejpam-6638	294	26	%	%	NOUN
ejpam-6638	294	27	in	in	ADP
ejpam-6638	294	28	psnr	psnr	NOUN
ejpam-6638	294	29	and	and	CCONJ
ejpam-6638	294	30	3−	3−	NUM
ejpam-6638	294	31	5	5	NUM
ejpam-6638	294	32	%	%	NOUN
ejpam-6638	294	33	in	in	ADP
ejpam-6638	294	34	ssim	ssim	NOUN
ejpam-6638	294	35	.	.	PUNCT
ejpam-6638	295	1	adam	adam	PROPN
ejpam-6638	295	2	occupies	occupy	VERB
ejpam-6638	295	3	a	a	DET
ejpam-6638	295	4	middle	middle	ADJ
ejpam-6638	295	5	ground	ground	NOUN
ejpam-6638	295	6	,	,	PUNCT
ejpam-6638	295	7	balancing	balance	VERB
ejpam-6638	295	8	performance	performance	NOUN
ejpam-6638	295	9	and	and	CCONJ
ejpam-6638	295	10	efficiency	efficiency	NOUN
ejpam-6638	295	11	,	,	PUNCT
ejpam-6638	295	12	but	but	CCONJ
ejpam-6638	295	13	exhibits	exhibit	VERB
ejpam-6638	295	14	higher	high	ADJ
ejpam-6638	295	15	variability	variability	NOUN
ejpam-6638	295	16	across	across	ADP
ejpam-6638	295	17	images	image	NOUN
ejpam-6638	295	18	,	,	PUNCT
ejpam-6638	295	19	particularly	particularly	ADV
ejpam-6638	295	20	in	in	ADP
ejpam-6638	295	21	subplots	subplot	NOUN
ejpam-6638	295	22	2	2	NUM
ejpam-6638	295	23	and	and	CCONJ
ejpam-6638	295	24	4	4	NUM
ejpam-6638	295	25	.	.	PUNCT
ejpam-6638	296	1	these	these	DET
ejpam-6638	296	2	findings	finding	NOUN
ejpam-6638	296	3	suggest	suggest	VERB
ejpam-6638	296	4	that	that	SCONJ
ejpam-6638	296	5	while	while	SCONJ
ejpam-6638	296	6	all	all	DET
ejpam-6638	296	7	optimizers	optimizer	NOUN
ejpam-6638	296	8	effectively	effectively	ADV
ejpam-6638	296	9	support	support	VERB
ejpam-6638	296	10	image	image	NOUN
ejpam-6638	296	11	denoising	denoise	VERB
ejpam-6638	296	12	within	within	ADP
ejpam-6638	296	13	our	our	PRON
ejpam-6638	296	14	architecture	architecture	NOUN
ejpam-6638	296	15	,	,	PUNCT
ejpam-6638	296	16	diffgrad	diffgrad	PROPN
ejpam-6638	296	17	offers	offer	VERB
ejpam-6638	296	18	the	the	DET
ejpam-6638	296	19	best	good	ADJ
ejpam-6638	296	20	overall	overall	ADJ
ejpam-6638	296	21	tradeoff	tradeoff	NOUN
ejpam-6638	296	22	between	between	ADP
ejpam-6638	296	23	reconstruction	reconstruction	NOUN
ejpam-6638	296	24	quality	quality	NOUN
ejpam-6638	296	25	and	and	CCONJ
ejpam-6638	296	26	computational	computational	ADJ
ejpam-6638	296	27	cost	cost	NOUN
ejpam-6638	296	28	.	.	PUNCT
ejpam-6638	297	1	(	(	PUNCT
ejpam-6638	297	2	a	a	X
ejpam-6638	297	3	)	)	PUNCT
ejpam-6638	297	4	(	(	PUNCT
ejpam-6638	297	5	b	b	X
ejpam-6638	297	6	)	)	PUNCT
ejpam-6638	297	7	(	(	PUNCT
ejpam-6638	297	8	c	c	X
ejpam-6638	297	9	)	)	PUNCT
ejpam-6638	297	10	(	(	PUNCT
ejpam-6638	297	11	d	d	X
ejpam-6638	297	12	)	)	PUNCT
ejpam-6638	297	13	(	(	PUNCT
ejpam-6638	297	14	e	e	NOUN
ejpam-6638	297	15	)	)	PUNCT
ejpam-6638	297	16	(	(	PUNCT
ejpam-6638	297	17	f	f	X
ejpam-6638	297	18	)	)	PUNCT
ejpam-6638	297	19	(	(	PUNCT
ejpam-6638	297	20	g	g	NOUN
ejpam-6638	297	21	)	)	PUNCT
ejpam-6638	297	22	(	(	PUNCT
ejpam-6638	297	23	h	h	NOUN
ejpam-6638	297	24	)	)	PUNCT
ejpam-6638	297	25	figure	figure	NOUN
ejpam-6638	297	26	9	9	NUM
ejpam-6638	297	27	:	:	PUNCT
ejpam-6638	297	28	3d	3d	NUM
ejpam-6638	297	29	plots	plot	NOUN
ejpam-6638	297	30	showing	show	VERB
ejpam-6638	297	31	the	the	DET
ejpam-6638	297	32	mean	mean	ADJ
ejpam-6638	297	33	psnr	psnr	NOUN
ejpam-6638	297	34	,	,	PUNCT
ejpam-6638	297	35	ssim	ssim	NOUN
ejpam-6638	297	36	,	,	PUNCT
ejpam-6638	297	37	and	and	CCONJ
ejpam-6638	297	38	update	update	VERB
ejpam-6638	297	39	cost	cost	NOUN
ejpam-6638	297	40	across	across	ADP
ejpam-6638	297	41	all	all	DET
ejpam-6638	297	42	epochs	epoch	NOUN
ejpam-6638	297	43	for	for	ADP
ejpam-6638	297	44	each	each	DET
ejpam-6638	297	45	optimizer	optimizer	NOUN
ejpam-6638	297	46	.	.	PUNCT
ejpam-6638	298	1	the	the	DET
ejpam-6638	298	2	three	three	NUM
ejpam-6638	298	3	axes	axis	NOUN
ejpam-6638	298	4	represent	represent	VERB
ejpam-6638	298	5	optimizers	optimizer	NOUN
ejpam-6638	298	6	,	,	PUNCT
ejpam-6638	298	7	evaluation	evaluation	NOUN
ejpam-6638	298	8	metrics	metric	NOUN
ejpam-6638	298	9	(	(	PUNCT
ejpam-6638	298	10	psnr	psnr	NOUN
ejpam-6638	298	11	,	,	PUNCT
ejpam-6638	298	12	ssim	ssim	NOUN
ejpam-6638	298	13	,	,	PUNCT
ejpam-6638	298	14	update	update	NOUN
ejpam-6638	298	15	cost	cost	NOUN
ejpam-6638	298	16	)	)	PUNCT
ejpam-6638	298	17	,	,	PUNCT
ejpam-6638	298	18	and	and	CCONJ
ejpam-6638	298	19	their	their	PRON
ejpam-6638	298	20	corresponding	corresponding	ADJ
ejpam-6638	298	21	mean	mean	ADJ
ejpam-6638	298	22	values	value	NOUN
ejpam-6638	298	23	over	over	ADP
ejpam-6638	298	24	3000	3000	NUM
ejpam-6638	298	25	epochs	epoch	NOUN
ejpam-6638	298	26	.	.	PUNCT
ejpam-6638	299	1	first	first	ADJ
ejpam-6638	299	2	row	row	NOUN
ejpam-6638	299	3	for	for	ADP
ejpam-6638	299	4	denoising	denoising	NOUN
ejpam-6638	299	5	and	and	CCONJ
ejpam-6638	299	6	second	second	ADJ
ejpam-6638	299	7	row	row	NOUN
ejpam-6638	299	8	for	for	ADP
ejpam-6638	299	9	inpainting	inpainte	VERB
ejpam-6638	299	10	.	.	PUNCT
ejpam-6638	300	1	4.3.2	4.3.2	X
ejpam-6638	300	2	.	.	PUNCT
ejpam-6638	301	1	inpainting	inpainte	VERB
ejpam-6638	301	2	the	the	DET
ejpam-6638	301	3	ablation	ablation	NOUN
ejpam-6638	301	4	study	study	NOUN
ejpam-6638	301	5	outlined	outline	VERB
ejpam-6638	301	6	in	in	ADP
ejpam-6638	301	7	table	table	NOUN
ejpam-6638	301	8	4	4	NUM
ejpam-6638	301	9	assesses	assess	VERB
ejpam-6638	301	10	the	the	DET
ejpam-6638	301	11	effectiveness	effectiveness	NOUN
ejpam-6638	301	12	of	of	ADP
ejpam-6638	301	13	the	the	DET
ejpam-6638	301	14	proposed	propose	VERB
ejpam-6638	301	15	dip	dip	NOUN
ejpam-6638	301	16	-	-	PUNCT
ejpam-6638	301	17	svd	svd	PROPN
ejpam-6638	301	18	technique	technique	NOUN
ejpam-6638	301	19	for	for	ADP
ejpam-6638	301	20	image	image	NOUN
ejpam-6638	301	21	inpainting	inpainte	VERB
ejpam-6638	301	22	by	by	ADP
ejpam-6638	301	23	utilizing	utilize	VERB
ejpam-6638	301	24	three	three	NUM
ejpam-6638	301	25	distinct	distinct	ADJ
ejpam-6638	301	26	optimizers	optimizer	NOUN
ejpam-6638	301	27	:	:	PUNCT
ejpam-6638	301	28	adam	adam	PROPN
ejpam-6638	301	29	,	,	PUNCT
ejpam-6638	301	30	apollo	apollo	PROPN
ejpam-6638	301	31	,	,	PUNCT
ejpam-6638	301	32	and	and	CCONJ
ejpam-6638	301	33	diffgrad	diffgrad	PROPN
ejpam-6638	301	34	.	.	PUNCT
ejpam-6638	302	1	two	two	NUM
ejpam-6638	302	2	common	common	ADJ
ejpam-6638	302	3	evaluation	evaluation	NOUN
ejpam-6638	302	4	metrics	metric	NOUN
ejpam-6638	302	5	—	—	PUNCT
ejpam-6638	302	6	peak	peak	NOUN
ejpam-6638	302	7	signal	signal	NOUN
ejpam-6638	302	8	-	-	PUNCT
ejpam-6638	302	9	to	to	ADP
ejpam-6638	302	10	-	-	PUNCT
ejpam-6638	302	11	noise	noise	NOUN
ejpam-6638	302	12	ratio	ratio	NOUN
ejpam-6638	302	13	(	(	PUNCT
ejpam-6638	302	14	psnr	psnr	NOUN
ejpam-6638	302	15	)	)	PUNCT
ejpam-6638	302	16	and	and	CCONJ
ejpam-6638	302	17	structural	structural	ADJ
ejpam-6638	302	18	similarity	similarity	NOUN
ejpam-6638	302	19	index	index	NOUN
ejpam-6638	302	20	measure	measure	NOUN
ejpam-6638	302	21	(	(	PUNCT
ejpam-6638	302	22	ssim)—are	ssim)—are	AUX
ejpam-6638	302	23	employed	employ	VERB
ejpam-6638	302	24	to	to	PART
ejpam-6638	302	25	evaluate	evaluate	VERB
ejpam-6638	302	26	the	the	DET
ejpam-6638	302	27	quality	quality	NOUN
ejpam-6638	302	28	of	of	ADP
ejpam-6638	302	29	the	the	DET
ejpam-6638	302	30	restored	restore	VERB
ejpam-6638	302	31	images	image	NOUN
ejpam-6638	302	32	.	.	PUNCT
ejpam-6638	303	1	the	the	DET
ejpam-6638	303	2	findings	finding	NOUN
ejpam-6638	303	3	reveal	reveal	VERB
ejpam-6638	303	4	that	that	SCONJ
ejpam-6638	303	5	the	the	DET
ejpam-6638	303	6	selection	selection	NOUN
ejpam-6638	303	7	of	of	ADP
ejpam-6638	303	8	optimizer	optimizer	NOUN
ejpam-6638	303	9	has	have	VERB
ejpam-6638	303	10	a	a	DET
ejpam-6638	303	11	considerable	considerable	ADJ
ejpam-6638	303	12	impact	impact	NOUN
ejpam-6638	303	13	on	on	ADP
ejpam-6638	303	14	the	the	DET
ejpam-6638	303	15	quality	quality	NOUN
ejpam-6638	303	16	of	of	ADP
ejpam-6638	303	17	inpainting	inpainte	VERB
ejpam-6638	303	18	.	.	PUNCT
ejpam-6638	304	1	remarkably	remarkably	ADV
ejpam-6638	304	2	,	,	PUNCT
ejpam-6638	304	3	both	both	CCONJ
ejpam-6638	304	4	the	the	DET
ejpam-6638	304	5	adam	adam	PROPN
ejpam-6638	304	6	and	and	CCONJ
ejpam-6638	304	7	diff	diff	PROPN
ejpam-6638	304	8	-	-	PUNCT
ejpam-6638	304	9	grad	grad	NOUN
ejpam-6638	304	10	optimizers	optimizer	NOUN
ejpam-6638	304	11	consistently	consistently	ADV
ejpam-6638	304	12	surpass	surpass	VERB
ejpam-6638	304	13	apollo	apollo	PROPN
ejpam-6638	304	14	across	across	ADP
ejpam-6638	304	15	all	all	DET
ejpam-6638	304	16	four	four	NUM
ejpam-6638	304	17	images	image	NOUN
ejpam-6638	304	18	,	,	PUNCT
ejpam-6638	304	19	showing	show	VERB
ejpam-6638	304	20	superior	superior	ADJ
ejpam-6638	304	21	psnr	psnr	NOUN
ejpam-6638	304	22	and	and	CCONJ
ejpam-6638	304	23	ssim	ssim	NOUN
ejpam-6638	304	24	values	value	NOUN
ejpam-6638	304	25	.	.	PUNCT
ejpam-6638	305	1	in	in	ADP
ejpam-6638	305	2	particular	particular	ADJ
ejpam-6638	305	3	,	,	PUNCT
ejpam-6638	305	4	diff	diff	NOUN
ejpam-6638	305	5	-	-	PUNCT
ejpam-6638	305	6	grad	grad	NOUN
ejpam-6638	305	7	achieves	achieve	VERB
ejpam-6638	305	8	an	an	DET
ejpam-6638	305	9	admirable	admirable	ADJ
ejpam-6638	305	10	equilibrium	equilibrium	NOUN
ejpam-6638	305	11	between	between	ADP
ejpam-6638	305	12	reconstruction	reconstruction	NOUN
ejpam-6638	305	13	quality	quality	NOUN
ejpam-6638	305	14	and	and	CCONJ
ejpam-6638	305	15	structural	structural	ADJ
ejpam-6638	305	16	integrity	integrity	NOUN
ejpam-6638	305	17	,	,	PUNCT
ejpam-6638	305	18	reaching	reach	VERB
ejpam-6638	305	19	psnr	psnr	NOUN
ejpam-6638	305	20	values	value	NOUN
ejpam-6638	305	21	as	as	ADV
ejpam-6638	305	22	high	high	ADJ
ejpam-6638	305	23	as	as	ADP
ejpam-6638	305	24	35.88	35.88	NUM
ejpam-6638	305	25	and	and	CCONJ
ejpam-6638	305	26	ssim	ssim	NOUN
ejpam-6638	305	27	scores	score	NOUN
ejpam-6638	305	28	that	that	PRON
ejpam-6638	305	29	are	be	AUX
ejpam-6638	305	30	remarkably	remarkably	ADV
ejpam-6638	305	31	close	close	ADJ
ejpam-6638	305	32	to	to	ADP
ejpam-6638	305	33	0.99	0.99	NUM
ejpam-6638	305	34	.	.	PUNCT
ejpam-6638	306	1	these	these	DET
ejpam-6638	306	2	results	result	NOUN
ejpam-6638	306	3	demonstrate	demonstrate	VERB
ejpam-6638	306	4	that	that	SCONJ
ejpam-6638	306	5	diff	diff	NOUN
ejpam-6638	306	6	-	-	PUNCT
ejpam-6638	306	7	grad	grad	NOUN
ejpam-6638	306	8	acts	act	NOUN
ejpam-6638	306	9	as	as	ADP
ejpam-6638	306	10	an	an	DET
ejpam-6638	306	11	exceptionally	exceptionally	ADV
ejpam-6638	306	12	effective	effective	ADJ
ejpam-6638	306	13	optimizer	optimizer	NOUN
ejpam-6638	306	14	for	for	ADP
ejpam-6638	306	15	training	train	VERB
ejpam-6638	306	16	the	the	DET
ejpam-6638	306	17	dip	dip	NOUN
ejpam-6638	306	18	-	-	PUNCT
ejpam-6638	306	19	svd	svd	PROPN
ejpam-6638	306	20	model	model	NOUN
ejpam-6638	306	21	,	,	PUNCT
ejpam-6638	306	22	enabling	enable	VERB
ejpam-6638	306	23	it	it	PRON
ejpam-6638	306	24	to	to	PART
ejpam-6638	306	25	restore	restore	VERB
ejpam-6638	306	26	missing	missing	ADJ
ejpam-6638	306	27	regions	region	NOUN
ejpam-6638	306	28	in	in	ADP
ejpam-6638	306	29	images	image	NOUN
ejpam-6638	306	30	with	with	ADP
ejpam-6638	306	31	outstanding	outstanding	ADJ
ejpam-6638	306	32	clarity	clarity	NOUN
ejpam-6638	306	33	and	and	CCONJ
ejpam-6638	306	34	structural	structural	ADJ
ejpam-6638	306	35	accuracy	accuracy	NOUN
ejpam-6638	306	36	.	.	PUNCT
ejpam-6638	307	1	the	the	DET
ejpam-6638	307	2	relatively	relatively	ADV
ejpam-6638	307	3	lower	low	ADJ
ejpam-6638	307	4	performance	performance	NOUN
ejpam-6638	307	5	of	of	ADP
ejpam-6638	307	6	apollo	apollo	PROPN
ejpam-6638	307	7	underscores	underscore	VERB
ejpam-6638	307	8	the	the	DET
ejpam-6638	307	9	importance	importance	NOUN
ejpam-6638	307	10	of	of	ADP
ejpam-6638	307	11	selecting	select	VERB
ejpam-6638	307	12	the	the	DET
ejpam-6638	307	13	appropriate	appropriate	ADJ
ejpam-6638	307	14	optimizer	optimizer	NOUN
ejpam-6638	307	15	within	within	ADP
ejpam-6638	307	16	deep	deep	ADJ
ejpam-6638	307	17	image	image	NOUN
ejpam-6638	307	18	prior	prior	ADV
ejpam-6638	307	19	-	-	PUNCT
ejpam-6638	307	20	based	base	VERB
ejpam-6638	307	21	systems	system	NOUN
ejpam-6638	307	22	.	.	PUNCT
ejpam-6638	308	1	this	this	DET
ejpam-6638	308	2	study	study	NOUN
ejpam-6638	308	3	emphasizes	emphasize	VERB
ejpam-6638	308	4	the	the	DET
ejpam-6638	308	5	critical	critical	ADJ
ejpam-6638	308	6	role	role	NOUN
ejpam-6638	308	7	that	that	PRON
ejpam-6638	308	8	optimizers	optimizer	NOUN
ejpam-6638	308	9	play	play	VERB
ejpam-6638	308	10	in	in	ADP
ejpam-6638	308	11	enhancing	enhance	VERB
ejpam-6638	308	12	inpainting	inpainte	VERB
ejpam-6638	308	13	outcomes	outcome	NOUN
ejpam-6638	308	14	in	in	ADP
ejpam-6638	308	15	neural	neural	ADJ
ejpam-6638	308	16	network	network	NOUN
ejpam-6638	308	17	-	-	PUNCT
ejpam-6638	308	18	based	base	VERB
ejpam-6638	308	19	image	image	NOUN
ejpam-6638	308	20	restoration	restoration	NOUN
ejpam-6638	308	21	efforts	effort	NOUN
ejpam-6638	308	22	.	.	PUNCT
ejpam-6638	309	1	m.	m.	NOUN
ejpam-6638	309	2	israr	israr	PROPN
ejpam-6638	309	3	et	et	PROPN
ejpam-6638	309	4	al	al	PROPN
ejpam-6638	309	5	.	.	PUNCT
ejpam-6638	309	6	/	/	SYM
ejpam-6638	309	7	eur	eur	PROPN
ejpam-6638	309	8	.	.	PUNCT
ejpam-6638	310	1	j.	j.	PROPN
ejpam-6638	310	2	pure	pure	PROPN
ejpam-6638	310	3	appl	appl	PROPN
ejpam-6638	310	4	.	.	PROPN
ejpam-6638	310	5	math	math	PROPN
ejpam-6638	310	6	,	,	PUNCT
ejpam-6638	310	7	18	18	NUM
ejpam-6638	310	8	(	(	PUNCT
ejpam-6638	310	9	4	4	NUM
ejpam-6638	310	10	)	)	PUNCT
ejpam-6638	310	11	(	(	PUNCT
ejpam-6638	310	12	2025	2025	NUM
ejpam-6638	310	13	)	)	PUNCT
ejpam-6638	310	14	,	,	PUNCT
ejpam-6638	310	15	6638	6638	NUM
ejpam-6638	310	16	17	17	NUM
ejpam-6638	310	17	of	of	ADP
ejpam-6638	310	18	22	22	NUM
ejpam-6638	310	19	optimizer	optimizer	NOUN
ejpam-6638	310	20	/	/	SYM
ejpam-6638	310	21	image	image	NOUN
ejpam-6638	310	22	1	1	NUM
ejpam-6638	310	23	2	2	NUM
ejpam-6638	310	24	3	3	NUM
ejpam-6638	310	25	4	4	NUM
ejpam-6638	310	26	psnr	psnr	NOUN
ejpam-6638	310	27	(	(	PUNCT
ejpam-6638	310	28	db	db	PROPN
ejpam-6638	310	29	)	)	PUNCT
ejpam-6638	310	30	adam	adam	PROPN
ejpam-6638	310	31	32.55	32.55	NUM
ejpam-6638	310	32	38.65	38.65	NUM
ejpam-6638	310	33	30.87	30.87	NUM
ejpam-6638	310	34	37.32	37.32	NUM
ejpam-6638	310	35	apollo	apollo	PROPN
ejpam-6638	310	36	25.51	25.51	NUM
ejpam-6638	310	37	24.59	24.59	NUM
ejpam-6638	310	38	22.59	22.59	NUM
ejpam-6638	310	39	23.62	23.62	NUM
ejpam-6638	310	40	diff	diff	NOUN
ejpam-6638	310	41	-	-	PUNCT
ejpam-6638	310	42	grad	grad	NOUN
ejpam-6638	310	43	32.04	32.04	NUM
ejpam-6638	310	44	35.88	35.88	NUM
ejpam-6638	310	45	30.99	30.99	NUM
ejpam-6638	310	46	34.47	34.47	NUM
ejpam-6638	310	47	ssim	ssim	NOUN
ejpam-6638	310	48	adam	adam	PROPN
ejpam-6638	310	49	0.98	0.98	NUM
ejpam-6638	310	50	0.98	0.98	NUM
ejpam-6638	310	51	0.59	0.59	NUM
ejpam-6638	310	52	0.99	0.99	NUM
ejpam-6638	310	53	apollo	apollo	PROPN
ejpam-6638	310	54	0.91	0.91	NUM
ejpam-6638	310	55	0.81	0.81	NUM
ejpam-6638	310	56	0.79	0.79	NUM
ejpam-6638	310	57	0.88	0.88	NUM
ejpam-6638	310	58	diff	diff	NOUN
ejpam-6638	310	59	-	-	PUNCT
ejpam-6638	310	60	grad	grad	NOUN
ejpam-6638	310	61	0.97	0.97	NUM
ejpam-6638	310	62	0.96	0.96	NUM
ejpam-6638	310	63	0.94	0.94	NUM
ejpam-6638	310	64	0.99	0.99	NUM
ejpam-6638	310	65	table	table	NOUN
ejpam-6638	310	66	4	4	NUM
ejpam-6638	310	67	:	:	PUNCT
ejpam-6638	310	68	ablation	ablation	NOUN
ejpam-6638	310	69	study	study	NOUN
ejpam-6638	310	70	:	:	PUNCT
ejpam-6638	310	71	inpainting	inpainte	VERB
ejpam-6638	310	72	performance	performance	NOUN
ejpam-6638	310	73	(	(	PUNCT
ejpam-6638	310	74	psnr	psnr	NOUN
ejpam-6638	310	75	and	and	CCONJ
ejpam-6638	310	76	ssim	ssim	NOUN
ejpam-6638	310	77	)	)	PUNCT
ejpam-6638	310	78	of	of	ADP
ejpam-6638	310	79	dip	dip	NOUN
ejpam-6638	310	80	-	-	PUNCT
ejpam-6638	310	81	svd	svd	PROPN
ejpam-6638	310	82	using	use	VERB
ejpam-6638	310	83	different	different	ADJ
ejpam-6638	310	84	optimizers	optimizer	NOUN
ejpam-6638	310	85	across	across	ADP
ejpam-6638	310	86	four	four	NUM
ejpam-6638	310	87	sample	sample	NOUN
ejpam-6638	310	88	images	image	NOUN
ejpam-6638	310	89	.	.	PUNCT
ejpam-6638	311	1	•	•	NOUN
ejpam-6638	311	2	training	training	NOUN
ejpam-6638	311	3	loss	loss	NOUN
ejpam-6638	311	4	analysis	analysis	NOUN
ejpam-6638	311	5	:	:	PUNCT
ejpam-6638	311	6	figure	figure	NOUN
ejpam-6638	311	7	10	10	NUM
ejpam-6638	311	8	’s	’s	PART
ejpam-6638	311	9	top	top	NOUN
ejpam-6638	311	10	row	row	NOUN
ejpam-6638	311	11	reveals	reveal	VERB
ejpam-6638	311	12	diffgrad	diffgrad	VERB
ejpam-6638	311	13	’s	’s	PART
ejpam-6638	311	14	superior	superior	ADJ
ejpam-6638	311	15	convergence	convergence	NOUN
ejpam-6638	311	16	in	in	ADP
ejpam-6638	311	17	inpainting	inpainte	VERB
ejpam-6638	311	18	tasks	task	NOUN
ejpam-6638	311	19	,	,	PUNCT
ejpam-6638	311	20	reaching	reach	VERB
ejpam-6638	311	21	stable	stable	ADJ
ejpam-6638	311	22	loss	loss	NOUN
ejpam-6638	311	23	values	value	NOUN
ejpam-6638	311	24	approximately	approximately	ADV
ejpam-6638	311	25	15	15	NUM
ejpam-6638	311	26	%	%	NOUN
ejpam-6638	311	27	lower	low	ADJ
ejpam-6638	311	28	than	than	ADP
ejpam-6638	311	29	adam	adam	PROPN
ejpam-6638	311	30	and	and	CCONJ
ejpam-6638	311	31	apollo	apollo	PROPN
ejpam-6638	311	32	by	by	ADP
ejpam-6638	311	33	epoch	epoch	PROPN
ejpam-6638	311	34	2000	2000	NUM
ejpam-6638	311	35	.	.	PUNCT
ejpam-6638	312	1	while	while	SCONJ
ejpam-6638	312	2	all	all	DET
ejpam-6638	312	3	optimizers	optimizer	NOUN
ejpam-6638	312	4	show	show	VERB
ejpam-6638	312	5	similar	similar	ADJ
ejpam-6638	312	6	initial	initial	ADJ
ejpam-6638	312	7	descent	descent	NOUN
ejpam-6638	312	8	,	,	PUNCT
ejpam-6638	312	9	diffgrad	diffgrad	PROPN
ejpam-6638	312	10	maintains	maintain	VERB
ejpam-6638	312	11	more	more	ADV
ejpam-6638	312	12	consistent	consistent	ADJ
ejpam-6638	312	13	optimization	optimization	NOUN
ejpam-6638	312	14	trajectories	trajectory	NOUN
ejpam-6638	312	15	in	in	ADP
ejpam-6638	312	16	later	later	ADJ
ejpam-6638	312	17	epochs	epoch	NOUN
ejpam-6638	312	18	,	,	PUNCT
ejpam-6638	312	19	particularly	particularly	ADV
ejpam-6638	312	20	visible	visible	ADJ
ejpam-6638	312	21	in	in	ADP
ejpam-6638	312	22	the	the	DET
ejpam-6638	312	23	library	library	NOUN
ejpam-6638	312	24	subplot	subplot	NOUN
ejpam-6638	312	25	.	.	PUNCT
ejpam-6638	313	1	our	our	PRON
ejpam-6638	313	2	dip	dip	NOUN
ejpam-6638	313	3	-	-	PUNCT
ejpam-6638	313	4	svd	svd	PROPN
ejpam-6638	313	5	architecture	architecture	NOUN
ejpam-6638	313	6	proves	prove	VERB
ejpam-6638	313	7	robust	robust	ADJ
ejpam-6638	313	8	across	across	ADP
ejpam-6638	313	9	optimizers	optimizer	NOUN
ejpam-6638	313	10	,	,	PUNCT
ejpam-6638	313	11	with	with	ADP
ejpam-6638	313	12	all	all	DET
ejpam-6638	313	13	methods	method	NOUN
ejpam-6638	313	14	converging	converge	VERB
ejpam-6638	313	15	below	below	ADP
ejpam-6638	313	16	0.05	0.05	NUM
ejpam-6638	313	17	loss	loss	NOUN
ejpam-6638	313	18	.	.	PUNCT
ejpam-6638	314	1	•	•	NUM
ejpam-6638	314	2	psnr	psnr	NOUN
ejpam-6638	314	3	performance	performance	NOUN
ejpam-6638	314	4	:	:	PUNCT
ejpam-6638	314	5	the	the	DET
ejpam-6638	314	6	middle	middle	PROPN
ejpam-6638	314	7	row	row	NOUN
ejpam-6638	314	8	in	in	ADP
ejpam-6638	314	9	figure	figure	NOUN
ejpam-6638	314	10	10	10	NUM
ejpam-6638	314	11	’s	’s	PART
ejpam-6638	314	12	demonstrates	demonstrate	VERB
ejpam-6638	314	13	diffgrad	diffgrad	PROPN
ejpam-6638	314	14	’s	’s	PART
ejpam-6638	314	15	advantage	advantage	NOUN
ejpam-6638	314	16	in	in	ADP
ejpam-6638	314	17	reconstruction	reconstruction	NOUN
ejpam-6638	314	18	quality	quality	NOUN
ejpam-6638	314	19	,	,	PUNCT
ejpam-6638	314	20	consistently	consistently	ADV
ejpam-6638	314	21	achieving	achieve	VERB
ejpam-6638	314	22	1.5	1.5	NUM
ejpam-6638	314	23	-	-	PUNCT
ejpam-6638	314	24	2db	2db	NUM
ejpam-6638	314	25	higher	high	ADJ
ejpam-6638	314	26	psnr	psnr	NOUN
ejpam-6638	314	27	than	than	ADP
ejpam-6638	314	28	competitors	competitor	NOUN
ejpam-6638	314	29	while	while	SCONJ
ejpam-6638	314	30	maintaining	maintain	VERB
ejpam-6638	314	31	dip	dip	NOUN
ejpam-6638	314	32	-	-	PUNCT
ejpam-6638	314	33	svd	svd	PROPN
ejpam-6638	314	34	’s	’s	PART
ejpam-6638	314	35	optimizer	optimizer	NOUN
ejpam-6638	314	36	-	-	PUNCT
ejpam-6638	314	37	agnostic	agnostic	ADJ
ejpam-6638	314	38	stability	stability	NOUN
ejpam-6638	314	39	.	.	PUNCT
ejpam-6638	315	1	the	the	DET
ejpam-6638	315	2	jet	jet	NOUN
ejpam-6638	315	3	image	image	NOUN
ejpam-6638	315	4	subplot	subplot	NOUN
ejpam-6638	315	5	shows	show	VERB
ejpam-6638	315	6	particularly	particularly	ADV
ejpam-6638	315	7	notable	notable	ADJ
ejpam-6638	315	8	gains	gain	NOUN
ejpam-6638	315	9	,	,	PUNCT
ejpam-6638	315	10	where	where	SCONJ
ejpam-6638	315	11	diffgrad	diffgrad	PROPN
ejpam-6638	315	12	surpasses	surpass	VERB
ejpam-6638	315	13	32db	32db	NUM
ejpam-6638	315	14	psnr	psnr	NOUN
ejpam-6638	315	15	while	while	SCONJ
ejpam-6638	315	16	others	other	NOUN
ejpam-6638	315	17	plateau	plateau	VERB
ejpam-6638	315	18	near	near	ADP
ejpam-6638	315	19	30db	30db	NOUN
ejpam-6638	315	20	.	.	PUNCT
ejpam-6638	316	1	all	all	DET
ejpam-6638	316	2	optimizers	optimizer	NOUN
ejpam-6638	316	3	achieve	achieve	VERB
ejpam-6638	316	4	professional	professional	ADJ
ejpam-6638	316	5	-	-	PUNCT
ejpam-6638	316	6	grade	grade	NOUN
ejpam-6638	316	7	results	result	NOUN
ejpam-6638	316	8	(	(	PUNCT
ejpam-6638	316	9	>	>	X
ejpam-6638	316	10	28db	28db	ADJ
ejpam-6638	316	11	psnr	psnr	NOUN
ejpam-6638	316	12	)	)	PUNCT
ejpam-6638	316	13	on	on	ADP
ejpam-6638	316	14	to	to	ADP
ejpam-6638	316	15	dip	dip	NOUN
ejpam-6638	316	16	-	-	PUNCT
ejpam-6638	316	17	svd	svd	PROPN
ejpam-6638	316	18	’s	’s	PART
ejpam-6638	316	19	svd	svd	PROPN
ejpam-6638	316	20	pooling	pool	VERB
ejpam-6638	316	21	layer	layer	NOUN
ejpam-6638	316	22	.	.	PUNCT
ejpam-6638	317	1	•	•	NUM
ejpam-6638	317	2	ssim	ssim	NOUN
ejpam-6638	317	3	evaluation	evaluation	NOUN
ejpam-6638	317	4	:	:	PUNCT
ejpam-6638	317	5	the	the	DET
ejpam-6638	317	6	bottom	bottom	NOUN
ejpam-6638	317	7	row	row	NOUN
ejpam-6638	317	8	in	in	ADP
ejpam-6638	317	9	figure	figure	NOUN
ejpam-6638	317	10	10	10	NUM
ejpam-6638	317	11	’s	’s	PART
ejpam-6638	317	12	confirms	confirm	VERB
ejpam-6638	317	13	diffgrad	diffgrad	PROPN
ejpam-6638	317	14	’s	’s	PART
ejpam-6638	317	15	superior	superior	ADJ
ejpam-6638	317	16	perceptual	perceptual	ADJ
ejpam-6638	317	17	results	result	NOUN
ejpam-6638	317	18	(	(	PUNCT
ejpam-6638	317	19	0.02−	0.02−	NUM
ejpam-6638	317	20	0.03	0.03	NUM
ejpam-6638	317	21	higher	high	ADJ
ejpam-6638	317	22	ssim	ssim	NOUN
ejpam-6638	317	23	)	)	PUNCT
ejpam-6638	317	24	while	while	SCONJ
ejpam-6638	317	25	showing	show	VERB
ejpam-6638	317	26	dip	dip	NOUN
ejpam-6638	317	27	-	-	PUNCT
ejpam-6638	317	28	svd	svd	PROPN
ejpam-6638	317	29	’s	’s	PART
ejpam-6638	317	30	consistent	consistent	ADJ
ejpam-6638	317	31	structural	structural	ADJ
ejpam-6638	317	32	preservation	preservation	NOUN
ejpam-6638	317	33	across	across	ADP
ejpam-6638	317	34	optimizers	optimizer	NOUN
ejpam-6638	317	35	.	.	PUNCT
ejpam-6638	318	1	all	all	DET
ejpam-6638	318	2	methods	method	NOUN
ejpam-6638	318	3	exhibit	exhibit	VERB
ejpam-6638	318	4	rapid	rapid	ADJ
ejpam-6638	318	5	early	early	ADJ
ejpam-6638	318	6	improvement	improvement	NOUN
ejpam-6638	318	7	followed	follow	VERB
ejpam-6638	318	8	by	by	ADP
ejpam-6638	318	9	gradual	gradual	ADJ
ejpam-6638	318	10	refinement	refinement	NOUN
ejpam-6638	318	11	,	,	PUNCT
ejpam-6638	318	12	achieving	achieve	VERB
ejpam-6638	318	13	ssim	ssim	NOUN
ejpam-6638	318	14	>	>	X
ejpam-6638	318	15	0.90	0.90	NUM
ejpam-6638	318	16	in	in	ADP
ejpam-6638	318	17	all	all	DET
ejpam-6638	318	18	cases	case	NOUN
ejpam-6638	318	19	.	.	PUNCT
ejpam-6638	319	1	the	the	DET
ejpam-6638	319	2	architecture	architecture	NOUN
ejpam-6638	319	3	ensures	ensure	VERB
ejpam-6638	319	4	final	final	ADJ
ejpam-6638	319	5	ssim	ssim	NOUN
ejpam-6638	319	6	values	value	NOUN
ejpam-6638	319	7	cluster	cluster	VERB
ejpam-6638	319	8	tightly	tightly	ADV
ejpam-6638	319	9	(	(	PUNCT
ejpam-6638	319	10	>	>	X
ejpam-6638	319	11	0.92for	0.92for	ADP
ejpam-6638	319	12	diffgrad	diffgrad	PROPN
ejpam-6638	319	13	)	)	PUNCT
ejpam-6638	319	14	,	,	PUNCT
ejpam-6638	319	15	demonstrating	demonstrate	VERB
ejpam-6638	319	16	effective	effective	ADJ
ejpam-6638	319	17	texture	texture	NOUN
ejpam-6638	319	18	and	and	CCONJ
ejpam-6638	319	19	structure	structure	NOUN
ejpam-6638	319	20	preservation	preservation	NOUN
ejpam-6638	319	21	.	.	PUNCT
ejpam-6638	320	1	•	•	NUM
ejpam-6638	320	2	inpainting	inpainte	VERB
ejpam-6638	320	3	results	result	NOUN
ejpam-6638	320	4	(	(	PUNCT
ejpam-6638	320	5	second	second	ADJ
ejpam-6638	320	6	row	row	NOUN
ejpam-6638	320	7	):	):	PUNCT
ejpam-6638	320	8	the	the	DET
ejpam-6638	320	9	bottom	bottom	NOUN
ejpam-6638	320	10	row	row	NOUN
ejpam-6638	320	11	of	of	ADP
ejpam-6638	320	12	figure	figure	NOUN
ejpam-6638	320	13	9	9	NUM
ejpam-6638	320	14	presents	present	NOUN
ejpam-6638	320	15	3d	3d	NOUN
ejpam-6638	320	16	plots	plot	NOUN
ejpam-6638	320	17	of	of	ADP
ejpam-6638	320	18	inpainting	inpainte	VERB
ejpam-6638	320	19	performance	performance	NOUN
ejpam-6638	320	20	.	.	PUNCT
ejpam-6638	321	1	diffgrad	diffgrad	PROPN
ejpam-6638	321	2	once	once	ADV
ejpam-6638	321	3	again	again	ADV
ejpam-6638	321	4	leads	lead	VERB
ejpam-6638	321	5	,	,	PUNCT
ejpam-6638	321	6	achieving	achieve	VERB
ejpam-6638	321	7	psnr	psnr	NOUN
ejpam-6638	321	8	values	value	NOUN
ejpam-6638	321	9	above	above	ADP
ejpam-6638	321	10	48	48	NUM
ejpam-6638	321	11	and	and	CCONJ
ejpam-6638	321	12	ssim	ssim	NOUN
ejpam-6638	321	13	scores	score	NOUN
ejpam-6638	321	14	(	(	PUNCT
ejpam-6638	321	15	scaled	scale	VERB
ejpam-6638	321	16	by	by	ADP
ejpam-6638	321	17	100	100	NUM
ejpam-6638	321	18	)	)	PUNCT
ejpam-6638	321	19	exceeding	exceed	VERB
ejpam-6638	321	20	92	92	NUM
ejpam-6638	321	21	.	.	PUNCT
ejpam-6638	322	1	apollo	apollo	PROPN
ejpam-6638	322	2	remains	remain	VERB
ejpam-6638	322	3	the	the	DET
ejpam-6638	322	4	most	most	ADV
ejpam-6638	322	5	efficient	efficient	ADJ
ejpam-6638	322	6	,	,	PUNCT
ejpam-6638	322	7	with	with	ADP
ejpam-6638	322	8	update	update	NOUN
ejpam-6638	322	9	costs	cost	NOUN
ejpam-6638	322	10	approximately	approximately	ADV
ejpam-6638	322	11	30	30	NUM
ejpam-6638	322	12	%	%	NOUN
ejpam-6638	322	13	lower	low	ADJ
ejpam-6638	322	14	than	than	ADP
ejpam-6638	322	15	its	its	PRON
ejpam-6638	322	16	counterparts	counterpart	NOUN
ejpam-6638	322	17	.	.	PUNCT
ejpam-6638	323	1	in	in	ADP
ejpam-6638	323	2	particular	particular	ADJ
ejpam-6638	323	3	,	,	PUNCT
ejpam-6638	323	4	the	the	DET
ejpam-6638	323	5	jet	jet	NOUN
ejpam-6638	323	6	and	and	CCONJ
ejpam-6638	323	7	vase	vase	NOUN
ejpam-6638	323	8	subplots	subplot	NOUN
ejpam-6638	323	9	emphasize	emphasize	VERB
ejpam-6638	323	10	diffgrad	diffgrad	PROPN
ejpam-6638	323	11	’s	’s	PART
ejpam-6638	323	12	stability	stability	NOUN
ejpam-6638	323	13	,	,	PUNCT
ejpam-6638	323	14	showing	show	VERB
ejpam-6638	323	15	minimal	minimal	ADJ
ejpam-6638	323	16	variation	variation	NOUN
ejpam-6638	323	17	in	in	ADP
ejpam-6638	323	18	metrics	metric	NOUN
ejpam-6638	323	19	compared	compare	VERB
ejpam-6638	323	20	to	to	ADP
ejpam-6638	323	21	the	the	DET
ejpam-6638	323	22	more	more	ADV
ejpam-6638	323	23	fluctuating	fluctuate	VERB
ejpam-6638	323	24	performance	performance	NOUN
ejpam-6638	323	25	of	of	ADP
ejpam-6638	323	26	adam	adam	PROPN
ejpam-6638	323	27	.	.	PUNCT
ejpam-6638	324	1	despite	despite	SCONJ
ejpam-6638	324	2	differences	difference	NOUN
ejpam-6638	324	3	in	in	ADP
ejpam-6638	324	4	optimizer	optimizer	NOUN
ejpam-6638	324	5	behavior	behavior	NOUN
ejpam-6638	324	6	,	,	PUNCT
ejpam-6638	324	7	all	all	DET
ejpam-6638	324	8	three	three	NUM
ejpam-6638	324	9	achieve	achieve	VERB
ejpam-6638	324	10	professional	professional	ADJ
ejpam-6638	324	11	-	-	PUNCT
ejpam-6638	324	12	grade	grade	NOUN
ejpam-6638	324	13	inpainting	inpainting	NOUN
ejpam-6638	324	14	(	(	PUNCT
ejpam-6638	324	15	psnr	psnr	NOUN
ejpam-6638	324	16	>	>	X
ejpam-6638	324	17	45	45	NUM
ejpam-6638	324	18	,	,	PUNCT
ejpam-6638	324	19	ssim	ssim	NOUN
ejpam-6638	324	20	×100	×100	PROPN
ejpam-6638	324	21	>	>	X
ejpam-6638	324	22	90	90	NUM
ejpam-6638	324	23	)	)	PUNCT
ejpam-6638	324	24	,	,	PUNCT
ejpam-6638	324	25	underscoring	underscore	VERB
ejpam-6638	324	26	the	the	DET
ejpam-6638	324	27	robustness	robustness	NOUN
ejpam-6638	324	28	of	of	ADP
ejpam-6638	324	29	our	our	PRON
ejpam-6638	324	30	architecture	architecture	NOUN
ejpam-6638	324	31	across	across	ADP
ejpam-6638	324	32	both	both	DET
ejpam-6638	324	33	optimizer	optimizer	NOUN
ejpam-6638	324	34	choices	choice	NOUN
ejpam-6638	324	35	and	and	CCONJ
ejpam-6638	324	36	varying	vary	VERB
ejpam-6638	324	37	image	image	NOUN
ejpam-6638	324	38	complexities	complexity	NOUN
ejpam-6638	324	39	.	.	PUNCT
ejpam-6638	325	1	training	training	NOUN
ejpam-6638	325	2	details	detail	NOUN
ejpam-6638	325	3	we	we	PRON
ejpam-6638	325	4	have	have	AUX
ejpam-6638	325	5	updated	update	VERB
ejpam-6638	325	6	the	the	DET
ejpam-6638	325	7	manuscript	manuscript	NOUN
ejpam-6638	325	8	to	to	PART
ejpam-6638	325	9	include	include	VERB
ejpam-6638	325	10	detailed	detailed	ADJ
ejpam-6638	325	11	training	training	NOUN
ejpam-6638	325	12	information	information	NOUN
ejpam-6638	325	13	.	.	PUNCT
ejpam-6638	326	1	specifically	specifically	ADV
ejpam-6638	326	2	,	,	PUNCT
ejpam-6638	326	3	for	for	ADP
ejpam-6638	326	4	all	all	DET
ejpam-6638	326	5	experiments	experiment	NOUN
ejpam-6638	326	6	:	:	PUNCT
ejpam-6638	326	7	•	•	NUM
ejpam-6638	326	8	training	training	NOUN
ejpam-6638	326	9	time	time	NOUN
ejpam-6638	326	10	per	per	ADP
ejpam-6638	326	11	image	image	NOUN
ejpam-6638	326	12	:	:	PUNCT
ejpam-6638	326	13	each	each	DET
ejpam-6638	326	14	image	image	NOUN
ejpam-6638	326	15	was	be	AUX
ejpam-6638	326	16	trained	train	VERB
ejpam-6638	326	17	for	for	ADP
ejpam-6638	326	18	approximately	approximately	ADV
ejpam-6638	326	19	xx	xx	NUM
ejpam-6638	326	20	minutes	minute	NOUN
ejpam-6638	326	21	(	(	PUNCT
ejpam-6638	326	22	depending	depend	VERB
ejpam-6638	326	23	on	on	ADP
ejpam-6638	326	24	resolution	resolution	NOUN
ejpam-6638	326	25	and	and	CCONJ
ejpam-6638	326	26	complexity	complexity	NOUN
ejpam-6638	326	27	)	)	PUNCT
ejpam-6638	326	28	.	.	PUNCT
ejpam-6638	327	1	m.	m.	NOUN
ejpam-6638	327	2	israr	israr	PROPN
ejpam-6638	327	3	et	et	PROPN
ejpam-6638	327	4	al	al	PROPN
ejpam-6638	327	5	.	.	PUNCT
ejpam-6638	327	6	/	/	SYM
ejpam-6638	327	7	eur	eur	PROPN
ejpam-6638	327	8	.	.	PUNCT
ejpam-6638	328	1	j.	j.	PROPN
ejpam-6638	328	2	pure	pure	PROPN
ejpam-6638	328	3	appl	appl	PROPN
ejpam-6638	328	4	.	.	PROPN
ejpam-6638	328	5	math	math	PROPN
ejpam-6638	328	6	,	,	PUNCT
ejpam-6638	328	7	18	18	NUM
ejpam-6638	328	8	(	(	PUNCT
ejpam-6638	328	9	4	4	NUM
ejpam-6638	328	10	)	)	PUNCT
ejpam-6638	328	11	(	(	PUNCT
ejpam-6638	328	12	2025	2025	NUM
ejpam-6638	328	13	)	)	PUNCT
ejpam-6638	328	14	,	,	PUNCT
ejpam-6638	328	15	6638	6638	NUM
ejpam-6638	328	16	18	18	NUM
ejpam-6638	328	17	of	of	ADP
ejpam-6638	328	18	22	22	NUM
ejpam-6638	328	19	(	(	PUNCT
ejpam-6638	328	20	a	a	NOUN
ejpam-6638	328	21	)	)	PUNCT
ejpam-6638	328	22	(	(	PUNCT
ejpam-6638	328	23	b	b	X
ejpam-6638	328	24	)	)	PUNCT
ejpam-6638	328	25	(	(	PUNCT
ejpam-6638	328	26	c	c	X
ejpam-6638	328	27	)	)	PUNCT
ejpam-6638	328	28	(	(	PUNCT
ejpam-6638	328	29	d	d	X
ejpam-6638	328	30	)	)	PUNCT
ejpam-6638	328	31	(	(	PUNCT
ejpam-6638	328	32	e	e	NOUN
ejpam-6638	328	33	)	)	PUNCT
ejpam-6638	328	34	(	(	PUNCT
ejpam-6638	328	35	f	f	X
ejpam-6638	328	36	)	)	PUNCT
ejpam-6638	328	37	(	(	PUNCT
ejpam-6638	328	38	g	g	NOUN
ejpam-6638	328	39	)	)	PUNCT
ejpam-6638	328	40	(	(	PUNCT
ejpam-6638	328	41	h	h	NOUN
ejpam-6638	328	42	)	)	PUNCT
ejpam-6638	328	43	(	(	PUNCT
ejpam-6638	328	44	i	i	NOUN
ejpam-6638	328	45	)	)	PUNCT
ejpam-6638	328	46	figure	figure	VERB
ejpam-6638	328	47	10	10	NUM
ejpam-6638	328	48	:	:	PUNCT
ejpam-6638	328	49	training	training	NOUN
ejpam-6638	328	50	loss	loss	NOUN
ejpam-6638	328	51	,	,	PUNCT
ejpam-6638	328	52	psnr	psnr	NOUN
ejpam-6638	328	53	,	,	PUNCT
ejpam-6638	328	54	and	and	CCONJ
ejpam-6638	328	55	ssim	ssim	NOUN
ejpam-6638	328	56	across	across	ADP
ejpam-6638	328	57	3000	3000	NUM
ejpam-6638	328	58	epochs	epoch	NOUN
ejpam-6638	328	59	for	for	ADP
ejpam-6638	328	60	each	each	DET
ejpam-6638	328	61	optimizer	optimizer	NOUN
ejpam-6638	328	62	on	on	ADP
ejpam-6638	328	63	four	four	NUM
ejpam-6638	328	64	selected	select	VERB
ejpam-6638	328	65	images	image	NOUN
ejpam-6638	328	66	in	in	ADP
ejpam-6638	328	67	case	case	NOUN
ejpam-6638	328	68	of	of	ADP
ejpam-6638	328	69	inpainting	inpainte	VERB
ejpam-6638	328	70	.	.	PUNCT
ejpam-6638	329	1	the	the	DET
ejpam-6638	329	2	first	first	ADJ
ejpam-6638	329	3	row	row	NOUN
ejpam-6638	329	4	shows	show	VERB
ejpam-6638	329	5	the	the	DET
ejpam-6638	329	6	training	training	NOUN
ejpam-6638	329	7	loss	loss	NOUN
ejpam-6638	329	8	,	,	PUNCT
ejpam-6638	329	9	the	the	DET
ejpam-6638	329	10	second	second	ADJ
ejpam-6638	329	11	row	row	NOUN
ejpam-6638	329	12	displays	display	VERB
ejpam-6638	329	13	psnr	psnr	NOUN
ejpam-6638	329	14	,	,	PUNCT
ejpam-6638	329	15	and	and	CCONJ
ejpam-6638	329	16	the	the	DET
ejpam-6638	329	17	third	third	ADJ
ejpam-6638	329	18	row	row	NOUN
ejpam-6638	329	19	presents	present	NOUN
ejpam-6638	329	20	ssim	ssim	NOUN
ejpam-6638	329	21	.	.	PUNCT
ejpam-6638	330	1	•	•	NUM
ejpam-6638	330	2	hardware	hardware	NOUN
ejpam-6638	330	3	used	use	VERB
ejpam-6638	330	4	:	:	PUNCT
ejpam-6638	330	5	experiments	experiment	NOUN
ejpam-6638	330	6	were	be	AUX
ejpam-6638	330	7	conducted	conduct	VERB
ejpam-6638	330	8	on	on	ADP
ejpam-6638	330	9	an	an	DET
ejpam-6638	330	10	nvidia	nvidia	PROPN
ejpam-6638	330	11	rtx	rtx	PROPN
ejpam-6638	330	12	4090	4090	NUM
ejpam-6638	330	13	gpu	gpu	NOUN
ejpam-6638	330	14	with	with	ADP
ejpam-6638	330	15	24	24	NUM
ejpam-6638	330	16	gb	gb	NOUN
ejpam-6638	330	17	memory	memory	NOUN
ejpam-6638	330	18	.	.	PUNCT
ejpam-6638	331	1	cpu	cpu	PROPN
ejpam-6638	331	2	computations	computation	NOUN
ejpam-6638	331	3	were	be	AUX
ejpam-6638	331	4	performed	perform	VERB
ejpam-6638	331	5	on	on	ADP
ejpam-6638	331	6	an	an	DET
ejpam-6638	331	7	intel	intel	PROPN
ejpam-6638	331	8	xeon	xeon	PROPN
ejpam-6638	331	9	w-2295	w-2295	PROPN
ejpam-6638	331	10	processor	processor	NOUN
ejpam-6638	331	11	.	.	PUNCT
ejpam-6638	332	1	•	•	NUM
ejpam-6638	332	2	learning	learn	VERB
ejpam-6638	332	3	rate	rate	NOUN
ejpam-6638	332	4	schedules	schedule	NOUN
ejpam-6638	332	5	:	:	PUNCT
ejpam-6638	332	6	a	a	DET
ejpam-6638	332	7	fixed	fix	VERB
ejpam-6638	332	8	learning	learning	NOUN
ejpam-6638	332	9	rate	rate	NOUN
ejpam-6638	332	10	of	of	ADP
ejpam-6638	332	11	0.001	0.001	NUM
ejpam-6638	332	12	was	be	AUX
ejpam-6638	332	13	used	use	VERB
ejpam-6638	332	14	for	for	ADP
ejpam-6638	332	15	all	all	DET
ejpam-6638	332	16	optimizers	optimizer	NOUN
ejpam-6638	332	17	,	,	PUNCT
ejpam-6638	332	18	without	without	ADP
ejpam-6638	332	19	any	any	DET
ejpam-6638	332	20	decay	decay	NOUN
ejpam-6638	332	21	schedule	schedule	NOUN
ejpam-6638	332	22	.	.	PUNCT
ejpam-6638	333	1	early	early	ADJ
ejpam-6638	333	2	stopping	stopping	NOUN
ejpam-6638	333	3	was	be	AUX
ejpam-6638	333	4	not	not	PART
ejpam-6638	333	5	employed	employ	VERB
ejpam-6638	333	6	;	;	PUNCT
ejpam-6638	333	7	instead	instead	ADV
ejpam-6638	333	8	,	,	PUNCT
ejpam-6638	333	9	a	a	DET
ejpam-6638	333	10	fixed	fix	VERB
ejpam-6638	333	11	number	number	NOUN
ejpam-6638	333	12	of	of	ADP
ejpam-6638	333	13	epochs	epoch	NOUN
ejpam-6638	333	14	was	be	AUX
ejpam-6638	333	15	run	run	VERB
ejpam-6638	333	16	for	for	ADP
ejpam-6638	333	17	each	each	DET
ejpam-6638	333	18	training	training	NOUN
ejpam-6638	333	19	session	session	NOUN
ejpam-6638	333	20	.	.	PUNCT
ejpam-6638	334	1	5	5	X
ejpam-6638	334	2	.	.	X
ejpam-6638	334	3	conclusion	conclusion	NOUN
ejpam-6638	334	4	in	in	ADP
ejpam-6638	334	5	this	this	DET
ejpam-6638	334	6	research	research	NOUN
ejpam-6638	334	7	,	,	PUNCT
ejpam-6638	334	8	we	we	PRON
ejpam-6638	334	9	presented	present	VERB
ejpam-6638	334	10	an	an	DET
ejpam-6638	334	11	enhanced	enhanced	ADJ
ejpam-6638	334	12	deep	deep	ADJ
ejpam-6638	334	13	image	image	NOUN
ejpam-6638	334	14	prior	prior	ADV
ejpam-6638	334	15	(	(	PUNCT
ejpam-6638	334	16	dip	dip	NOUN
ejpam-6638	334	17	)	)	PUNCT
ejpam-6638	334	18	framework	framework	NOUN
ejpam-6638	334	19	tailored	tailor	VERB
ejpam-6638	334	20	for	for	ADP
ejpam-6638	334	21	image	image	NOUN
ejpam-6638	334	22	restoration	restoration	NOUN
ejpam-6638	334	23	tasks	task	NOUN
ejpam-6638	334	24	,	,	PUNCT
ejpam-6638	334	25	featuring	feature	VERB
ejpam-6638	334	26	notable	notable	ADJ
ejpam-6638	334	27	architectural	architectural	ADJ
ejpam-6638	334	28	improvements	improvement	NOUN
ejpam-6638	334	29	like	like	ADP
ejpam-6638	334	30	svd	svd	PROPN
ejpam-6638	334	31	pooling	pool	VERB
ejpam-6638	334	32	to	to	PART
ejpam-6638	334	33	preserve	preserve	VERB
ejpam-6638	334	34	overall	overall	ADJ
ejpam-6638	334	35	image	image	NOUN
ejpam-6638	334	36	structures	structure	NOUN
ejpam-6638	334	37	,	,	PUNCT
ejpam-6638	334	38	softplus	softplus	ADJ
ejpam-6638	334	39	activation	activation	NOUN
ejpam-6638	334	40	to	to	PART
ejpam-6638	334	41	ensure	ensure	VERB
ejpam-6638	334	42	stable	stable	ADJ
ejpam-6638	334	43	gradient	gradient	NOUN
ejpam-6638	334	44	flow	flow	NOUN
ejpam-6638	334	45	,	,	PUNCT
ejpam-6638	334	46	and	and	CCONJ
ejpam-6638	334	47	a	a	DET
ejpam-6638	334	48	multiscale	multiscale	ADJ
ejpam-6638	334	49	attention	attention	NOUN
ejpam-6638	334	50	mechanism	mechanism	NOUN
ejpam-6638	334	51	with	with	ADP
ejpam-6638	334	52	dilated	dilated	ADJ
ejpam-6638	334	53	convolutions	convolution	NOUN
ejpam-6638	334	54	to	to	PART
ejpam-6638	334	55	improve	improve	VERB
ejpam-6638	334	56	detail	detail	NOUN
ejpam-6638	334	57	reconstruction	reconstruction	NOUN
ejpam-6638	334	58	.	.	PUNCT
ejpam-6638	335	1	to	to	PART
ejpam-6638	335	2	further	far	ADV
ejpam-6638	335	3	enhance	enhance	VERB
ejpam-6638	335	4	convergence	convergence	NOUN
ejpam-6638	335	5	and	and	CCONJ
ejpam-6638	335	6	generalization	generalization	NOUN
ejpam-6638	335	7	,	,	PUNCT
ejpam-6638	335	8	we	we	PRON
ejpam-6638	335	9	evaluated	evaluate	VERB
ejpam-6638	335	10	the	the	DET
ejpam-6638	335	11	performance	performance	NOUN
ejpam-6638	335	12	of	of	ADP
ejpam-6638	335	13	three	three	NUM
ejpam-6638	335	14	optimizers	optimizer	NOUN
ejpam-6638	335	15	—	—	PUNCT
ejpam-6638	335	16	adam	adam	PROPN
ejpam-6638	335	17	,	,	PUNCT
ejpam-6638	335	18	apollo	apollo	PROPN
ejpam-6638	335	19	,	,	PUNCT
ejpam-6638	335	20	and	and	CCONJ
ejpam-6638	335	21	diffgrad	diffgrad	PROPN
ejpam-6638	335	22	—	—	PUNCT
ejpam-6638	335	23	within	within	ADP
ejpam-6638	335	24	our	our	PRON
ejpam-6638	335	25	dip	dip	NOUN
ejpam-6638	335	26	-	-	PUNCT
ejpam-6638	335	27	svd	svd	PROPN
ejpam-6638	335	28	framework	framework	NOUN
ejpam-6638	335	29	.	.	PUNCT
ejpam-6638	336	1	the	the	DET
ejpam-6638	336	2	experimental	experimental	ADJ
ejpam-6638	336	3	assessment	assessment	NOUN
ejpam-6638	336	4	,	,	PUNCT
ejpam-6638	336	5	carried	carry	VERB
ejpam-6638	336	6	out	out	ADP
ejpam-6638	336	7	on	on	ADP
ejpam-6638	336	8	both	both	PRON
ejpam-6638	336	9	denoising	denoise	VERB
ejpam-6638	336	10	and	and	CCONJ
ejpam-6638	336	11	inpainting	inpainte	VERB
ejpam-6638	336	12	tasks	task	NOUN
ejpam-6638	336	13	,	,	PUNCT
ejpam-6638	336	14	showed	show	VERB
ejpam-6638	336	15	that	that	SCONJ
ejpam-6638	336	16	our	our	PRON
ejpam-6638	336	17	model	model	NOUN
ejpam-6638	336	18	attains	attain	VERB
ejpam-6638	336	19	superior	superior	ADJ
ejpam-6638	336	20	restoration	restoration	NOUN
ejpam-6638	336	21	quality	quality	NOUN
ejpam-6638	336	22	without	without	ADP
ejpam-6638	336	23	the	the	DET
ejpam-6638	336	24	need	need	NOUN
ejpam-6638	336	25	for	for	ADP
ejpam-6638	336	26	extensive	extensive	ADJ
ejpam-6638	336	27	training	training	NOUN
ejpam-6638	336	28	data	datum	NOUN
ejpam-6638	336	29	.	.	PUNCT
ejpam-6638	337	1	significantly	significantly	ADV
ejpam-6638	337	2	,	,	PUNCT
ejpam-6638	337	3	diffgrad	diffgrad	PROPN
ejpam-6638	337	4	consistently	consistently	ADV
ejpam-6638	337	5	surpassed	surpass	VERB
ejpam-6638	337	6	both	both	CCONJ
ejpam-6638	337	7	adam	adam	PROPN
ejpam-6638	337	8	and	and	CCONJ
ejpam-6638	337	9	apollo	apollo	PROPN
ejpam-6638	337	10	regarding	regard	VERB
ejpam-6638	337	11	psnr	psnr	NOUN
ejpam-6638	337	12	and	and	CCONJ
ejpam-6638	337	13	ssim	ssim	NOUN
ejpam-6638	337	14	metrics	metric	NOUN
ejpam-6638	337	15	,	,	PUNCT
ejpam-6638	337	16	demonstrating	demonstrate	VERB
ejpam-6638	337	17	its	its	PRON
ejpam-6638	337	18	efficacy	efficacy	NOUN
ejpam-6638	337	19	in	in	ADP
ejpam-6638	337	20	enhancing	enhance	VERB
ejpam-6638	337	21	reconstruction	reconstruction	NOUN
ejpam-6638	337	22	quality	quality	NOUN
ejpam-6638	337	23	.	.	PUNCT
ejpam-6638	338	1	these	these	DET
ejpam-6638	338	2	findings	finding	NOUN
ejpam-6638	338	3	emphasize	emphasize	VERB
ejpam-6638	338	4	the	the	DET
ejpam-6638	338	5	advantages	advantage	NOUN
ejpam-6638	338	6	of	of	ADP
ejpam-6638	338	7	employing	employ	VERB
ejpam-6638	338	8	adaptive	adaptive	ADJ
ejpam-6638	338	9	optimization	optimization	NOUN
ejpam-6638	338	10	techniques	technique	NOUN
ejpam-6638	338	11	alongside	alongside	ADP
ejpam-6638	338	12	structural	structural	ADJ
ejpam-6638	338	13	inductive	inductive	ADJ
ejpam-6638	338	14	biases	bias	NOUN
ejpam-6638	338	15	for	for	ADP
ejpam-6638	338	16	unsupervised	unsupervised	ADJ
ejpam-6638	338	17	image	image	NOUN
ejpam-6638	338	18	restoration	restoration	NOUN
ejpam-6638	338	19	.	.	PUNCT
ejpam-6638	339	1	our	our	PRON
ejpam-6638	339	2	study	study	NOUN
ejpam-6638	339	3	includes	include	VERB
ejpam-6638	339	4	multiple	multiple	ADJ
ejpam-6638	339	5	visual	visual	ADJ
ejpam-6638	339	6	analyses	analysis	NOUN
ejpam-6638	339	7	to	to	PART
ejpam-6638	339	8	validate	validate	VERB
ejpam-6638	339	9	the	the	DET
ejpam-6638	339	10	effectiveness	effectiveness	NOUN
ejpam-6638	339	11	of	of	ADP
ejpam-6638	339	12	the	the	DET
ejpam-6638	339	13	proposed	propose	VERB
ejpam-6638	339	14	method	method	NOUN
ejpam-6638	339	15	.	.	PUNCT
ejpam-6638	340	1	we	we	PRON
ejpam-6638	340	2	present	present	VERB
ejpam-6638	340	3	the	the	DET
ejpam-6638	340	4	best	good	ADJ
ejpam-6638	340	5	runs	run	NOUN
ejpam-6638	340	6	based	base	VERB
ejpam-6638	340	7	on	on	ADP
ejpam-6638	340	8	learned	learn	VERB
ejpam-6638	340	9	weights	weight	NOUN
ejpam-6638	340	10	for	for	ADP
ejpam-6638	340	11	each	each	DET
ejpam-6638	340	12	image	image	NOUN
ejpam-6638	340	13	in	in	ADP
ejpam-6638	340	14	both	both	PRON
ejpam-6638	340	15	denoising	denoise	VERB
ejpam-6638	340	16	and	and	CCONJ
ejpam-6638	340	17	inpainting	inpainte	VERB
ejpam-6638	340	18	tasks	task	NOUN
ejpam-6638	340	19	,	,	PUNCT
ejpam-6638	340	20	along	along	ADP
ejpam-6638	340	21	with	with	ADP
ejpam-6638	340	22	training	training	NOUN
ejpam-6638	340	23	curves	curve	NOUN
ejpam-6638	340	24	showing	show	VERB
ejpam-6638	340	25	the	the	DET
ejpam-6638	340	26	evolution	evolution	NOUN
ejpam-6638	340	27	of	of	ADP
ejpam-6638	340	28	loss	loss	NOUN
ejpam-6638	340	29	,	,	PUNCT
ejpam-6638	340	30	psnr	psnr	NOUN
ejpam-6638	340	31	,	,	PUNCT
ejpam-6638	340	32	and	and	CCONJ
ejpam-6638	340	33	ssim	ssim	NOUN
ejpam-6638	340	34	m.	m.	NOUN
ejpam-6638	340	35	israr	israr	PROPN
ejpam-6638	340	36	et	et	PROPN
ejpam-6638	340	37	al	al	PROPN
ejpam-6638	340	38	.	.	PUNCT
ejpam-6638	340	39	/	/	SYM
ejpam-6638	340	40	eur	eur	PROPN
ejpam-6638	340	41	.	.	PUNCT
ejpam-6638	341	1	j.	j.	PROPN
ejpam-6638	341	2	pure	pure	PROPN
ejpam-6638	341	3	appl	appl	PROPN
ejpam-6638	341	4	.	.	PROPN
ejpam-6638	341	5	math	math	PROPN
ejpam-6638	341	6	,	,	PUNCT
ejpam-6638	341	7	18	18	NUM
ejpam-6638	341	8	(	(	PUNCT
ejpam-6638	341	9	4	4	NUM
ejpam-6638	341	10	)	)	PUNCT
ejpam-6638	341	11	(	(	PUNCT
ejpam-6638	341	12	2025	2025	NUM
ejpam-6638	341	13	)	)	PUNCT
ejpam-6638	341	14	,	,	PUNCT
ejpam-6638	341	15	6638	6638	NUM
ejpam-6638	341	16	19	19	NUM
ejpam-6638	341	17	of	of	ADP
ejpam-6638	341	18	22	22	NUM
ejpam-6638	341	19	for	for	ADP
ejpam-6638	341	20	each	each	DET
ejpam-6638	341	21	optimizer	optimizer	NOUN
ejpam-6638	341	22	.	.	PUNCT
ejpam-6638	342	1	additionally	additionally	ADV
ejpam-6638	342	2	,	,	PUNCT
ejpam-6638	342	3	3d	3d	NUM
ejpam-6638	342	4	plots	plot	NOUN
ejpam-6638	342	5	summarize	summarize	VERB
ejpam-6638	342	6	the	the	DET
ejpam-6638	342	7	mean	mean	ADJ
ejpam-6638	342	8	psnr	psnr	NOUN
ejpam-6638	342	9	,	,	PUNCT
ejpam-6638	342	10	ssim	ssim	NOUN
ejpam-6638	342	11	,	,	PUNCT
ejpam-6638	342	12	and	and	CCONJ
ejpam-6638	342	13	update	update	VERB
ejpam-6638	342	14	cost	cost	NOUN
ejpam-6638	342	15	across	across	ADP
ejpam-6638	342	16	all	all	DET
ejpam-6638	342	17	epochs	epoch	NOUN
ejpam-6638	342	18	,	,	PUNCT
ejpam-6638	342	19	highlighting	highlight	VERB
ejpam-6638	342	20	variations	variation	NOUN
ejpam-6638	342	21	in	in	ADP
ejpam-6638	342	22	convergence	convergence	NOUN
ejpam-6638	342	23	behavior	behavior	NOUN
ejpam-6638	342	24	and	and	CCONJ
ejpam-6638	342	25	computational	computational	ADJ
ejpam-6638	342	26	efficiency	efficiency	NOUN
ejpam-6638	342	27	.	.	PUNCT
ejpam-6638	343	1	future	future	ADJ
ejpam-6638	343	2	directions	direction	NOUN
ejpam-6638	343	3	include	include	VERB
ejpam-6638	343	4	unifying	unify	VERB
ejpam-6638	343	5	svd	svd	PROPN
ejpam-6638	343	6	-	-	PUNCT
ejpam-6638	343	7	dip	dip	NOUN
ejpam-6638	343	8	with	with	ADP
ejpam-6638	343	9	variational	variational	ADJ
ejpam-6638	343	10	methods	method	NOUN
ejpam-6638	343	11	(	(	PUNCT
ejpam-6638	343	12	tv	tv	NOUN
ejpam-6638	343	13	/	/	SYM
ejpam-6638	343	14	mean	mean	NOUN
ejpam-6638	343	15	curvature	curvature	NOUN
ejpam-6638	343	16	)	)	PUNCT
ejpam-6638	343	17	through	through	ADP
ejpam-6638	343	18	joint	joint	ADJ
ejpam-6638	343	19	optimization	optimization	NOUN
ejpam-6638	343	20	to	to	PART
ejpam-6638	343	21	strengthen	strengthen	VERB
ejpam-6638	343	22	geometric	geometric	ADJ
ejpam-6638	343	23	priors	prior	NOUN
ejpam-6638	343	24	in	in	ADP
ejpam-6638	343	25	unsupervised	unsupervised	ADJ
ejpam-6638	343	26	restoration	restoration	NOUN
ejpam-6638	343	27	.	.	PUNCT
ejpam-6638	344	1	for	for	ADP
ejpam-6638	344	2	largedata	largedata	ADJ
ejpam-6638	344	3	scenarios	scenario	NOUN
ejpam-6638	344	4	,	,	PUNCT
ejpam-6638	344	5	we	we	PRON
ejpam-6638	344	6	will	will	AUX
ejpam-6638	344	7	investigate	investigate	VERB
ejpam-6638	344	8	distillation	distillation	NOUN
ejpam-6638	344	9	techniques	technique	NOUN
ejpam-6638	344	10	to	to	PART
ejpam-6638	344	11	transfer	transfer	VERB
ejpam-6638	344	12	svd	svd	NOUN
ejpam-6638	344	13	pooling	pool	VERB
ejpam-6638	344	14	benefits	benefit	NOUN
ejpam-6638	344	15	to	to	ADP
ejpam-6638	344	16	conventional	conventional	ADJ
ejpam-6638	344	17	cnns	cnn	NOUN
ejpam-6638	344	18	while	while	SCONJ
ejpam-6638	344	19	preserving	preserve	VERB
ejpam-6638	344	20	dip	dip	NOUN
ejpam-6638	344	21	’s	’s	PART
ejpam-6638	344	22	data	datum	NOUN
ejpam-6638	344	23	efficiency	efficiency	NOUN
ejpam-6638	344	24	.	.	PUNCT
ejpam-6638	345	1	adaptive	adaptive	ADJ
ejpam-6638	345	2	rank	rank	NOUN
ejpam-6638	345	3	selection	selection	NOUN
ejpam-6638	345	4	during	during	ADP
ejpam-6638	345	5	svd	svd	PROPN
ejpam-6638	345	6	decomposition	decomposition	NOUN
ejpam-6638	345	7	could	could	AUX
ejpam-6638	345	8	dynamically	dynamically	ADV
ejpam-6638	345	9	balance	balance	VERB
ejpam-6638	345	10	texture	texture	ADJ
ejpam-6638	345	11	retention	retention	NOUN
ejpam-6638	345	12	and	and	CCONJ
ejpam-6638	345	13	noise	noise	NOUN
ejpam-6638	345	14	suppression	suppression	NOUN
ejpam-6638	345	15	.	.	PUNCT
ejpam-6638	346	1	extensions	extension	NOUN
ejpam-6638	346	2	to	to	ADP
ejpam-6638	346	3	coupled	couple	VERB
ejpam-6638	346	4	degradation	degradation	NOUN
ejpam-6638	346	5	models	model	NOUN
ejpam-6638	346	6	(	(	PUNCT
ejpam-6638	346	7	e.g.	e.g.	ADV
ejpam-6638	346	8	,	,	PUNCT
ejpam-6638	346	9	blur+noise	blur+noise	PROPN
ejpam-6638	346	10	)	)	PUNCT
ejpam-6638	346	11	may	may	AUX
ejpam-6638	346	12	further	far	ADV
ejpam-6638	346	13	validate	validate	VERB
ejpam-6638	346	14	the	the	DET
ejpam-6638	346	15	framework	framework	NOUN
ejpam-6638	346	16	’s	’s	PART
ejpam-6638	346	17	robustness	robustness	NOUN
ejpam-6638	346	18	for	for	ADP
ejpam-6638	346	19	real	real	ADJ
ejpam-6638	346	20	-	-	PUNCT
ejpam-6638	346	21	world	world	NOUN
ejpam-6638	346	22	low	low	ADJ
ejpam-6638	346	23	-	-	PUNCT
ejpam-6638	346	24	level	level	NOUN
ejpam-6638	346	25	restoration	restoration	NOUN
ejpam-6638	346	26	.	.	PUNCT
ejpam-6638	347	1	acknowledgements	acknowledgement	NOUN
ejpam-6638	347	2	the	the	DET
ejpam-6638	347	3	authors	author	NOUN
ejpam-6638	347	4	would	would	AUX
ejpam-6638	347	5	like	like	VERB
ejpam-6638	347	6	to	to	PART
ejpam-6638	347	7	thank	thank	VERB
ejpam-6638	347	8	the	the	DET
ejpam-6638	347	9	anonymous	anonymous	ADJ
ejpam-6638	347	10	reviewers	reviewer	NOUN
ejpam-6638	347	11	for	for	ADP
ejpam-6638	347	12	their	their	PRON
ejpam-6638	347	13	valuable	valuable	ADJ
ejpam-6638	347	14	suggestions	suggestion	NOUN
ejpam-6638	347	15	and	and	CCONJ
ejpam-6638	347	16	comments	comment	NOUN
ejpam-6638	347	17	,	,	PUNCT
ejpam-6638	347	18	which	which	PRON
ejpam-6638	347	19	helped	help	VERB
ejpam-6638	347	20	to	to	PART
ejpam-6638	347	21	improve	improve	VERB
ejpam-6638	347	22	the	the	DET
ejpam-6638	347	23	quality	quality	NOUN
ejpam-6638	347	24	of	of	ADP
ejpam-6638	347	25	this	this	DET
ejpam-6638	347	26	manuscript	manuscript	NOUN
ejpam-6638	347	27	.	.	PUNCT
ejpam-6638	348	1	references	reference	NOUN
ejpam-6638	348	2	[	[	X
ejpam-6638	348	3	1	1	NUM
ejpam-6638	348	4	]	]	X
ejpam-6638	348	5	l.i	l.i	PROPN
ejpam-6638	348	6	.	.	PROPN
ejpam-6638	348	7	rudin	rudin	PROPN
ejpam-6638	348	8	,	,	PUNCT
ejpam-6638	348	9	s.	s.	PROPN
ejpam-6638	348	10	osher	osher	PROPN
ejpam-6638	348	11	,	,	PUNCT
ejpam-6638	348	12	and	and	CCONJ
ejpam-6638	348	13	e.	e.	PROPN
ejpam-6638	348	14	fatemi	fatemi	PROPN
ejpam-6638	348	15	.	.	PUNCT
ejpam-6638	349	1	nonlinear	nonlinear	ADJ
ejpam-6638	349	2	total	total	ADJ
ejpam-6638	349	3	variation	variation	NOUN
ejpam-6638	349	4	based	base	VERB
ejpam-6638	349	5	noise	noise	NOUN
ejpam-6638	349	6	removal	removal	NOUN
ejpam-6638	349	7	algorithms	algorithm	NOUN
ejpam-6638	349	8	.	.	PUNCT
ejpam-6638	350	1	physica	physica	PROPN
ejpam-6638	350	2	d	d	PROPN
ejpam-6638	350	3	,	,	PUNCT
ejpam-6638	350	4	60:259–268	60:259–268	NUM
ejpam-6638	350	5	,	,	PUNCT
ejpam-6638	350	6	1992	1992	NUM
ejpam-6638	350	7	.	.	PUNCT
ejpam-6638	351	1	[	[	X
ejpam-6638	351	2	2	2	X
ejpam-6638	351	3	]	]	X
ejpam-6638	351	4	s.	s.	PROPN
ejpam-6638	351	5	roth	roth	PROPN
ejpam-6638	351	6	and	and	CCONJ
ejpam-6638	351	7	m.j	m.j	PROPN
ejpam-6638	351	8	.	.	PROPN
ejpam-6638	351	9	black	black	PROPN
ejpam-6638	351	10	.	.	PUNCT
ejpam-6638	352	1	fields	field	NOUN
ejpam-6638	352	2	of	of	ADP
ejpam-6638	352	3	experts	expert	NOUN
ejpam-6638	352	4	:	:	PUNCT
ejpam-6638	352	5	a	a	DET
ejpam-6638	352	6	framework	framework	NOUN
ejpam-6638	352	7	for	for	ADP
ejpam-6638	352	8	learning	learn	VERB
ejpam-6638	352	9	image	image	NOUN
ejpam-6638	352	10	priors	prior	NOUN
ejpam-6638	352	11	.	.	PUNCT
ejpam-6638	353	1	international	international	ADJ
ejpam-6638	353	2	journal	journal	PROPN
ejpam-6638	353	3	of	of	ADP
ejpam-6638	353	4	computer	computer	NOUN
ejpam-6638	353	5	vision	vision	NOUN
ejpam-6638	353	6	,	,	PUNCT
ejpam-6638	353	7	82:205–229	82:205–229	NUM
ejpam-6638	353	8	,	,	PUNCT
ejpam-6638	353	9	2009	2009	NUM
ejpam-6638	353	10	.	.	PUNCT
ejpam-6638	354	1	[	[	X
ejpam-6638	354	2	3	3	NUM
ejpam-6638	354	3	]	]	PUNCT
ejpam-6638	354	4	a.	a.	NOUN
ejpam-6638	354	5	buades	buades	PROPN
ejpam-6638	354	6	,	,	PUNCT
ejpam-6638	354	7	b.	b.	PROPN
ejpam-6638	354	8	coll	coll	PROPN
ejpam-6638	354	9	,	,	PUNCT
ejpam-6638	354	10	and	and	CCONJ
ejpam-6638	354	11	j.m	j.m	PROPN
ejpam-6638	354	12	.	.	PROPN
ejpam-6638	354	13	morel	morel	PROPN
ejpam-6638	354	14	.	.	PUNCT
ejpam-6638	355	1	nonlocal	nonlocal	ADJ
ejpam-6638	355	2	image	image	NOUN
ejpam-6638	355	3	and	and	CCONJ
ejpam-6638	355	4	movie	movie	NOUN
ejpam-6638	355	5	denoising	denoising	NOUN
ejpam-6638	355	6	.	.	PUNCT
ejpam-6638	356	1	international	international	ADJ
ejpam-6638	356	2	journal	journal	NOUN
ejpam-6638	356	3	of	of	ADP
ejpam-6638	356	4	computer	computer	NOUN
ejpam-6638	356	5	vision	vision	NOUN
ejpam-6638	356	6	,	,	PUNCT
ejpam-6638	356	7	76:123–139	76:123–139	NUM
ejpam-6638	356	8	,	,	PUNCT
ejpam-6638	356	9	2008	2008	NUM
ejpam-6638	356	10	.	.	PUNCT
ejpam-6638	357	1	[	[	X
ejpam-6638	357	2	4	4	X
ejpam-6638	357	3	]	]	X
ejpam-6638	357	4	j.	j.	PROPN
ejpam-6638	357	5	portilla	portilla	PROPN
ejpam-6638	357	6	,	,	PUNCT
ejpam-6638	357	7	v.	v.	PROPN
ejpam-6638	357	8	strela	strela	PROPN
ejpam-6638	357	9	,	,	PUNCT
ejpam-6638	357	10	m.j	m.j	PROPN
ejpam-6638	357	11	.	.	PROPN
ejpam-6638	357	12	wainwright	wainwright	PROPN
ejpam-6638	357	13	,	,	PUNCT
ejpam-6638	357	14	and	and	CCONJ
ejpam-6638	357	15	e.p	e.p	PROPN
ejpam-6638	357	16	.	.	PROPN
ejpam-6638	357	17	simoncelli	simoncelli	PROPN
ejpam-6638	357	18	.	.	PUNCT
ejpam-6638	358	1	image	image	NOUN
ejpam-6638	358	2	denoising	denoise	VERB
ejpam-6638	358	3	using	use	VERB
ejpam-6638	358	4	scale	scale	NOUN
ejpam-6638	358	5	mixtures	mixture	NOUN
ejpam-6638	358	6	of	of	ADP
ejpam-6638	358	7	gaussians	gaussian	NOUN
ejpam-6638	358	8	in	in	ADP
ejpam-6638	358	9	the	the	DET
ejpam-6638	358	10	wavelet	wavelet	NOUN
ejpam-6638	358	11	domain	domain	NOUN
ejpam-6638	358	12	.	.	PUNCT
ejpam-6638	359	1	ieee	ieee	NOUN
ejpam-6638	359	2	transactions	transaction	NOUN
ejpam-6638	359	3	on	on	ADP
ejpam-6638	359	4	image	image	NOUN
ejpam-6638	359	5	processing	processing	NOUN
ejpam-6638	359	6	,	,	PUNCT
ejpam-6638	359	7	12:1338–1351	12:1338–1351	NUM
ejpam-6638	359	8	,	,	PUNCT
ejpam-6638	359	9	2003	2003	NUM
ejpam-6638	359	10	.	.	PUNCT
ejpam-6638	360	1	[	[	X
ejpam-6638	360	2	5	5	NUM
ejpam-6638	360	3	]	]	PUNCT
ejpam-6638	360	4	c.	c.	PROPN
ejpam-6638	360	5	dong	dong	PROPN
ejpam-6638	360	6	,	,	PUNCT
ejpam-6638	360	7	c.c	c.c	PROPN
ejpam-6638	360	8	.	.	PROPN
ejpam-6638	360	9	loy	loy	PROPN
ejpam-6638	360	10	,	,	PUNCT
ejpam-6638	360	11	k.	k.	PROPN
ejpam-6638	361	1	he	he	PRON
ejpam-6638	361	2	,	,	PUNCT
ejpam-6638	361	3	and	and	CCONJ
ejpam-6638	361	4	x.	x.	NOUN
ejpam-6638	361	5	tang	tang	PROPN
ejpam-6638	361	6	.	.	PUNCT
ejpam-6638	362	1	image	image	NOUN
ejpam-6638	362	2	super	super	NOUN
ejpam-6638	362	3	-	-	NOUN
ejpam-6638	362	4	resolution	resolution	NOUN
ejpam-6638	362	5	using	use	VERB
ejpam-6638	362	6	deep	deep	ADJ
ejpam-6638	362	7	convolutional	convolutional	ADJ
ejpam-6638	362	8	networks	network	NOUN
ejpam-6638	362	9	.	.	PUNCT
ejpam-6638	363	1	ieee	ieee	NOUN
ejpam-6638	363	2	transactions	transaction	NOUN
ejpam-6638	363	3	on	on	ADP
ejpam-6638	363	4	pattern	pattern	NOUN
ejpam-6638	363	5	analysis	analysis	NOUN
ejpam-6638	363	6	and	and	CCONJ
ejpam-6638	363	7	machine	machine	NOUN
ejpam-6638	363	8	intelligence	intelligence	NOUN
ejpam-6638	363	9	,	,	PUNCT
ejpam-6638	363	10	38:295–307	38:295–307	PROPN
ejpam-6638	363	11	,	,	PUNCT
ejpam-6638	363	12	2016	2016	NUM
ejpam-6638	363	13	.	.	PUNCT
ejpam-6638	364	1	[	[	X
ejpam-6638	364	2	6	6	NUM
ejpam-6638	364	3	]	]	X
ejpam-6638	364	4	c.	c.	PROPN
ejpam-6638	364	5	ledig	ledig	PROPN
ejpam-6638	364	6	,	,	PUNCT
ejpam-6638	364	7	l.	l.	PROPN
ejpam-6638	364	8	theis	theis	PROPN
ejpam-6638	364	9	,	,	PUNCT
ejpam-6638	364	10	f.	f.	PROPN
ejpam-6638	364	11	huszár	huszár	PROPN
ejpam-6638	364	12	,	,	PUNCT
ejpam-6638	364	13	j.	j.	PROPN
ejpam-6638	364	14	caballero	caballero	PROPN
ejpam-6638	364	15	,	,	PUNCT
ejpam-6638	364	16	a.	a.	PROPN
ejpam-6638	364	17	cunningham	cunningham	PROPN
ejpam-6638	364	18	,	,	PUNCT
ejpam-6638	364	19	a.	a.	NOUN
ejpam-6638	364	20	acosta	acosta	PROPN
ejpam-6638	364	21	,	,	PUNCT
ejpam-6638	364	22	a.p	a.p	PROPN
ejpam-6638	364	23	.	.	PROPN
ejpam-6638	364	24	aitken	aitken	PROPN
ejpam-6638	364	25	,	,	PUNCT
ejpam-6638	364	26	a.	a.	NOUN
ejpam-6638	364	27	tejani	tejani	PROPN
ejpam-6638	364	28	,	,	PUNCT
ejpam-6638	364	29	j.	j.	PROPN
ejpam-6638	364	30	totz	totz	PROPN
ejpam-6638	364	31	,	,	PUNCT
ejpam-6638	364	32	z.	z.	PROPN
ejpam-6638	364	33	wang	wang	PROPN
ejpam-6638	364	34	,	,	PUNCT
ejpam-6638	364	35	and	and	CCONJ
ejpam-6638	364	36	w.	w.	PROPN
ejpam-6638	364	37	shi	shi	PROPN
ejpam-6638	364	38	.	.	PUNCT
ejpam-6638	365	1	photo	photo	NOUN
ejpam-6638	365	2	-	-	PUNCT
ejpam-6638	365	3	realistic	realistic	ADJ
ejpam-6638	365	4	single	single	ADJ
ejpam-6638	365	5	image	image	NOUN
ejpam-6638	365	6	super	super	NOUN
ejpam-6638	365	7	-	-	NOUN
ejpam-6638	365	8	resolution	resolution	NOUN
ejpam-6638	365	9	using	use	VERB
ejpam-6638	365	10	a	a	DET
ejpam-6638	365	11	generative	generative	ADJ
ejpam-6638	365	12	adversarial	adversarial	ADJ
ejpam-6638	365	13	network	network	NOUN
ejpam-6638	365	14	.	.	PUNCT
ejpam-6638	366	1	in	in	ADP
ejpam-6638	366	2	proceedings	proceeding	NOUN
ejpam-6638	366	3	of	of	ADP
ejpam-6638	366	4	the	the	DET
ejpam-6638	366	5	ieee	ieee	NOUN
ejpam-6638	366	6	conference	conference	NOUN
ejpam-6638	366	7	on	on	ADP
ejpam-6638	366	8	computer	computer	NOUN
ejpam-6638	366	9	vision	vision	NOUN
ejpam-6638	366	10	and	and	CCONJ
ejpam-6638	366	11	pattern	pattern	NOUN
ejpam-6638	366	12	recognition	recognition	NOUN
ejpam-6638	366	13	(	(	PUNCT
ejpam-6638	366	14	cvpr	cvpr	NOUN
ejpam-6638	366	15	)	)	PUNCT
ejpam-6638	366	16	,	,	PUNCT
ejpam-6638	366	17	pages	page	NOUN
ejpam-6638	366	18	4681–4690	4681–4690	NUM
ejpam-6638	366	19	,	,	PUNCT
ejpam-6638	366	20	2017	2017	NUM
ejpam-6638	366	21	.	.	PUNCT
ejpam-6638	367	1	[	[	X
ejpam-6638	367	2	7	7	X
ejpam-6638	367	3	]	]	PUNCT
ejpam-6638	367	4	k.	k.	PROPN
ejpam-6638	367	5	zhang	zhang	PROPN
ejpam-6638	367	6	,	,	PUNCT
ejpam-6638	367	7	w.	w.	PROPN
ejpam-6638	367	8	zuo	zuo	PROPN
ejpam-6638	367	9	,	,	PUNCT
ejpam-6638	367	10	y.	y.	PROPN
ejpam-6638	367	11	chen	chen	PROPN
ejpam-6638	367	12	,	,	PUNCT
ejpam-6638	367	13	d.	d.	PROPN
ejpam-6638	367	14	meng	meng	PROPN
ejpam-6638	367	15	,	,	PUNCT
ejpam-6638	367	16	and	and	CCONJ
ejpam-6638	367	17	l.	l.	PROPN
ejpam-6638	367	18	zhang	zhang	PROPN
ejpam-6638	367	19	.	.	PUNCT
ejpam-6638	368	1	beyond	beyond	ADP
ejpam-6638	368	2	a	a	DET
ejpam-6638	368	3	gaussian	gaussian	ADJ
ejpam-6638	368	4	denoiser	denoiser	NOUN
ejpam-6638	368	5	:	:	PUNCT
ejpam-6638	368	6	residual	residual	ADJ
ejpam-6638	368	7	learning	learning	NOUN
ejpam-6638	368	8	of	of	ADP
ejpam-6638	368	9	deep	deep	ADJ
ejpam-6638	368	10	cnn	cnn	PROPN
ejpam-6638	368	11	for	for	ADP
ejpam-6638	368	12	image	image	NOUN
ejpam-6638	368	13	denoising	denoise	VERB
ejpam-6638	368	14	.	.	PUNCT
ejpam-6638	369	1	ieee	ieee	NOUN
ejpam-6638	369	2	transactions	transaction	NOUN
ejpam-6638	369	3	on	on	ADP
ejpam-6638	369	4	image	image	NOUN
ejpam-6638	369	5	processing	processing	NOUN
ejpam-6638	369	6	,	,	PUNCT
ejpam-6638	369	7	26:3142	26:3142	NUM
ejpam-6638	369	8	–	–	PUNCT
ejpam-6638	369	9	3155	3155	NUM
ejpam-6638	369	10	,	,	PUNCT
ejpam-6638	369	11	2017	2017	NUM
ejpam-6638	369	12	.	.	PUNCT
ejpam-6638	370	1	[	[	X
ejpam-6638	370	2	8	8	NUM
ejpam-6638	370	3	]	]	X
ejpam-6638	370	4	d.	d.	PROPN
ejpam-6638	370	5	ulyanov	ulyanov	PROPN
ejpam-6638	370	6	,	,	PUNCT
ejpam-6638	370	7	a.	a.	NOUN
ejpam-6638	370	8	vedaldi	vedaldi	NOUN
ejpam-6638	370	9	,	,	PUNCT
ejpam-6638	370	10	and	and	CCONJ
ejpam-6638	370	11	v.	v.	ADP
ejpam-6638	370	12	lempitsky	lempitsky	NOUN
ejpam-6638	370	13	.	.	PUNCT
ejpam-6638	371	1	deep	deep	ADJ
ejpam-6638	371	2	image	image	NOUN
ejpam-6638	371	3	prior	prior	ADV
ejpam-6638	371	4	.	.	PUNCT
ejpam-6638	372	1	international	international	ADJ
ejpam-6638	372	2	journal	journal	NOUN
ejpam-6638	372	3	of	of	ADP
ejpam-6638	372	4	computer	computer	NOUN
ejpam-6638	372	5	vision	vision	NOUN
ejpam-6638	372	6	,	,	PUNCT
ejpam-6638	372	7	128:1867–1888	128:1867–1888	NUM
ejpam-6638	372	8	,	,	PUNCT
ejpam-6638	372	9	2020	2020	NUM
ejpam-6638	372	10	.	.	PUNCT
ejpam-6638	373	1	[	[	X
ejpam-6638	373	2	9	9	NUM
ejpam-6638	373	3	]	]	PUNCT
ejpam-6638	373	4	r.	r.	PROPN
ejpam-6638	373	5	heckel	heckel	PROPN
ejpam-6638	373	6	and	and	CCONJ
ejpam-6638	373	7	p.	p.	PROPN
ejpam-6638	373	8	hand	hand	NOUN
ejpam-6638	373	9	.	.	PUNCT
ejpam-6638	374	1	deep	deep	ADJ
ejpam-6638	374	2	decoder	decoder	NOUN
ejpam-6638	374	3	:	:	PUNCT
ejpam-6638	374	4	concise	concise	ADJ
ejpam-6638	374	5	image	image	NOUN
ejpam-6638	374	6	representations	representation	NOUN
ejpam-6638	374	7	from	from	ADP
ejpam-6638	374	8	untrained	untrained	ADJ
ejpam-6638	374	9	non	non	ADJ
ejpam-6638	374	10	-	-	ADJ
ejpam-6638	374	11	convolutional	convolutional	ADJ
ejpam-6638	374	12	networks	network	NOUN
ejpam-6638	374	13	.	.	PUNCT
ejpam-6638	375	1	international	international	ADJ
ejpam-6638	375	2	journal	journal	PROPN
ejpam-6638	375	3	of	of	ADP
ejpam-6638	375	4	computer	computer	NOUN
ejpam-6638	375	5	vision	vision	NOUN
ejpam-6638	375	6	,	,	PUNCT
ejpam-6638	375	7	129:2536–2555	129:2536–2555	NUM
ejpam-6638	375	8	,	,	PUNCT
ejpam-6638	375	9	2021	2021	NUM
ejpam-6638	375	10	.	.	PUNCT
ejpam-6638	376	1	[	[	X
ejpam-6638	376	2	10	10	NUM
ejpam-6638	376	3	]	]	PUNCT
ejpam-6638	376	4	z.	z.	PROPN
ejpam-6638	376	5	xie	xie	PROPN
ejpam-6638	376	6	,	,	PUNCT
ejpam-6638	376	7	l.	l.	PROPN
ejpam-6638	376	8	liu	liu	PROPN
ejpam-6638	376	9	,	,	PUNCT
ejpam-6638	376	10	z.	z.	PROPN
ejpam-6638	376	11	luo	luo	PROPN
ejpam-6638	376	12	,	,	PUNCT
ejpam-6638	376	13	and	and	CCONJ
ejpam-6638	376	14	j.	j.	PROPN
ejpam-6638	376	15	huang	huang	PROPN
ejpam-6638	376	16	.	.	PUNCT
ejpam-6638	376	17	image	image	NOUN
ejpam-6638	376	18	denoising	denoise	VERB
ejpam-6638	376	19	using	use	VERB
ejpam-6638	376	20	nonlocal	nonlocal	ADJ
ejpam-6638	376	21	regularized	regularize	VERB
ejpam-6638	376	22	deep	deep	ADJ
ejpam-6638	376	23	image	image	NOUN
ejpam-6638	376	24	prior	prior	ADV
ejpam-6638	376	25	.	.	PUNCT
ejpam-6638	377	1	symmetry	symmetry	NOUN
ejpam-6638	377	2	,	,	PUNCT
ejpam-6638	377	3	13:2114	13:2114	NUM
ejpam-6638	377	4	,	,	PUNCT
ejpam-6638	377	5	2021	2021	NUM
ejpam-6638	377	6	.	.	PUNCT
ejpam-6638	378	1	[	[	X
ejpam-6638	378	2	11	11	NUM
ejpam-6638	378	3	]	]	PUNCT
ejpam-6638	378	4	x.	x.	NOUN
ejpam-6638	378	5	wang	wang	PROPN
ejpam-6638	378	6	,	,	PUNCT
ejpam-6638	378	7	r.	r.	PROPN
ejpam-6638	378	8	girshick	girshick	PROPN
ejpam-6638	378	9	,	,	PUNCT
ejpam-6638	378	10	a.	a.	NOUN
ejpam-6638	378	11	gupta	gupta	PROPN
ejpam-6638	378	12	,	,	PUNCT
ejpam-6638	378	13	and	and	CCONJ
ejpam-6638	378	14	k.	k.	PROPN
ejpam-6638	378	15	he	he	PRON
ejpam-6638	378	16	.	.	PUNCT
ejpam-6638	379	1	non	non	ADJ
ejpam-6638	379	2	-	-	ADJ
ejpam-6638	379	3	local	local	ADJ
ejpam-6638	379	4	neural	neural	ADJ
ejpam-6638	379	5	networks	network	NOUN
ejpam-6638	379	6	.	.	PUNCT
ejpam-6638	380	1	springer	springer	NOUN
ejpam-6638	380	2	lecture	lecture	NOUN
ejpam-6638	380	3	notes	note	NOUN
ejpam-6638	380	4	in	in	ADP
ejpam-6638	380	5	computer	computer	NOUN
ejpam-6638	380	6	science	science	NOUN
ejpam-6638	380	7	,	,	PUNCT
ejpam-6638	380	8	11211:779–796	11211:779–796	NUM
ejpam-6638	380	9	,	,	PUNCT
ejpam-6638	380	10	2018	2018	NUM
ejpam-6638	380	11	.	.	PUNCT
ejpam-6638	381	1	[	[	X
ejpam-6638	381	2	12	12	NUM
ejpam-6638	381	3	]	]	PUNCT
ejpam-6638	381	4	s.	s.	PROPN
ejpam-6638	381	5	woo	woo	PROPN
ejpam-6638	381	6	,	,	PUNCT
ejpam-6638	381	7	j.	j.	PROPN
ejpam-6638	381	8	park	park	PROPN
ejpam-6638	381	9	,	,	PUNCT
ejpam-6638	381	10	j.y	j.y	PROPN
ejpam-6638	381	11	.	.	PROPN
ejpam-6638	381	12	lee	lee	PROPN
ejpam-6638	381	13	,	,	PUNCT
ejpam-6638	381	14	and	and	CCONJ
ejpam-6638	381	15	i.s	i.s	PROPN
ejpam-6638	381	16	.	.	PROPN
ejpam-6638	381	17	kweon	kweon	PROPN
ejpam-6638	381	18	.	.	PUNCT
ejpam-6638	382	1	cbam	cbam	NOUN
ejpam-6638	382	2	:	:	PUNCT
ejpam-6638	382	3	convolutional	convolutional	ADJ
ejpam-6638	382	4	block	block	NOUN
ejpam-6638	382	5	attention	attention	NOUN
ejpam-6638	382	6	module	module	NOUN
ejpam-6638	382	7	.	.	PUNCT
ejpam-6638	383	1	sensors	sensor	NOUN
ejpam-6638	383	2	,	,	PUNCT
ejpam-6638	383	3	18:4271	18:4271	NUM
ejpam-6638	383	4	,	,	PUNCT
ejpam-6638	383	5	2018	2018	NUM
ejpam-6638	383	6	.	.	PUNCT
ejpam-6638	384	1	[	[	X
ejpam-6638	384	2	13	13	NUM
ejpam-6638	384	3	]	]	PUNCT
ejpam-6638	384	4	j.	j.	PROPN
ejpam-6638	384	5	fu	fu	PROPN
ejpam-6638	384	6	,	,	PUNCT
ejpam-6638	384	7	j.	j.	PROPN
ejpam-6638	384	8	liu	liu	PROPN
ejpam-6638	384	9	,	,	PUNCT
ejpam-6638	384	10	h.	h.	PROPN
ejpam-6638	384	11	tian	tian	PROPN
ejpam-6638	384	12	,	,	PUNCT
ejpam-6638	384	13	y.	y.	PROPN
ejpam-6638	384	14	li	li	PROPN
ejpam-6638	384	15	,	,	PUNCT
ejpam-6638	384	16	y.	y.	PROPN
ejpam-6638	384	17	bao	bao	PROPN
ejpam-6638	384	18	,	,	PUNCT
ejpam-6638	384	19	z.	z.	PROPN
ejpam-6638	384	20	fang	fang	PROPN
ejpam-6638	384	21	,	,	PUNCT
ejpam-6638	384	22	and	and	CCONJ
ejpam-6638	384	23	h.	h.	PROPN
ejpam-6638	384	24	lu	lu	PROPN
ejpam-6638	384	25	.	.	PUNCT
ejpam-6638	385	1	dual	dual	ADJ
ejpam-6638	385	2	attention	attention	NOUN
ejpam-6638	385	3	network	network	NOUN
ejpam-6638	385	4	for	for	ADP
ejpam-6638	385	5	scene	scene	NOUN
ejpam-6638	385	6	segmentation	segmentation	NOUN
ejpam-6638	385	7	.	.	PUNCT
ejpam-6638	386	1	pattern	pattern	NOUN
ejpam-6638	386	2	recognition	recognition	PROPN
ejpam-6638	386	3	,	,	PUNCT
ejpam-6638	386	4	100:107110	100:107110	NUM
ejpam-6638	386	5	,	,	PUNCT
ejpam-6638	386	6	2020	2020	NUM
ejpam-6638	386	7	.	.	PUNCT
ejpam-6638	387	1	[	[	X
ejpam-6638	387	2	14	14	NUM
ejpam-6638	387	3	]	]	X
ejpam-6638	387	4	k.	k.	PROPN
ejpam-6638	387	5	li	li	PROPN
ejpam-6638	387	6	,	,	PUNCT
ejpam-6638	387	7	y.	y.	PROPN
ejpam-6638	387	8	you	you	PRON
ejpam-6638	387	9	,	,	PUNCT
ejpam-6638	387	10	y.	y.	PROPN
ejpam-6638	387	11	lu	lu	PROPN
ejpam-6638	387	12	,	,	PUNCT
ejpam-6638	387	13	and	and	CCONJ
ejpam-6638	387	14	x.	x.	NOUN
ejpam-6638	387	15	wang	wang	PROPN
ejpam-6638	387	16	.	.	PUNCT
ejpam-6638	388	1	revisiting	revisit	VERB
ejpam-6638	388	2	pooling	pooling	NOUN
ejpam-6638	388	3	in	in	ADP
ejpam-6638	388	4	deep	deep	ADJ
ejpam-6638	388	5	cnn	cnn	PROPN
ejpam-6638	388	6	for	for	ADP
ejpam-6638	388	7	vision	vision	NOUN
ejpam-6638	388	8	tasks	task	NOUN
ejpam-6638	388	9	.	.	PUNCT
ejpam-6638	389	1	pattern	pattern	NOUN
ejpam-6638	389	2	recognition	recognition	NOUN
ejpam-6638	389	3	letters	letter	NOUN
ejpam-6638	389	4	,	,	PUNCT
ejpam-6638	389	5	133:192–198	133:192–198	NUM
ejpam-6638	389	6	,	,	PUNCT
ejpam-6638	389	7	2020	2020	NUM
ejpam-6638	389	8	.	.	PUNCT
ejpam-6638	390	1	m.	m.	NOUN
ejpam-6638	390	2	israr	israr	PROPN
ejpam-6638	390	3	et	et	PROPN
ejpam-6638	390	4	al	al	PROPN
ejpam-6638	390	5	.	.	PUNCT
ejpam-6638	390	6	/	/	SYM
ejpam-6638	390	7	eur	eur	PROPN
ejpam-6638	390	8	.	.	PUNCT
ejpam-6638	391	1	j.	j.	PROPN
ejpam-6638	391	2	pure	pure	PROPN
ejpam-6638	391	3	appl	appl	PROPN
ejpam-6638	391	4	.	.	PROPN
ejpam-6638	391	5	math	math	PROPN
ejpam-6638	391	6	,	,	PUNCT
ejpam-6638	391	7	18	18	NUM
ejpam-6638	391	8	(	(	PUNCT
ejpam-6638	391	9	4	4	NUM
ejpam-6638	391	10	)	)	PUNCT
ejpam-6638	391	11	(	(	PUNCT
ejpam-6638	391	12	2025	2025	NUM
ejpam-6638	391	13	)	)	PUNCT
ejpam-6638	391	14	,	,	PUNCT
ejpam-6638	391	15	6638	6638	NUM
ejpam-6638	391	16	20	20	NUM
ejpam-6638	391	17	of	of	ADP
ejpam-6638	391	18	22	22	NUM
ejpam-6638	392	1	[	[	X
ejpam-6638	392	2	15	15	NUM
ejpam-6638	392	3	]	]	X
ejpam-6638	392	4	y.	y.	PROPN
ejpam-6638	392	5	zhang	zhang	PROPN
ejpam-6638	392	6	,	,	PUNCT
ejpam-6638	392	7	s.	s.	PROPN
ejpam-6638	392	8	liu	liu	PROPN
ejpam-6638	392	9	,	,	PUNCT
ejpam-6638	392	10	z.	z.	PROPN
ejpam-6638	392	11	lin	lin	PROPN
ejpam-6638	392	12	,	,	PUNCT
ejpam-6638	392	13	h.	h.	PROPN
ejpam-6638	392	14	qi	qi	PROPN
ejpam-6638	392	15	,	,	PUNCT
ejpam-6638	392	16	and	and	CCONJ
ejpam-6638	392	17	p.	p.	PROPN
ejpam-6638	392	18	luo	luo	PROPN
ejpam-6638	392	19	.	.	PUNCT
ejpam-6638	393	1	residual	residual	ADJ
ejpam-6638	393	2	non	non	ADJ
ejpam-6638	393	3	-	-	ADJ
ejpam-6638	393	4	local	local	ADJ
ejpam-6638	393	5	attention	attention	NOUN
ejpam-6638	393	6	networks	network	NOUN
ejpam-6638	393	7	for	for	ADP
ejpam-6638	393	8	image	image	NOUN
ejpam-6638	393	9	restoration	restoration	NOUN
ejpam-6638	393	10	.	.	PUNCT
ejpam-6638	394	1	aims	aim	VERB
ejpam-6638	394	2	mathematics	mathematic	NOUN
ejpam-6638	394	3	,	,	PUNCT
ejpam-6638	394	4	4:964–977	4:964–977	NOUN
ejpam-6638	394	5	,	,	PUNCT
ejpam-6638	394	6	2019	2019	NUM
ejpam-6638	394	7	.	.	PUNCT
ejpam-6638	395	1	[	[	X
ejpam-6638	395	2	16	16	NUM
ejpam-6638	395	3	]	]	X
ejpam-6638	395	4	shuang	shuang	PROPN
ejpam-6638	395	5	chen	chen	PROPN
ejpam-6638	395	6	,	,	PUNCT
ejpam-6638	395	7	amir	amir	PROPN
ejpam-6638	395	8	atapour	atapour	PROPN
ejpam-6638	395	9	-	-	PUNCT
ejpam-6638	395	10	abarghouei	abarghouei	PROPN
ejpam-6638	395	11	,	,	PUNCT
ejpam-6638	395	12	and	and	CCONJ
ejpam-6638	395	13	hubert	hubert	PROPN
ejpam-6638	395	14	p.h	p.h	PROPN
ejpam-6638	395	15	.	.	PROPN
ejpam-6638	395	16	shum	shum	PROPN
ejpam-6638	395	17	.	.	PUNCT
ejpam-6638	396	1	hint	hint	NOUN
ejpam-6638	396	2	:	:	PUNCT
ejpam-6638	396	3	high	high	ADJ
ejpam-6638	396	4	-	-	PUNCT
ejpam-6638	396	5	quality	quality	NOUN
ejpam-6638	396	6	inpainting	inpainting	ADJ
ejpam-6638	396	7	transformer	transformer	NOUN
ejpam-6638	396	8	with	with	ADP
ejpam-6638	396	9	mask	mask	NOUN
ejpam-6638	396	10	-	-	PUNCT
ejpam-6638	396	11	aware	aware	ADJ
ejpam-6638	396	12	encoding	encoding	NOUN
ejpam-6638	396	13	and	and	CCONJ
ejpam-6638	396	14	enhanced	enhance	VERB
ejpam-6638	396	15	attention	attention	NOUN
ejpam-6638	396	16	.	.	PUNCT
ejpam-6638	397	1	ieee	ieee	NOUN
ejpam-6638	397	2	transactions	transaction	NOUN
ejpam-6638	397	3	on	on	ADP
ejpam-6638	397	4	multimedia	multimedia	NOUN
ejpam-6638	397	5	,	,	PUNCT
ejpam-6638	397	6	xx(yy):zz	xx(yy):zz	NUM
ejpam-6638	397	7	–	–	PUNCT
ejpam-6638	397	8	aa	aa	NOUN
ejpam-6638	397	9	,	,	PUNCT
ejpam-6638	397	10	2024	2024	NUM
ejpam-6638	397	11	.	.	PUNCT
ejpam-6638	398	1	[	[	X
ejpam-6638	398	2	17	17	NUM
ejpam-6638	398	3	]	]	X
ejpam-6638	398	4	cai	cai	PROPN
ejpam-6638	398	5	ran	run	VERB
ejpam-6638	398	6	,	,	PUNCT
ejpam-6638	398	7	xinfu	xinfu	PROPN
ejpam-6638	398	8	li	li	PROPN
ejpam-6638	398	9	,	,	PUNCT
ejpam-6638	398	10	and	and	CCONJ
ejpam-6638	398	11	fang	fang	PROPN
ejpam-6638	398	12	yang	yang	PROPN
ejpam-6638	398	13	.	.	PUNCT
ejpam-6638	399	1	multi	multi	ADJ
ejpam-6638	399	2	-	-	ADJ
ejpam-6638	399	3	step	step	ADJ
ejpam-6638	399	4	structure	structure	NOUN
ejpam-6638	399	5	image	image	NOUN
ejpam-6638	399	6	inpainting	inpainte	VERB
ejpam-6638	399	7	model	model	NOUN
ejpam-6638	399	8	with	with	ADP
ejpam-6638	399	9	attention	attention	NOUN
ejpam-6638	399	10	mechanism	mechanism	NOUN
ejpam-6638	399	11	.	.	PUNCT
ejpam-6638	400	1	sensors	sensor	NOUN
ejpam-6638	400	2	,	,	PUNCT
ejpam-6638	400	3	23(4):2316	23(4):2316	NUM
ejpam-6638	400	4	,	,	PUNCT
ejpam-6638	400	5	2023	2023	NUM
ejpam-6638	400	6	.	.	PUNCT
ejpam-6638	401	1	[	[	X
ejpam-6638	401	2	18	18	NUM
ejpam-6638	401	3	]	]	X
ejpam-6638	401	4	l.	l.	PROPN
ejpam-6638	401	5	a.	a.	PROPN
ejpam-6638	401	6	martinho	martinho	PROPN
ejpam-6638	401	7	,	,	PUNCT
ejpam-6638	401	8	j.	j.	PROPN
ejpam-6638	401	9	m.	m.	PROPN
ejpam-6638	401	10	b.	b.	PROPN
ejpam-6638	401	11	cavalcanti	cavalcanti	PROPN
ejpam-6638	401	12	,	,	PUNCT
ejpam-6638	401	13	j.	j.	PROPN
ejpam-6638	401	14	l.	l.	PROPN
ejpam-6638	401	15	pio	pio	PROPN
ejpam-6638	401	16	,	,	PUNCT
ejpam-6638	401	17	et	et	PROPN
ejpam-6638	401	18	al	al	PROPN
ejpam-6638	401	19	.	.	PUNCT
ejpam-6638	401	20	diving	diving	NOUN
ejpam-6638	401	21	into	into	ADP
ejpam-6638	401	22	clarity	clarity	NOUN
ejpam-6638	401	23	:	:	PUNCT
ejpam-6638	401	24	restoring	restore	VERB
ejpam-6638	401	25	underwater	underwater	ADJ
ejpam-6638	401	26	images	image	NOUN
ejpam-6638	401	27	using	use	VERB
ejpam-6638	401	28	deep	deep	ADJ
ejpam-6638	401	29	learning	learning	NOUN
ejpam-6638	401	30	.	.	PUNCT
ejpam-6638	402	1	journal	journal	PROPN
ejpam-6638	402	2	of	of	ADP
ejpam-6638	402	3	intelligent	intelligent	ADJ
ejpam-6638	402	4	&	&	CCONJ
ejpam-6638	402	5	robotic	robotic	ADJ
ejpam-6638	402	6	systems	system	NOUN
ejpam-6638	402	7	,	,	PUNCT
ejpam-6638	402	8	110:32	110:32	NUM
ejpam-6638	402	9	,	,	PUNCT
ejpam-6638	402	10	2024	2024	NUM
ejpam-6638	402	11	.	.	PUNCT
ejpam-6638	403	1	[	[	X
ejpam-6638	403	2	19	19	NUM
ejpam-6638	403	3	]	]	PUNCT
ejpam-6638	403	4	b.	b.	PROPN
ejpam-6638	403	5	lin	lin	PROPN
ejpam-6638	403	6	,	,	PUNCT
ejpam-6638	403	7	y.	y.	PROPN
ejpam-6638	403	8	tian	tian	PROPN
ejpam-6638	403	9	,	,	PUNCT
ejpam-6638	403	10	y.	y.	PROPN
ejpam-6638	403	11	zhang	zhang	PROPN
ejpam-6638	403	12	,	,	PUNCT
ejpam-6638	403	13	z.	z.	PROPN
ejpam-6638	403	14	zhu	zhu	PROPN
ejpam-6638	403	15	,	,	PUNCT
ejpam-6638	403	16	and	and	CCONJ
ejpam-6638	403	17	d.	d.	PROPN
ejpam-6638	403	18	wang	wang	PROPN
ejpam-6638	403	19	.	.	PUNCT
ejpam-6638	404	1	deep	deep	ADJ
ejpam-6638	404	2	learning	learning	NOUN
ejpam-6638	404	3	methods	method	NOUN
ejpam-6638	404	4	for	for	ADP
ejpam-6638	404	5	high	high	ADJ
ejpam-6638	404	6	-	-	PUNCT
ejpam-6638	404	7	resolution	resolution	NOUN
ejpam-6638	404	8	microscale	microscale	ADJ
ejpam-6638	404	9	light	light	ADJ
ejpam-6638	404	10	-	-	PUNCT
ejpam-6638	404	11	field	field	NOUN
ejpam-6638	404	12	image	image	NOUN
ejpam-6638	404	13	reconstruction	reconstruction	NOUN
ejpam-6638	404	14	:	:	PUNCT
ejpam-6638	404	15	a	a	DET
ejpam-6638	404	16	survey	survey	NOUN
ejpam-6638	404	17	.	.	PUNCT
ejpam-6638	405	1	frontiers	frontier	NOUN
ejpam-6638	405	2	in	in	ADP
ejpam-6638	405	3	bioengineering	bioengineering	NOUN
ejpam-6638	405	4	and	and	CCONJ
ejpam-6638	405	5	biotechnology	biotechnology	NOUN
ejpam-6638	405	6	,	,	PUNCT
ejpam-6638	405	7	12:1500270	12:1500270	NUM
ejpam-6638	405	8	,	,	PUNCT
ejpam-6638	405	9	2024	2024	NUM
ejpam-6638	405	10	.	.	PUNCT
ejpam-6638	406	1	[	[	X
ejpam-6638	406	2	20	20	NUM
ejpam-6638	406	3	]	]	PUNCT
ejpam-6638	406	4	jun	jun	PROPN
ejpam-6638	406	5	li	li	PROPN
ejpam-6638	406	6	,	,	PUNCT
ejpam-6638	406	7	heran	heran	PROPN
ejpam-6638	406	8	wang	wang	PROPN
ejpam-6638	406	9	,	,	PUNCT
ejpam-6638	406	10	yingjie	yingjie	PROPN
ejpam-6638	406	11	li	li	PROPN
ejpam-6638	406	12	,	,	PUNCT
ejpam-6638	406	13	and	and	CCONJ
ejpam-6638	406	14	haochuan	haochuan	PROPN
ejpam-6638	406	15	zhang	zhang	PROPN
ejpam-6638	406	16	.	.	PUNCT
ejpam-6638	407	1	a	a	DET
ejpam-6638	407	2	comprehensive	comprehensive	ADJ
ejpam-6638	407	3	review	review	NOUN
ejpam-6638	407	4	of	of	ADP
ejpam-6638	407	5	image	image	NOUN
ejpam-6638	407	6	restoration	restoration	NOUN
ejpam-6638	407	7	research	research	NOUN
ejpam-6638	407	8	based	base	VERB
ejpam-6638	407	9	on	on	ADP
ejpam-6638	407	10	diffusion	diffusion	NOUN
ejpam-6638	407	11	models	model	NOUN
ejpam-6638	407	12	.	.	PUNCT
ejpam-6638	408	1	mathematics	mathematic	NOUN
ejpam-6638	408	2	,	,	PUNCT
ejpam-6638	408	3	13(13):2079	13(13):2079	NUM
ejpam-6638	408	4	,	,	PUNCT
ejpam-6638	408	5	2025	2025	NUM
ejpam-6638	408	6	.	.	PUNCT
ejpam-6638	409	1	[	[	X
ejpam-6638	409	2	21	21	NUM
ejpam-6638	409	3	]	]	X
ejpam-6638	409	4	p.	p.	PROPN
ejpam-6638	409	5	gupta	gupta	PROPN
ejpam-6638	409	6	,	,	PUNCT
ejpam-6638	409	7	r.	r.	PROPN
ejpam-6638	409	8	grosse	grosse	PROPN
ejpam-6638	409	9	,	,	PUNCT
ejpam-6638	409	10	and	and	CCONJ
ejpam-6638	409	11	d.	d.	PROPN
ejpam-6638	409	12	goldfarb	goldfarb	PROPN
ejpam-6638	409	13	.	.	PUNCT
ejpam-6638	410	1	shampoo	shampoo	PROPN
ejpam-6638	410	2	:	:	PUNCT
ejpam-6638	410	3	preconditioned	precondition	VERB
ejpam-6638	410	4	stochastic	stochastic	ADJ
ejpam-6638	410	5	tensor	tensor	NOUN
ejpam-6638	410	6	optimization	optimization	NOUN
ejpam-6638	410	7	.	.	PUNCT
ejpam-6638	411	1	journal	journal	PROPN
ejpam-6638	411	2	of	of	ADP
ejpam-6638	411	3	machine	machine	NOUN
ejpam-6638	411	4	learning	learn	VERB
ejpam-6638	411	5	research	research	NOUN
ejpam-6638	411	6	,	,	PUNCT
ejpam-6638	411	7	19:1–22	19:1–22	NUM
ejpam-6638	411	8	,	,	PUNCT
ejpam-6638	411	9	2018	2018	NUM
ejpam-6638	411	10	.	.	PUNCT
ejpam-6638	412	1	[	[	X
ejpam-6638	412	2	22	22	NUM
ejpam-6638	412	3	]	]	PUNCT
ejpam-6638	412	4	j.	j.	PROPN
ejpam-6638	412	5	martens	martens	PROPN
ejpam-6638	412	6	and	and	CCONJ
ejpam-6638	412	7	r.	r.	PROPN
ejpam-6638	412	8	grosse	grosse	PROPN
ejpam-6638	412	9	.	.	PUNCT
ejpam-6638	413	1	optimizing	optimize	VERB
ejpam-6638	413	2	neural	neural	ADJ
ejpam-6638	413	3	networks	network	NOUN
ejpam-6638	413	4	with	with	ADP
ejpam-6638	413	5	kronecker	kronecker	NOUN
ejpam-6638	413	6	-	-	PUNCT
ejpam-6638	413	7	factored	factor	VERB
ejpam-6638	413	8	approximate	approximate	ADJ
ejpam-6638	413	9	curvature	curvature	NOUN
ejpam-6638	413	10	.	.	PUNCT
ejpam-6638	414	1	springer	springer	NOUN
ejpam-6638	414	2	lecture	lecture	NOUN
ejpam-6638	414	3	notes	note	NOUN
ejpam-6638	414	4	in	in	ADP
ejpam-6638	414	5	computer	computer	NOUN
ejpam-6638	414	6	science	science	NOUN
ejpam-6638	414	7	,	,	PUNCT
ejpam-6638	414	8	37:2408–2417	37:2408–2417	NUM
ejpam-6638	414	9	,	,	PUNCT
ejpam-6638	414	10	2015	2015	NUM
ejpam-6638	414	11	.	.	PUNCT
ejpam-6638	415	1	[	[	X
ejpam-6638	415	2	23	23	NUM
ejpam-6638	415	3	]	]	PUNCT
ejpam-6638	415	4	t.	t.	PROPN
ejpam-6638	415	5	chen	chen	PROPN
ejpam-6638	415	6	,	,	PUNCT
ejpam-6638	415	7	m.	m.	PROPN
ejpam-6638	415	8	lucic	lucic	PROPN
ejpam-6638	415	9	,	,	PUNCT
ejpam-6638	415	10	n.	n.	PROPN
ejpam-6638	415	11	houlsby	houlsby	PROPN
ejpam-6638	415	12	,	,	PUNCT
ejpam-6638	415	13	and	and	CCONJ
ejpam-6638	415	14	y.	y.	PROPN
ejpam-6638	415	15	wang	wang	PROPN
ejpam-6638	415	16	.	.	PUNCT
ejpam-6638	415	17	natural	natural	ADJ
ejpam-6638	415	18	gradient	gradient	ADJ
ejpam-6638	415	19	descent	descent	NOUN
ejpam-6638	415	20	for	for	ADP
ejpam-6638	415	21	overparameterized	overparameterized	ADJ
ejpam-6638	415	22	networks	network	NOUN
ejpam-6638	415	23	.	.	PUNCT
ejpam-6638	416	1	neurocomputing	neurocomputing	NOUN
ejpam-6638	416	2	,	,	PUNCT
ejpam-6638	416	3	337:39–47	337:39–47	PROPN
ejpam-6638	416	4	,	,	PUNCT
ejpam-6638	416	5	2018	2018	NUM
ejpam-6638	416	6	.	.	PUNCT
ejpam-6638	417	1	[	[	X
ejpam-6638	417	2	24	24	NUM
ejpam-6638	417	3	]	]	PUNCT
ejpam-6638	417	4	x.	x.	NOUN
ejpam-6638	417	5	dong	dong	PROPN
ejpam-6638	417	6	,	,	PUNCT
ejpam-6638	417	7	j.	j.	PROPN
ejpam-6638	417	8	pan	pan	PROPN
ejpam-6638	417	9	,	,	PUNCT
ejpam-6638	417	10	t.	t.	PROPN
ejpam-6638	417	11	xiang	xiang	PROPN
ejpam-6638	417	12	,	,	PUNCT
ejpam-6638	417	13	w.	w.	PROPN
ejpam-6638	417	14	zuo	zuo	PROPN
ejpam-6638	417	15	,	,	PUNCT
ejpam-6638	417	16	and	and	CCONJ
ejpam-6638	417	17	c.c	c.c	PROPN
ejpam-6638	417	18	.	.	PROPN
ejpam-6638	417	19	loy	loy	PROPN
ejpam-6638	417	20	.	.	PUNCT
ejpam-6638	418	1	attention	attention	NOUN
ejpam-6638	418	2	-	-	PUNCT
ejpam-6638	418	3	augmented	augment	VERB
ejpam-6638	418	4	image	image	NOUN
ejpam-6638	418	5	restoration	restoration	NOUN
ejpam-6638	418	6	with	with	ADP
ejpam-6638	418	7	deep	deep	ADJ
ejpam-6638	418	8	image	image	NOUN
ejpam-6638	418	9	prior	prior	ADV
ejpam-6638	418	10	.	.	PUNCT
ejpam-6638	419	1	ieee	ieee	NOUN
ejpam-6638	419	2	access	access	NOUN
ejpam-6638	419	3	,	,	PUNCT
ejpam-6638	419	4	8:15319–15329	8:15319–15329	NUM
ejpam-6638	419	5	,	,	PUNCT
ejpam-6638	419	6	2020	2020	NUM
ejpam-6638	419	7	.	.	PUNCT
ejpam-6638	420	1	[	[	X
ejpam-6638	420	2	25	25	NUM
ejpam-6638	420	3	]	]	PUNCT
ejpam-6638	420	4	z.	z.	PROPN
ejpam-6638	420	5	yao	yao	PROPN
ejpam-6638	420	6	,	,	PUNCT
ejpam-6638	420	7	a.	a.	NOUN
ejpam-6638	420	8	gholami	gholami	PROPN
ejpam-6638	420	9	,	,	PUNCT
ejpam-6638	420	10	k.	k.	PROPN
ejpam-6638	420	11	keutzer	keutzer	PROPN
ejpam-6638	420	12	,	,	PUNCT
ejpam-6638	420	13	and	and	CCONJ
ejpam-6638	420	14	m.w	m.w	PROPN
ejpam-6638	420	15	.	.	PROPN
ejpam-6638	420	16	mahoney	mahoney	PROPN
ejpam-6638	420	17	.	.	PUNCT
ejpam-6638	421	1	large	large	ADJ
ejpam-6638	421	2	batch	batch	NOUN
ejpam-6638	421	3	size	size	NOUN
ejpam-6638	421	4	training	training	NOUN
ejpam-6638	421	5	of	of	ADP
ejpam-6638	421	6	neural	neural	ADJ
ejpam-6638	421	7	networks	network	NOUN
ejpam-6638	421	8	with	with	ADP
ejpam-6638	421	9	adversarial	adversarial	ADJ
ejpam-6638	421	10	robustness	robustness	NOUN
ejpam-6638	421	11	.	.	PUNCT
ejpam-6638	422	1	journal	journal	NOUN
ejpam-6638	422	2	of	of	ADP
ejpam-6638	422	3	computer	computer	NOUN
ejpam-6638	422	4	science	science	NOUN
ejpam-6638	422	5	and	and	CCONJ
ejpam-6638	422	6	technology	technology	NOUN
ejpam-6638	422	7	,	,	PUNCT
ejpam-6638	422	8	33:109–123	33:109–123	PROPN
ejpam-6638	422	9	,	,	PUNCT
ejpam-6638	422	10	2018	2018	NUM
ejpam-6638	422	11	.	.	PUNCT
ejpam-6638	423	1	[	[	X
ejpam-6638	423	2	26	26	NUM
ejpam-6638	423	3	]	]	PUNCT
ejpam-6638	423	4	x.	x.	NOUN
ejpam-6638	424	1	he	he	PRON
ejpam-6638	424	2	,	,	PUNCT
ejpam-6638	424	3	y.	y.	PROPN
ejpam-6638	424	4	zhang	zhang	PROPN
ejpam-6638	424	5	,	,	PUNCT
ejpam-6638	424	6	and	and	CCONJ
ejpam-6638	424	7	y.	y.	PROPN
ejpam-6638	424	8	tang	tang	PROPN
ejpam-6638	424	9	.	.	PUNCT
ejpam-6638	425	1	image	image	NOUN
ejpam-6638	425	2	structure	structure	NOUN
ejpam-6638	425	3	preservation	preservation	NOUN
ejpam-6638	425	4	via	via	ADP
ejpam-6638	425	5	svd	svd	PROPN
ejpam-6638	425	6	and	and	CCONJ
ejpam-6638	425	7	non	non	ADJ
ejpam-6638	425	8	-	-	ADJ
ejpam-6638	425	9	local	local	ADJ
ejpam-6638	425	10	attention	attention	NOUN
ejpam-6638	425	11	.	.	PUNCT
ejpam-6638	426	1	journal	journal	NOUN
ejpam-6638	426	2	of	of	ADP
ejpam-6638	426	3	visual	visual	ADJ
ejpam-6638	426	4	communication	communication	NOUN
ejpam-6638	426	5	and	and	CCONJ
ejpam-6638	426	6	image	image	NOUN
ejpam-6638	426	7	representation	representation	NOUN
ejpam-6638	426	8	,	,	PUNCT
ejpam-6638	426	9	71:102859	71:102859	NUM
ejpam-6638	426	10	,	,	PUNCT
ejpam-6638	426	11	2020	2020	NUM
ejpam-6638	426	12	.	.	PUNCT
ejpam-6638	427	1	[	[	X
ejpam-6638	427	2	27	27	NUM
ejpam-6638	427	3	]	]	X
ejpam-6638	427	4	xiaolong	xiaolong	PROPN
ejpam-6638	427	5	wang	wang	PROPN
ejpam-6638	427	6	,	,	PUNCT
ejpam-6638	427	7	ross	ross	PROPN
ejpam-6638	427	8	girshick	girshick	PROPN
ejpam-6638	427	9	,	,	PUNCT
ejpam-6638	427	10	abhinav	abhinav	PROPN
ejpam-6638	427	11	gupta	gupta	PROPN
ejpam-6638	427	12	,	,	PUNCT
ejpam-6638	427	13	and	and	CCONJ
ejpam-6638	427	14	kaiming	kaime	VERB
ejpam-6638	427	15	he	he	PRON
ejpam-6638	427	16	.	.	PUNCT
ejpam-6638	428	1	non	non	ADJ
ejpam-6638	428	2	-	-	ADJ
ejpam-6638	428	3	local	local	ADJ
ejpam-6638	428	4	neural	neural	ADJ
ejpam-6638	428	5	networks	network	NOUN
ejpam-6638	428	6	.	.	PUNCT
ejpam-6638	429	1	in	in	ADP
ejpam-6638	429	2	proceedings	proceeding	NOUN
ejpam-6638	429	3	of	of	ADP
ejpam-6638	429	4	the	the	DET
ejpam-6638	429	5	ieee	ieee	NOUN
ejpam-6638	429	6	conference	conference	NOUN
ejpam-6638	429	7	on	on	ADP
ejpam-6638	429	8	computer	computer	NOUN
ejpam-6638	429	9	vision	vision	NOUN
ejpam-6638	429	10	and	and	CCONJ
ejpam-6638	429	11	pattern	pattern	NOUN
ejpam-6638	429	12	recognition	recognition	NOUN
ejpam-6638	429	13	(	(	PUNCT
ejpam-6638	429	14	cvpr	cvpr	NOUN
ejpam-6638	429	15	)	)	PUNCT
ejpam-6638	429	16	,	,	PUNCT
ejpam-6638	429	17	pages	page	NOUN
ejpam-6638	429	18	7794–7803	7794–7803	NOUN
ejpam-6638	429	19	,	,	PUNCT
ejpam-6638	429	20	2018	2018	NUM
ejpam-6638	429	21	.	.	PUNCT
ejpam-6638	430	1	[	[	X
ejpam-6638	430	2	28	28	NUM
ejpam-6638	430	3	]	]	X
ejpam-6638	430	4	jun	jun	PROPN
ejpam-6638	430	5	fu	fu	PROPN
ejpam-6638	430	6	,	,	PUNCT
ejpam-6638	430	7	jing	jing	PROPN
ejpam-6638	430	8	liu	liu	PROPN
ejpam-6638	430	9	,	,	PUNCT
ejpam-6638	430	10	haijie	haijie	PROPN
ejpam-6638	430	11	tian	tian	PROPN
ejpam-6638	430	12	,	,	PUNCT
ejpam-6638	430	13	yong	yong	PROPN
ejpam-6638	430	14	li	li	PROPN
ejpam-6638	430	15	,	,	PUNCT
ejpam-6638	430	16	yongjun	yongjun	PROPN
ejpam-6638	430	17	bao	bao	PROPN
ejpam-6638	430	18	,	,	PUNCT
ejpam-6638	430	19	zhiwei	zhiwei	PROPN
ejpam-6638	430	20	fang	fang	PROPN
ejpam-6638	430	21	,	,	PUNCT
ejpam-6638	430	22	and	and	CCONJ
ejpam-6638	430	23	hanqing	hanqe	VERB
ejpam-6638	430	24	lu	lu	NOUN
ejpam-6638	430	25	.	.	PUNCT
ejpam-6638	431	1	dual	dual	ADJ
ejpam-6638	431	2	attention	attention	NOUN
ejpam-6638	431	3	network	network	NOUN
ejpam-6638	431	4	for	for	ADP
ejpam-6638	431	5	scene	scene	NOUN
ejpam-6638	431	6	segmentation	segmentation	NOUN
ejpam-6638	431	7	.	.	PUNCT
ejpam-6638	432	1	in	in	ADP
ejpam-6638	432	2	proceedings	proceeding	NOUN
ejpam-6638	432	3	of	of	ADP
ejpam-6638	432	4	the	the	DET
ejpam-6638	432	5	ieee	ieee	NOUN
ejpam-6638	432	6	/	/	SYM
ejpam-6638	432	7	cvf	cvf	NOUN
ejpam-6638	432	8	conference	conference	NOUN
ejpam-6638	432	9	on	on	ADP
ejpam-6638	432	10	computer	computer	NOUN
ejpam-6638	432	11	vision	vision	NOUN
ejpam-6638	432	12	and	and	CCONJ
ejpam-6638	432	13	pattern	pattern	NOUN
ejpam-6638	432	14	recognition	recognition	NOUN
ejpam-6638	432	15	(	(	PUNCT
ejpam-6638	432	16	cvpr	cvpr	NOUN
ejpam-6638	432	17	)	)	PUNCT
ejpam-6638	432	18	,	,	PUNCT
ejpam-6638	432	19	pages	page	NOUN
ejpam-6638	432	20	3141–3149	3141–3149	NUM
ejpam-6638	432	21	,	,	PUNCT
ejpam-6638	432	22	2019	2019	NUM
ejpam-6638	432	23	.	.	PUNCT
ejpam-6638	433	1	[	[	X
ejpam-6638	433	2	29	29	NUM
ejpam-6638	433	3	]	]	X
ejpam-6638	433	4	y.	y.	PROPN
ejpam-6638	433	5	sun	sun	PROPN
ejpam-6638	433	6	,	,	PUNCT
ejpam-6638	433	7	l.	l.	PROPN
ejpam-6638	433	8	zheng	zheng	PROPN
ejpam-6638	433	9	,	,	PUNCT
ejpam-6638	433	10	w.	w.	PROPN
ejpam-6638	433	11	deng	deng	PROPN
ejpam-6638	433	12	,	,	PUNCT
ejpam-6638	433	13	and	and	CCONJ
ejpam-6638	433	14	s.	s.	PROPN
ejpam-6638	433	15	wang	wang	PROPN
ejpam-6638	433	16	.	.	PUNCT
ejpam-6638	433	17	svdnet	svdnet	PROPN
ejpam-6638	433	18	for	for	ADP
ejpam-6638	433	19	pedestrian	pedestrian	NOUN
ejpam-6638	433	20	retrieval	retrieval	NOUN
ejpam-6638	433	21	.	.	PUNCT
ejpam-6638	434	1	in	in	ADP
ejpam-6638	434	2	proceedings	proceeding	NOUN
ejpam-6638	434	3	of	of	ADP
ejpam-6638	434	4	the	the	DET
ejpam-6638	434	5	ieee	ieee	NOUN
ejpam-6638	434	6	international	international	PROPN
ejpam-6638	434	7	conference	conference	NOUN
ejpam-6638	434	8	on	on	ADP
ejpam-6638	434	9	computer	computer	NOUN
ejpam-6638	434	10	vision	vision	NOUN
ejpam-6638	434	11	,	,	PUNCT
ejpam-6638	434	12	pages	page	NOUN
ejpam-6638	434	13	3800–3808	3800–3808	NUM
ejpam-6638	434	14	,	,	PUNCT
ejpam-6638	434	15	2017	2017	NUM
ejpam-6638	434	16	.	.	PUNCT
ejpam-6638	435	1	[	[	X
ejpam-6638	435	2	30	30	NUM
ejpam-6638	435	3	]	]	X
ejpam-6638	435	4	xizhou	xizhou	PROPN
ejpam-6638	435	5	zhang	zhang	PROPN
ejpam-6638	435	6	,	,	PUNCT
ejpam-6638	435	7	xiaosong	xiaosong	PROPN
ejpam-6638	435	8	ma	ma	PROPN
ejpam-6638	435	9	,	,	PUNCT
ejpam-6638	435	10	jian	jian	PROPN
ejpam-6638	435	11	sun	sun	PROPN
ejpam-6638	435	12	,	,	PUNCT
ejpam-6638	435	13	and	and	CCONJ
ejpam-6638	435	14	zhiwei	zhiwei	PROPN
ejpam-6638	435	15	xiong	xiong	PROPN
ejpam-6638	435	16	.	.	PUNCT
ejpam-6638	436	1	spectral	spectral	ADJ
ejpam-6638	436	2	pooling	pool	VERB
ejpam-6638	436	3	for	for	ADP
ejpam-6638	436	4	dense	dense	ADJ
ejpam-6638	436	5	prediction	prediction	NOUN
ejpam-6638	436	6	tasks	task	NOUN
ejpam-6638	436	7	.	.	PUNCT
ejpam-6638	437	1	in	in	ADP
ejpam-6638	437	2	proceedings	proceeding	NOUN
ejpam-6638	437	3	of	of	ADP
ejpam-6638	437	4	the	the	DET
ejpam-6638	437	5	ieee	ieee	NOUN
ejpam-6638	437	6	international	international	PROPN
ejpam-6638	437	7	conference	conference	NOUN
ejpam-6638	437	8	on	on	ADP
ejpam-6638	437	9	computer	computer	NOUN
ejpam-6638	437	10	vision	vision	NOUN
ejpam-6638	437	11	(	(	PUNCT
ejpam-6638	437	12	iccv	iccv	PROPN
ejpam-6638	437	13	)	)	PUNCT
ejpam-6638	437	14	,	,	PUNCT
ejpam-6638	437	15	pages	page	NOUN
ejpam-6638	437	16	8601–8610	8601–8610	NUM
ejpam-6638	437	17	,	,	PUNCT
ejpam-6638	437	18	2019	2019	NUM
ejpam-6638	437	19	.	.	PUNCT
ejpam-6638	438	1	[	[	X
ejpam-6638	438	2	31	31	NUM
ejpam-6638	438	3	]	]	PUNCT
ejpam-6638	438	4	g.	g.	PROPN
ejpam-6638	438	5	liu	liu	PROPN
ejpam-6638	438	6	,	,	PUNCT
ejpam-6638	438	7	f.a	f.a	PROPN
ejpam-6638	438	8	.	.	PROPN
ejpam-6638	438	9	reda	reda	PROPN
ejpam-6638	438	10	,	,	PUNCT
ejpam-6638	438	11	k.j	k.j	PROPN
ejpam-6638	438	12	.	.	PROPN
ejpam-6638	438	13	shih	shih	PROPN
ejpam-6638	438	14	,	,	PUNCT
ejpam-6638	438	15	t.c	t.c	PROPN
ejpam-6638	438	16	.	.	PROPN
ejpam-6638	438	17	wang	wang	PROPN
ejpam-6638	438	18	,	,	PUNCT
ejpam-6638	438	19	a.	a.	NOUN
ejpam-6638	438	20	tao	tao	PROPN
ejpam-6638	438	21	,	,	PUNCT
ejpam-6638	438	22	and	and	CCONJ
ejpam-6638	438	23	b.	b.	PROPN
ejpam-6638	438	24	catanzaro	catanzaro	PROPN
ejpam-6638	438	25	.	.	PUNCT
ejpam-6638	439	1	image	image	NOUN
ejpam-6638	439	2	inpainting	inpainte	VERB
ejpam-6638	439	3	for	for	ADP
ejpam-6638	439	4	irregular	irregular	ADJ
ejpam-6638	439	5	holes	hole	NOUN
ejpam-6638	439	6	using	use	VERB
ejpam-6638	439	7	partial	partial	ADJ
ejpam-6638	439	8	convolutions	convolution	NOUN
ejpam-6638	439	9	.	.	PUNCT
ejpam-6638	440	1	in	in	ADP
ejpam-6638	440	2	proceedings	proceeding	NOUN
ejpam-6638	440	3	of	of	ADP
ejpam-6638	440	4	the	the	DET
ejpam-6638	440	5	european	european	PROPN
ejpam-6638	440	6	conference	conference	PROPN
ejpam-6638	440	7	on	on	ADP
ejpam-6638	440	8	computer	computer	NOUN
ejpam-6638	440	9	vision	vision	NOUN
ejpam-6638	440	10	(	(	PUNCT
ejpam-6638	440	11	eccv	eccv	ADV
ejpam-6638	440	12	)	)	PUNCT
ejpam-6638	440	13	,	,	PUNCT
ejpam-6638	440	14	pages	page	NOUN
ejpam-6638	440	15	85–100	85–100	PROPN
ejpam-6638	440	16	,	,	PUNCT
ejpam-6638	440	17	2019	2019	NUM
ejpam-6638	440	18	.	.	PUNCT
ejpam-6638	441	1	[	[	X
ejpam-6638	441	2	32	32	NUM
ejpam-6638	441	3	]	]	X
ejpam-6638	441	4	y.	y.	PROPN
ejpam-6638	441	5	zeng	zeng	PROPN
ejpam-6638	441	6	,	,	PUNCT
ejpam-6638	441	7	j.	j.	PROPN
ejpam-6638	441	8	fu	fu	PROPN
ejpam-6638	441	9	,	,	PUNCT
ejpam-6638	441	10	h.	h.	PROPN
ejpam-6638	441	11	chao	chao	PROPN
ejpam-6638	441	12	,	,	PUNCT
ejpam-6638	441	13	and	and	CCONJ
ejpam-6638	441	14	y.	y.	PROPN
ejpam-6638	441	15	yu	yu	PROPN
ejpam-6638	441	16	.	.	PUNCT
ejpam-6638	442	1	learning	learn	VERB
ejpam-6638	442	2	pyramid	pyramid	NOUN
ejpam-6638	442	3	-	-	PUNCT
ejpam-6638	442	4	context	context	NOUN
ejpam-6638	442	5	encoder	encoder	NOUN
ejpam-6638	442	6	network	network	NOUN
ejpam-6638	442	7	for	for	ADP
ejpam-6638	442	8	highquality	highquality	NOUN
ejpam-6638	442	9	image	image	NOUN
ejpam-6638	442	10	inpainting	inpainte	VERB
ejpam-6638	442	11	.	.	PUNCT
ejpam-6638	443	1	in	in	ADP
ejpam-6638	443	2	proceedings	proceeding	NOUN
ejpam-6638	443	3	of	of	ADP
ejpam-6638	443	4	the	the	DET
ejpam-6638	443	5	ieee	ieee	NOUN
ejpam-6638	443	6	/	/	SYM
ejpam-6638	443	7	cvf	cvf	NOUN
ejpam-6638	443	8	conference	conference	NOUN
ejpam-6638	443	9	on	on	ADP
ejpam-6638	443	10	computer	computer	NOUN
ejpam-6638	443	11	vision	vision	NOUN
ejpam-6638	443	12	and	and	CCONJ
ejpam-6638	443	13	pattern	pattern	NOUN
ejpam-6638	443	14	recognition	recognition	NOUN
ejpam-6638	443	15	(	(	PUNCT
ejpam-6638	443	16	cvpr	cvpr	NOUN
ejpam-6638	443	17	)	)	PUNCT
ejpam-6638	443	18	,	,	PUNCT
ejpam-6638	443	19	pages	page	NOUN
ejpam-6638	443	20	1486–1494	1486–1494	NUM
ejpam-6638	443	21	,	,	PUNCT
ejpam-6638	443	22	2020	2020	NUM
ejpam-6638	443	23	.	.	PUNCT
ejpam-6638	444	1	[	[	X
ejpam-6638	444	2	33	33	NUM
ejpam-6638	444	3	]	]	PUNCT
ejpam-6638	444	4	k.	k.	PROPN
ejpam-6638	444	5	nazeri	nazeri	PROPN
ejpam-6638	444	6	,	,	PUNCT
ejpam-6638	444	7	e.	e.	PROPN
ejpam-6638	444	8	ng	ng	PROPN
ejpam-6638	444	9	,	,	PUNCT
ejpam-6638	444	10	t.	t.	PROPN
ejpam-6638	444	11	joseph	joseph	PROPN
ejpam-6638	444	12	,	,	PUNCT
ejpam-6638	444	13	f.	f.	PROPN
ejpam-6638	444	14	qureshi	qureshi	PROPN
ejpam-6638	444	15	,	,	PUNCT
ejpam-6638	444	16	and	and	CCONJ
ejpam-6638	444	17	m.	m.	PROPN
ejpam-6638	444	18	ebrahimi	ebrahimi	PROPN
ejpam-6638	444	19	.	.	PUNCT
ejpam-6638	444	20	edgeconnect	edgeconnect	PROPN
ejpam-6638	444	21	:	:	PUNCT
ejpam-6638	444	22	generative	generative	ADJ
ejpam-6638	444	23	image	image	NOUN
ejpam-6638	444	24	inpainting	inpainte	VERB
ejpam-6638	444	25	with	with	ADP
ejpam-6638	444	26	adversarial	adversarial	ADJ
ejpam-6638	444	27	edge	edge	NOUN
ejpam-6638	444	28	learning	learning	NOUN
ejpam-6638	444	29	.	.	PUNCT
ejpam-6638	445	1	in	in	ADP
ejpam-6638	445	2	proceedings	proceeding	NOUN
ejpam-6638	445	3	of	of	ADP
ejpam-6638	445	4	the	the	DET
ejpam-6638	445	5	ieee	ieee	NOUN
ejpam-6638	445	6	/	/	SYM
ejpam-6638	445	7	cvf	cvf	NOUN
ejpam-6638	445	8	international	international	ADJ
ejpam-6638	445	9	conference	conference	NOUN
ejpam-6638	445	10	on	on	ADP
ejpam-6638	445	11	computer	computer	NOUN
ejpam-6638	445	12	vision	vision	NOUN
ejpam-6638	445	13	workshops	workshop	NOUN
ejpam-6638	445	14	(	(	PUNCT
ejpam-6638	445	15	iccvw	iccvw	PROPN
ejpam-6638	445	16	)	)	PUNCT
ejpam-6638	445	17	,	,	PUNCT
ejpam-6638	445	18	pages	page	NOUN
ejpam-6638	445	19	3265–3274	3265–3274	NUM
ejpam-6638	445	20	,	,	PUNCT
ejpam-6638	445	21	2019	2019	NUM
ejpam-6638	445	22	.	.	PUNCT
ejpam-6638	446	1	[	[	X
ejpam-6638	446	2	34	34	NUM
ejpam-6638	446	3	]	]	X
ejpam-6638	446	4	y.	y.	PROPN
ejpam-6638	446	5	liu	liu	PROPN
ejpam-6638	446	6	,	,	PUNCT
ejpam-6638	446	7	z.	z.	PROPN
ejpam-6638	446	8	wang	wang	PROPN
ejpam-6638	446	9	,	,	PUNCT
ejpam-6638	446	10	y.	y.	PROPN
ejpam-6638	446	11	yang	yang	PROPN
ejpam-6638	446	12	,	,	PUNCT
ejpam-6638	446	13	and	and	CCONJ
ejpam-6638	446	14	g.	g.	PROPN
ejpam-6638	446	15	gao	gao	PROPN
ejpam-6638	446	16	.	.	PUNCT
ejpam-6638	447	1	shift	shift	NOUN
ejpam-6638	447	2	-	-	PUNCT
ejpam-6638	447	3	net	net	NOUN
ejpam-6638	447	4	:	:	PUNCT
ejpam-6638	447	5	image	image	NOUN
ejpam-6638	447	6	inpainting	inpainte	VERB
ejpam-6638	447	7	via	via	ADP
ejpam-6638	447	8	deep	deep	ADJ
ejpam-6638	447	9	feature	feature	NOUN
ejpam-6638	447	10	rearrangement	rearrangement	NOUN
ejpam-6638	447	11	.	.	PUNCT
ejpam-6638	448	1	signal	signal	ADJ
ejpam-6638	448	2	processing	processing	NOUN
ejpam-6638	448	3	:	:	PUNCT
ejpam-6638	448	4	image	image	NOUN
ejpam-6638	448	5	communication	communication	NOUN
ejpam-6638	448	6	,	,	PUNCT
ejpam-6638	448	7	81:115708	81:115708	NUM
ejpam-6638	448	8	,	,	PUNCT
ejpam-6638	448	9	2020	2020	NUM
ejpam-6638	448	10	.	.	PUNCT
ejpam-6638	449	1	[	[	X
ejpam-6638	449	2	35	35	NUM
ejpam-6638	449	3	]	]	PUNCT
ejpam-6638	449	4	x.	x.	NOUN
ejpam-6638	449	5	guo	guo	PROPN
ejpam-6638	449	6	,	,	PUNCT
ejpam-6638	449	7	y.	y.	PROPN
ejpam-6638	449	8	zhang	zhang	PROPN
ejpam-6638	449	9	,	,	PUNCT
ejpam-6638	449	10	h.	h.	PROPN
ejpam-6638	449	11	hu	hu	PROPN
ejpam-6638	449	12	,	,	PUNCT
ejpam-6638	449	13	b.	b.	PROPN
ejpam-6638	449	14	liu	liu	PROPN
ejpam-6638	449	15	,	,	PUNCT
ejpam-6638	449	16	and	and	CCONJ
ejpam-6638	449	17	j.	j.	PROPN
ejpam-6638	449	18	han	han	PROPN
ejpam-6638	449	19	.	.	PROPN
ejpam-6638	450	1	progressive	progressive	ADJ
ejpam-6638	450	2	image	image	NOUN
ejpam-6638	450	3	inpainting	inpainte	VERB
ejpam-6638	450	4	with	with	ADP
ejpam-6638	450	5	fullm	fullm	PRON
ejpam-6638	450	6	.	.	PUNCT
ejpam-6638	451	1	israr	israr	PROPN
ejpam-6638	451	2	et	et	PROPN
ejpam-6638	451	3	al	al	PROPN
ejpam-6638	451	4	.	.	PUNCT
ejpam-6638	451	5	/	/	SYM
ejpam-6638	451	6	eur	eur	PROPN
ejpam-6638	451	7	.	.	PUNCT
ejpam-6638	452	1	j.	j.	PROPN
ejpam-6638	452	2	pure	pure	PROPN
ejpam-6638	452	3	appl	appl	PROPN
ejpam-6638	452	4	.	.	PROPN
ejpam-6638	452	5	math	math	PROPN
ejpam-6638	452	6	,	,	PUNCT
ejpam-6638	452	7	18	18	NUM
ejpam-6638	452	8	(	(	PUNCT
ejpam-6638	452	9	4	4	NUM
ejpam-6638	452	10	)	)	PUNCT
ejpam-6638	452	11	(	(	PUNCT
ejpam-6638	452	12	2025	2025	NUM
ejpam-6638	452	13	)	)	PUNCT
ejpam-6638	452	14	,	,	PUNCT
ejpam-6638	452	15	6638	6638	NUM
ejpam-6638	452	16	21	21	NUM
ejpam-6638	452	17	of	of	ADP
ejpam-6638	452	18	22	22	NUM
ejpam-6638	452	19	resolution	resolution	NOUN
ejpam-6638	452	20	self	self	NOUN
ejpam-6638	452	21	-	-	PUNCT
ejpam-6638	452	22	attention	attention	NOUN
ejpam-6638	452	23	network	network	NOUN
ejpam-6638	452	24	.	.	PUNCT
ejpam-6638	453	1	information	information	NOUN
ejpam-6638	453	2	sciences	sciences	PROPN
ejpam-6638	453	3	,	,	PUNCT
ejpam-6638	453	4	550:265–277	550:265–277	NUM
ejpam-6638	453	5	,	,	PUNCT
ejpam-6638	453	6	2021	2021	NUM
ejpam-6638	453	7	.	.	PUNCT
ejpam-6638	454	1	[	[	X
ejpam-6638	454	2	36	36	NUM
ejpam-6638	454	3	]	]	PUNCT
ejpam-6638	454	4	m.	m.	PROPN
ejpam-6638	454	5	zhang	zhang	PROPN
ejpam-6638	454	6	,	,	PUNCT
ejpam-6638	454	7	y.	y.	PROPN
ejpam-6638	454	8	zhang	zhang	PROPN
ejpam-6638	454	9	,	,	PUNCT
ejpam-6638	454	10	and	and	CCONJ
ejpam-6638	454	11	y.	y.	PROPN
ejpam-6638	454	12	wang	wang	PROPN
ejpam-6638	454	13	.	.	PUNCT
ejpam-6638	455	1	ms	ms	PROPN
ejpam-6638	455	2	-	-	PUNCT
ejpam-6638	455	3	fusionnet	fusionnet	ADJ
ejpam-6638	455	4	:	:	PUNCT
ejpam-6638	455	5	multi	multi	ADJ
ejpam-6638	455	6	-	-	ADJ
ejpam-6638	455	7	scale	scale	ADJ
ejpam-6638	455	8	fusion	fusion	NOUN
ejpam-6638	455	9	network	network	NOUN
ejpam-6638	455	10	for	for	ADP
ejpam-6638	455	11	image	image	NOUN
ejpam-6638	455	12	inpainting	inpainte	VERB
ejpam-6638	455	13	.	.	PUNCT
ejpam-6638	456	1	mathematical	mathematical	ADJ
ejpam-6638	456	2	biosciences	bioscience	NOUN
ejpam-6638	456	3	and	and	CCONJ
ejpam-6638	456	4	engineering	engineering	NOUN
ejpam-6638	456	5	,	,	PUNCT
ejpam-6638	456	6	18:8280–8297	18:8280–8297	NUM
ejpam-6638	456	7	,	,	PUNCT
ejpam-6638	456	8	2021	2021	NUM
ejpam-6638	456	9	.	.	PUNCT
ejpam-6638	457	1	[	[	X
ejpam-6638	457	2	37	37	NUM
ejpam-6638	457	3	]	]	PUNCT
ejpam-6638	457	4	x.	x.	NOUN
ejpam-6638	457	5	li	li	PROPN
ejpam-6638	457	6	,	,	PUNCT
ejpam-6638	457	7	y.	y.	PROPN
ejpam-6638	457	8	liu	liu	PROPN
ejpam-6638	457	9	,	,	PUNCT
ejpam-6638	457	10	and	and	CCONJ
ejpam-6638	457	11	j.	j.	PROPN
ejpam-6638	457	12	zhao	zhao	PROPN
ejpam-6638	457	13	.	.	PUNCT
ejpam-6638	458	1	structure	structure	NOUN
ejpam-6638	458	2	and	and	CCONJ
ejpam-6638	458	3	semantic	semantic	ADJ
ejpam-6638	458	4	memory	memory	NOUN
ejpam-6638	458	5	guided	guide	VERB
ejpam-6638	458	6	image	image	NOUN
ejpam-6638	458	7	inpainting	inpainte	VERB
ejpam-6638	458	8	via	via	ADP
ejpam-6638	458	9	multi	multi	ADJ
ejpam-6638	458	10	-	-	ADJ
ejpam-6638	458	11	scale	scale	ADJ
ejpam-6638	458	12	cnn	cnn	PROPN
ejpam-6638	458	13	.	.	PUNCT
ejpam-6638	459	1	neurocomputing	neurocomputing	NOUN
ejpam-6638	459	2	,	,	PUNCT
ejpam-6638	459	3	489:168–180	489:168–180	NUM
ejpam-6638	459	4	,	,	PUNCT
ejpam-6638	459	5	2022	2022	NUM
ejpam-6638	459	6	.	.	PUNCT
ejpam-6638	460	1	[	[	X
ejpam-6638	460	2	38	38	NUM
ejpam-6638	460	3	]	]	PUNCT
ejpam-6638	460	4	h.	h.	PROPN
ejpam-6638	460	5	zhou	zhou	PROPN
ejpam-6638	460	6	,	,	PUNCT
ejpam-6638	460	7	j.	j.	PROPN
ejpam-6638	460	8	liu	liu	PROPN
ejpam-6638	460	9	,	,	PUNCT
ejpam-6638	460	10	k.	k.	PROPN
ejpam-6638	460	11	zhang	zhang	PROPN
ejpam-6638	460	12	,	,	PUNCT
ejpam-6638	460	13	and	and	CCONJ
ejpam-6638	460	14	y.	y.	PROPN
ejpam-6638	460	15	pan	pan	PROPN
ejpam-6638	460	16	.	.	PROPN
ejpam-6638	460	17	trans	tran	NOUN
ejpam-6638	460	18	-	-	NOUN
ejpam-6638	460	19	inpaint	inpaint	NOUN
ejpam-6638	460	20	:	:	PUNCT
ejpam-6638	460	21	transformer	transformer	NOUN
ejpam-6638	460	22	-	-	PUNCT
ejpam-6638	460	23	based	base	VERB
ejpam-6638	460	24	high	high	ADJ
ejpam-6638	460	25	-	-	PUNCT
ejpam-6638	460	26	fidelity	fidelity	NOUN
ejpam-6638	460	27	image	image	NOUN
ejpam-6638	460	28	inpainting	inpainte	VERB
ejpam-6638	460	29	via	via	ADP
ejpam-6638	460	30	context	context	NOUN
ejpam-6638	460	31	-	-	PUNCT
ejpam-6638	460	32	aware	aware	ADJ
ejpam-6638	460	33	masked	masked	ADJ
ejpam-6638	460	34	attention	attention	NOUN
ejpam-6638	460	35	.	.	PUNCT
ejpam-6638	461	1	pattern	pattern	NOUN
ejpam-6638	461	2	recognition	recognition	NOUN
ejpam-6638	461	3	,	,	PUNCT
ejpam-6638	461	4	146:110175	146:110175	NUM
ejpam-6638	461	5	,	,	PUNCT
ejpam-6638	461	6	2024	2024	NUM
ejpam-6638	461	7	.	.	PUNCT
ejpam-6638	462	1	[	[	X
ejpam-6638	462	2	39	39	NUM
ejpam-6638	462	3	]	]	PUNCT
ejpam-6638	462	4	h.	h.	PROPN
ejpam-6638	462	5	li	li	PROPN
ejpam-6638	462	6	,	,	PUNCT
ejpam-6638	462	7	l.	l.	PROPN
ejpam-6638	462	8	yuan	yuan	PROPN
ejpam-6638	462	9	,	,	PUNCT
ejpam-6638	462	10	and	and	CCONJ
ejpam-6638	462	11	n.	n.	PROPN
ejpam-6638	462	12	vasconcelos	vasconcelos	PROPN
ejpam-6638	462	13	.	.	PUNCT
ejpam-6638	463	1	revisiting	revisit	VERB
ejpam-6638	463	2	svd	svd	PROPN
ejpam-6638	463	3	for	for	ADP
ejpam-6638	463	4	robust	robust	ADJ
ejpam-6638	463	5	representation	representation	NOUN
ejpam-6638	463	6	learning	learning	NOUN
ejpam-6638	463	7	.	.	PUNCT
ejpam-6638	464	1	in	in	ADP
ejpam-6638	464	2	proceedings	proceeding	NOUN
ejpam-6638	464	3	of	of	ADP
ejpam-6638	464	4	the	the	DET
ejpam-6638	464	5	ieee	ieee	NOUN
ejpam-6638	464	6	/	/	SYM
ejpam-6638	464	7	cvf	cvf	NOUN
ejpam-6638	464	8	conference	conference	NOUN
ejpam-6638	464	9	on	on	ADP
ejpam-6638	464	10	computer	computer	NOUN
ejpam-6638	464	11	vision	vision	NOUN
ejpam-6638	464	12	and	and	CCONJ
ejpam-6638	464	13	pattern	pattern	NOUN
ejpam-6638	464	14	recognition	recognition	NOUN
ejpam-6638	464	15	(	(	PUNCT
ejpam-6638	464	16	cvpr	cvpr	NOUN
ejpam-6638	464	17	)	)	PUNCT
ejpam-6638	464	18	,	,	PUNCT
ejpam-6638	464	19	pages	page	NOUN
ejpam-6638	464	20	12376–12385	12376–12385	NUM
ejpam-6638	464	21	,	,	PUNCT
ejpam-6638	464	22	2021	2021	NUM
ejpam-6638	464	23	.	.	PUNCT
ejpam-6638	465	1	[	[	X
ejpam-6638	465	2	40	40	NUM
ejpam-6638	465	3	]	]	PUNCT
ejpam-6638	465	4	x.	x.	PROPN
ejpam-6638	465	5	chen	chen	PROPN
ejpam-6638	465	6	,	,	PUNCT
ejpam-6638	465	7	y.	y.	PROPN
ejpam-6638	465	8	cao	cao	PROPN
ejpam-6638	465	9	,	,	PUNCT
ejpam-6638	465	10	c.	c.	PROPN
ejpam-6638	465	11	hu	hu	PROPN
ejpam-6638	465	12	,	,	PUNCT
ejpam-6638	465	13	et	et	PROPN
ejpam-6638	465	14	al	al	PROPN
ejpam-6638	465	15	.	.	PUNCT
ejpam-6638	466	1	non	non	ADJ
ejpam-6638	466	2	-	-	ADJ
ejpam-6638	466	3	local	local	ADJ
ejpam-6638	466	4	neural	neural	ADJ
ejpam-6638	466	5	networks	network	NOUN
ejpam-6638	466	6	with	with	ADP
ejpam-6638	466	7	group	group	NOUN
ejpam-6638	466	8	attention	attention	NOUN
ejpam-6638	466	9	.	.	PUNCT
ejpam-6638	467	1	in	in	ADP
ejpam-6638	467	2	proceedings	proceeding	NOUN
ejpam-6638	467	3	of	of	ADP
ejpam-6638	467	4	the	the	DET
ejpam-6638	467	5	aaai	aaai	PROPN
ejpam-6638	467	6	conference	conference	NOUN
ejpam-6638	467	7	on	on	ADP
ejpam-6638	467	8	artificial	artificial	ADJ
ejpam-6638	467	9	intelligence	intelligence	NOUN
ejpam-6638	467	10	,	,	PUNCT
ejpam-6638	467	11	pages	page	NOUN
ejpam-6638	467	12	10504–10511	10504–10511	NUM
ejpam-6638	467	13	,	,	PUNCT
ejpam-6638	467	14	2020	2020	NUM
ejpam-6638	467	15	.	.	PUNCT
ejpam-6638	468	1	[	[	X
ejpam-6638	468	2	41	41	NUM
ejpam-6638	468	3	]	]	PUNCT
ejpam-6638	468	4	d.-a	d.-a	NOUN
ejpam-6638	468	5	.	.	PUNCT
ejpam-6638	469	1	clevert	clevert	PROPN
ejpam-6638	469	2	,	,	PUNCT
ejpam-6638	469	3	t.	t.	PROPN
ejpam-6638	469	4	unterthiner	unterthiner	PROPN
ejpam-6638	469	5	,	,	PUNCT
ejpam-6638	469	6	and	and	CCONJ
ejpam-6638	469	7	s.	s.	PROPN
ejpam-6638	469	8	hochreiter	hochreiter	PROPN
ejpam-6638	469	9	.	.	PUNCT
ejpam-6638	470	1	fast	fast	ADJ
ejpam-6638	470	2	and	and	CCONJ
ejpam-6638	470	3	accurate	accurate	ADJ
ejpam-6638	470	4	deep	deep	ADJ
ejpam-6638	470	5	network	network	NOUN
ejpam-6638	470	6	learning	learn	VERB
ejpam-6638	470	7	by	by	ADP
ejpam-6638	470	8	exponential	exponential	ADJ
ejpam-6638	470	9	linear	linear	PROPN
ejpam-6638	470	10	units	unit	NOUN
ejpam-6638	470	11	(	(	PUNCT
ejpam-6638	470	12	elus	elus	ADJ
ejpam-6638	470	13	)	)	PUNCT
ejpam-6638	470	14	.	.	PUNCT
ejpam-6638	471	1	arxiv	arxiv	PROPN
ejpam-6638	471	2	preprint	preprint	PROPN
ejpam-6638	471	3	arxiv:1511.07289	arxiv:1511.07289	PROPN
ejpam-6638	471	4	,	,	PUNCT
ejpam-6638	471	5	2016	2016	NUM
ejpam-6638	471	6	.	.	PUNCT
ejpam-6638	472	1	[	[	X
ejpam-6638	472	2	42	42	NUM
ejpam-6638	472	3	]	]	X
ejpam-6638	472	4	a.f	a.f	PROPN
ejpam-6638	472	5	.	.	PROPN
ejpam-6638	472	6	agarap	agarap	PROPN
ejpam-6638	472	7	.	.	PUNCT
ejpam-6638	473	1	deep	deep	ADJ
ejpam-6638	473	2	learning	learning	NOUN
ejpam-6638	473	3	using	use	VERB
ejpam-6638	473	4	rectified	rectify	VERB
ejpam-6638	473	5	linear	linear	ADJ
ejpam-6638	473	6	units	unit	NOUN
ejpam-6638	473	7	(	(	PUNCT
ejpam-6638	473	8	relu	relu	NOUN
ejpam-6638	473	9	)	)	PUNCT
ejpam-6638	473	10	.	.	PUNCT
ejpam-6638	474	1	arxiv	arxiv	PROPN
ejpam-6638	474	2	preprint	preprint	NOUN
ejpam-6638	474	3	arxiv:1803.08375	arxiv:1803.08375	NOUN
ejpam-6638	474	4	,	,	PUNCT
ejpam-6638	474	5	2018	2018	NUM
ejpam-6638	474	6	.	.	PUNCT
ejpam-6638	475	1	[	[	X
ejpam-6638	475	2	43	43	NUM
ejpam-6638	475	3	]	]	X
ejpam-6638	475	4	s.w	s.w	PROPN
ejpam-6638	475	5	.	.	PROPN
ejpam-6638	475	6	zamir	zamir	PROPN
ejpam-6638	475	7	,	,	PUNCT
ejpam-6638	475	8	a.	a.	PROPN
ejpam-6638	475	9	arora	arora	PROPN
ejpam-6638	475	10	,	,	PUNCT
ejpam-6638	475	11	s.h	s.h	PROPN
ejpam-6638	475	12	.	.	PROPN
ejpam-6638	475	13	khan	khan	PROPN
ejpam-6638	475	14	,	,	PUNCT
ejpam-6638	475	15	et	et	PROPN
ejpam-6638	475	16	al	al	PROPN
ejpam-6638	475	17	.	.	PUNCT
ejpam-6638	475	18	multi	multi	ADJ
ejpam-6638	475	19	-	-	ADJ
ejpam-6638	475	20	stage	stage	ADJ
ejpam-6638	475	21	progressive	progressive	ADJ
ejpam-6638	475	22	image	image	NOUN
ejpam-6638	475	23	restoration	restoration	NOUN
ejpam-6638	475	24	.	.	PUNCT
ejpam-6638	476	1	in	in	ADP
ejpam-6638	476	2	proceedings	proceeding	NOUN
ejpam-6638	476	3	of	of	ADP
ejpam-6638	476	4	the	the	DET
ejpam-6638	476	5	ieee	ieee	NOUN
ejpam-6638	476	6	/	/	SYM
ejpam-6638	476	7	cvf	cvf	NOUN
ejpam-6638	476	8	conference	conference	NOUN
ejpam-6638	476	9	on	on	ADP
ejpam-6638	476	10	computer	computer	NOUN
ejpam-6638	476	11	vision	vision	NOUN
ejpam-6638	476	12	and	and	CCONJ
ejpam-6638	476	13	pattern	pattern	NOUN
ejpam-6638	476	14	recognition	recognition	NOUN
ejpam-6638	476	15	(	(	PUNCT
ejpam-6638	476	16	cvpr	cvpr	NOUN
ejpam-6638	476	17	)	)	PUNCT
ejpam-6638	476	18	,	,	PUNCT
ejpam-6638	476	19	pages	page	NOUN
ejpam-6638	476	20	14821–14831	14821–14831	NUM
ejpam-6638	476	21	,	,	PUNCT
ejpam-6638	476	22	2021	2021	NUM
ejpam-6638	476	23	.	.	PUNCT
ejpam-6638	477	1	[	[	X
ejpam-6638	477	2	44	44	NUM
ejpam-6638	477	3	]	]	PUNCT
ejpam-6638	477	4	h.	h.	PROPN
ejpam-6638	477	5	liu	liu	PROPN
ejpam-6638	477	6	,	,	PUNCT
ejpam-6638	477	7	x.	x.	PROPN
ejpam-6638	477	8	zhang	zhang	PROPN
ejpam-6638	477	9	,	,	PUNCT
ejpam-6638	477	10	and	and	CCONJ
ejpam-6638	477	11	y.	y.	PROPN
ejpam-6638	477	12	lin	lin	PROPN
ejpam-6638	477	13	.	.	PUNCT
ejpam-6638	478	1	residual	residual	ADJ
ejpam-6638	478	2	non	non	ADJ
ejpam-6638	478	3	-	-	ADJ
ejpam-6638	478	4	local	local	ADJ
ejpam-6638	478	5	attention	attention	NOUN
ejpam-6638	478	6	networks	network	NOUN
ejpam-6638	478	7	for	for	ADP
ejpam-6638	478	8	image	image	NOUN
ejpam-6638	478	9	restoration	restoration	NOUN
ejpam-6638	478	10	.	.	PUNCT
ejpam-6638	479	1	in	in	ADP
ejpam-6638	479	2	proceedings	proceeding	NOUN
ejpam-6638	479	3	of	of	ADP
ejpam-6638	479	4	the	the	DET
ejpam-6638	479	5	ieee	ieee	NOUN
ejpam-6638	479	6	/	/	SYM
ejpam-6638	479	7	cvf	cvf	NOUN
ejpam-6638	479	8	conference	conference	NOUN
ejpam-6638	479	9	on	on	ADP
ejpam-6638	479	10	computer	computer	NOUN
ejpam-6638	479	11	vision	vision	NOUN
ejpam-6638	479	12	and	and	CCONJ
ejpam-6638	479	13	pattern	pattern	NOUN
ejpam-6638	479	14	recognition	recognition	NOUN
ejpam-6638	479	15	workshops	workshop	NOUN
ejpam-6638	479	16	(	(	PUNCT
ejpam-6638	479	17	cvprw	cvprw	PROPN
ejpam-6638	479	18	)	)	PUNCT
ejpam-6638	479	19	,	,	PUNCT
ejpam-6638	479	20	pages	page	NOUN
ejpam-6638	479	21	0–1	0–1	NOUN
ejpam-6638	479	22	,	,	PUNCT
ejpam-6638	479	23	2020	2020	NUM
ejpam-6638	479	24	.	.	PUNCT
ejpam-6638	480	1	[	[	X
ejpam-6638	480	2	45	45	NUM
ejpam-6638	480	3	]	]	PUNCT
ejpam-6638	480	4	c.	c.	PROPN
ejpam-6638	480	5	cheng	cheng	PROPN
ejpam-6638	480	6	,	,	PUNCT
ejpam-6638	480	7	j.b	j.b	PROPN
ejpam-6638	480	8	.	.	PUNCT
ejpam-6638	480	9	lin	lin	PROPN
ejpam-6638	480	10	,	,	PUNCT
ejpam-6638	480	11	z.	z.	PROPN
ejpam-6638	480	12	lin	lin	PROPN
ejpam-6638	480	13	,	,	PUNCT
ejpam-6638	481	1	et	et	PROPN
ejpam-6638	481	2	al	al	PROPN
ejpam-6638	481	3	.	.	PUNCT
ejpam-6638	481	4	apollo	apollo	PROPN
ejpam-6638	481	5	:	:	PUNCT
ejpam-6638	481	6	an	an	DET
ejpam-6638	481	7	adaptive	adaptive	ADJ
ejpam-6638	481	8	parameter	parameter	NOUN
ejpam-6638	481	9	-	-	PUNCT
ejpam-6638	481	10	wise	wise	ADJ
ejpam-6638	481	11	diagonal	diagonal	ADJ
ejpam-6638	481	12	quasinewton	quasinewton	NOUN
ejpam-6638	481	13	method	method	NOUN
ejpam-6638	481	14	for	for	ADP
ejpam-6638	481	15	nonconvex	nonconvex	NOUN
ejpam-6638	481	16	stochastic	stochastic	ADJ
ejpam-6638	481	17	optimization	optimization	NOUN
ejpam-6638	481	18	.	.	PUNCT
ejpam-6638	482	1	arxiv	arxiv	PROPN
ejpam-6638	482	2	preprint	preprint	VERB
ejpam-6638	482	3	arxiv:2203.13819	arxiv:2203.13819	NOUN
ejpam-6638	482	4	,	,	PUNCT
ejpam-6638	482	5	2022	2022	NUM
ejpam-6638	482	6	.	.	PUNCT
ejpam-6638	483	1	[	[	X
ejpam-6638	483	2	46	46	NUM
ejpam-6638	483	3	]	]	PUNCT
ejpam-6638	483	4	z.	z.	PROPN
ejpam-6638	483	5	liu	liu	PROPN
ejpam-6638	483	6	,	,	PUNCT
ejpam-6638	483	7	j.	j.	PROPN
ejpam-6638	483	8	zhang	zhang	PROPN
ejpam-6638	483	9	,	,	PUNCT
ejpam-6638	483	10	and	and	CCONJ
ejpam-6638	483	11	s.	s.	PROPN
ejpam-6638	483	12	zhang	zhang	PROPN
ejpam-6638	483	13	.	.	PUNCT
ejpam-6638	484	1	diffgrad	diffgrad	PROPN
ejpam-6638	484	2	:	:	PUNCT
ejpam-6638	484	3	an	an	DET
ejpam-6638	484	4	optimization	optimization	NOUN
ejpam-6638	484	5	method	method	NOUN
ejpam-6638	484	6	for	for	ADP
ejpam-6638	484	7	convolutional	convolutional	ADJ
ejpam-6638	484	8	neural	neural	ADJ
ejpam-6638	484	9	networks	network	NOUN
ejpam-6638	484	10	.	.	PUNCT
ejpam-6638	485	1	neurocomputing	neurocomputing	NOUN
ejpam-6638	485	2	,	,	PUNCT
ejpam-6638	485	3	393:139–148	393:139–148	NUM
ejpam-6638	485	4	,	,	PUNCT
ejpam-6638	485	5	2020	2020	NUM
ejpam-6638	485	6	.	.	PUNCT
ejpam-6638	486	1	[	[	X
ejpam-6638	486	2	47	47	NUM
ejpam-6638	486	3	]	]	X
ejpam-6638	486	4	a.g	a.g	PROPN
ejpam-6638	486	5	.	.	PROPN
ejpam-6638	486	6	baydin	baydin	NOUN
ejpam-6638	486	7	,	,	PUNCT
ejpam-6638	486	8	r.	r.	PROPN
ejpam-6638	486	9	cornish	cornish	PROPN
ejpam-6638	486	10	,	,	PUNCT
ejpam-6638	486	11	d.	d.	PROPN
ejpam-6638	486	12	martinez	martinez	PROPN
ejpam-6638	486	13	-	-	PROPN
ejpam-6638	486	14	rubio	rubio	PROPN
ejpam-6638	486	15	,	,	PUNCT
ejpam-6638	486	16	et	et	PROPN
ejpam-6638	486	17	al	al	PROPN
ejpam-6638	486	18	.	.	PROPN
ejpam-6638	487	1	online	online	PROPN
ejpam-6638	487	2	learning	learning	NOUN
ejpam-6638	487	3	rate	rate	NOUN
ejpam-6638	487	4	adaptation	adaptation	NOUN
ejpam-6638	487	5	with	with	ADP
ejpam-6638	487	6	hypergradient	hypergradient	ADJ
ejpam-6638	487	7	descent	descent	NOUN
ejpam-6638	487	8	.	.	PUNCT
ejpam-6638	488	1	arxiv	arxiv	PROPN
ejpam-6638	488	2	preprint	preprint	PROPN
ejpam-6638	488	3	arxiv:1703.04782	arxiv:1703.04782	PROPN
ejpam-6638	488	4	,	,	PUNCT
ejpam-6638	488	5	2018	2018	NUM
ejpam-6638	488	6	.	.	PUNCT
ejpam-6638	489	1	[	[	X
ejpam-6638	489	2	48	48	NUM
ejpam-6638	489	3	]	]	PUNCT
ejpam-6638	489	4	per	per	ADP
ejpam-6638	489	5	christian	christian	PROPN
ejpam-6638	489	6	hansen	hansen	PROPN
ejpam-6638	489	7	.	.	PROPN
ejpam-6638	489	8	discrete	discrete	ADJ
ejpam-6638	489	9	inverse	inverse	NOUN
ejpam-6638	489	10	problems	problem	NOUN
ejpam-6638	489	11	:	:	PUNCT
ejpam-6638	489	12	insight	insight	NOUN
ejpam-6638	489	13	and	and	CCONJ
ejpam-6638	489	14	algorithms	algorithm	NOUN
ejpam-6638	489	15	.	.	PUNCT
ejpam-6638	490	1	siam	siam	PROPN
ejpam-6638	490	2	,	,	PUNCT
ejpam-6638	490	3	philadelphia	philadelphia	PROPN
ejpam-6638	490	4	,	,	PUNCT
ejpam-6638	490	5	2010	2010	NUM
ejpam-6638	490	6	.	.	PUNCT
ejpam-6638	491	1	[	[	X
ejpam-6638	491	2	49	49	NUM
ejpam-6638	491	3	]	]	PUNCT
ejpam-6638	491	4	j.	j.	PROPN
ejpam-6638	491	5	xie	xie	PROPN
ejpam-6638	491	6	,	,	PUNCT
ejpam-6638	491	7	l.	l.	PROPN
ejpam-6638	491	8	xu	xu	PROPN
ejpam-6638	491	9	,	,	PUNCT
ejpam-6638	491	10	and	and	CCONJ
ejpam-6638	491	11	e.	e.	PROPN
ejpam-6638	491	12	chen	chen	PROPN
ejpam-6638	491	13	.	.	PUNCT
ejpam-6638	492	1	image	image	NOUN
ejpam-6638	492	2	denoising	denoise	VERB
ejpam-6638	492	3	and	and	CCONJ
ejpam-6638	492	4	inpainting	inpainte	VERB
ejpam-6638	492	5	with	with	ADP
ejpam-6638	492	6	deep	deep	ADJ
ejpam-6638	492	7	neural	neural	ADJ
ejpam-6638	492	8	networks	network	NOUN
ejpam-6638	492	9	.	.	PUNCT
ejpam-6638	493	1	in	in	ADP
ejpam-6638	493	2	advances	advance	NOUN
ejpam-6638	493	3	in	in	ADP
ejpam-6638	493	4	neural	neural	ADJ
ejpam-6638	493	5	information	information	NOUN
ejpam-6638	493	6	processing	processing	NOUN
ejpam-6638	493	7	systems	system	NOUN
ejpam-6638	493	8	,	,	PUNCT
ejpam-6638	493	9	pages	page	NOUN
ejpam-6638	493	10	341–349	341–349	NUM
ejpam-6638	493	11	,	,	PUNCT
ejpam-6638	493	12	2012	2012	NUM
ejpam-6638	493	13	.	.	PUNCT
ejpam-6638	494	1	[	[	X
ejpam-6638	494	2	50	50	NUM
ejpam-6638	494	3	]	]	X
ejpam-6638	494	4	d.p	d.p	PROPN
ejpam-6638	494	5	.	.	PROPN
ejpam-6638	494	6	kingma	kingma	PROPN
ejpam-6638	494	7	and	and	CCONJ
ejpam-6638	494	8	j.	j.	PROPN
ejpam-6638	494	9	ba	ba	PROPN
ejpam-6638	494	10	.	.	PUNCT
ejpam-6638	495	1	adam	adam	PROPN
ejpam-6638	495	2	:	:	PUNCT
ejpam-6638	495	3	a	a	DET
ejpam-6638	495	4	method	method	NOUN
ejpam-6638	495	5	for	for	ADP
ejpam-6638	495	6	stochastic	stochastic	ADJ
ejpam-6638	495	7	optimization	optimization	NOUN
ejpam-6638	495	8	.	.	PUNCT
ejpam-6638	496	1	arxiv	arxiv	PROPN
ejpam-6638	496	2	preprint	preprint	VERB
ejpam-6638	496	3	arxiv:1412.6980	arxiv:1412.6980	NUM
ejpam-6638	496	4	,	,	PUNCT
ejpam-6638	496	5	2014	2014	NUM
ejpam-6638	496	6	.	.	PUNCT
ejpam-6638	497	1	[	[	X
ejpam-6638	497	2	51	51	NUM
ejpam-6638	497	3	]	]	PUNCT
ejpam-6638	497	4	dmitry	dmitry	PROPN
ejpam-6638	497	5	ulyanov	ulyanov	PROPN
ejpam-6638	497	6	,	,	PUNCT
ejpam-6638	497	7	andrea	andrea	PROPN
ejpam-6638	497	8	vedaldi	vedaldi	PROPN
ejpam-6638	497	9	,	,	PUNCT
ejpam-6638	497	10	and	and	CCONJ
ejpam-6638	497	11	victor	victor	NOUN
ejpam-6638	497	12	lempitsky	lempitsky	NOUN
ejpam-6638	497	13	.	.	PUNCT
ejpam-6638	498	1	deep	deep	ADJ
ejpam-6638	498	2	image	image	NOUN
ejpam-6638	498	3	prior	prior	ADV
ejpam-6638	498	4	.	.	PUNCT
ejpam-6638	499	1	in	in	ADP
ejpam-6638	499	2	proceedings	proceeding	NOUN
ejpam-6638	499	3	of	of	ADP
ejpam-6638	499	4	the	the	DET
ejpam-6638	499	5	ieee	ieee	NOUN
ejpam-6638	499	6	conference	conference	NOUN
ejpam-6638	499	7	on	on	ADP
ejpam-6638	499	8	computer	computer	NOUN
ejpam-6638	499	9	vision	vision	NOUN
ejpam-6638	499	10	and	and	CCONJ
ejpam-6638	499	11	pattern	pattern	NOUN
ejpam-6638	499	12	recognition	recognition	NOUN
ejpam-6638	499	13	(	(	PUNCT
ejpam-6638	499	14	cvpr	cvpr	NOUN
ejpam-6638	499	15	)	)	PUNCT
ejpam-6638	499	16	,	,	PUNCT
ejpam-6638	499	17	pages	page	NOUN
ejpam-6638	499	18	9446–9454	9446–9454	NUM
ejpam-6638	499	19	,	,	PUNCT
ejpam-6638	499	20	2018	2018	NUM
ejpam-6638	499	21	.	.	PUNCT
ejpam-6638	500	1	[	[	X
ejpam-6638	500	2	52	52	NUM
ejpam-6638	500	3	]	]	X
ejpam-6638	500	4	c.	c.	PROPN
ejpam-6638	500	5	ma	ma	PROPN
ejpam-6638	500	6	,	,	PUNCT
ejpam-6638	500	7	t.	t.	PROPN
ejpam-6638	500	8	ma	ma	PROPN
ejpam-6638	500	9	,	,	PUNCT
ejpam-6638	500	10	y.	y.	PROPN
ejpam-6638	500	11	chi	chi	PROPN
ejpam-6638	500	12	,	,	PUNCT
ejpam-6638	500	13	and	and	CCONJ
ejpam-6638	500	14	t.	t.	PROPN
ejpam-6638	500	15	yang	yang	PROPN
ejpam-6638	500	16	.	.	PUNCT
ejpam-6638	501	1	apollo	apollo	PROPN
ejpam-6638	501	2	:	:	PUNCT
ejpam-6638	501	3	an	an	DET
ejpam-6638	501	4	adaptive	adaptive	ADJ
ejpam-6638	501	5	parameter	parameter	NOUN
ejpam-6638	501	6	-	-	PUNCT
ejpam-6638	501	7	wise	wise	ADJ
ejpam-6638	501	8	diagonal	diagonal	ADJ
ejpam-6638	501	9	quasinewton	quasinewton	NOUN
ejpam-6638	501	10	method	method	NOUN
ejpam-6638	501	11	for	for	ADP
ejpam-6638	501	12	nonconvex	nonconvex	NOUN
ejpam-6638	501	13	stochastic	stochastic	ADJ
ejpam-6638	501	14	optimization	optimization	NOUN
ejpam-6638	501	15	.	.	PUNCT
ejpam-6638	502	1	in	in	ADP
ejpam-6638	502	2	proceedings	proceeding	NOUN
ejpam-6638	502	3	of	of	ADP
ejpam-6638	502	4	the	the	DET
ejpam-6638	502	5	38th	38th	ADJ
ejpam-6638	502	6	international	international	ADJ
ejpam-6638	502	7	conference	conference	NOUN
ejpam-6638	502	8	on	on	ADP
ejpam-6638	502	9	machine	machine	NOUN
ejpam-6638	502	10	learning	learning	NOUN
ejpam-6638	502	11	(	(	PUNCT
ejpam-6638	502	12	icml	icml	PROPN
ejpam-6638	502	13	)	)	PUNCT
ejpam-6638	502	14	,	,	PUNCT
ejpam-6638	502	15	pages	page	NOUN
ejpam-6638	502	16	7412–7423	7412–7423	NUM
ejpam-6638	502	17	,	,	PUNCT
ejpam-6638	502	18	2021	2021	NUM
ejpam-6638	502	19	.	.	PUNCT
ejpam-6638	503	1	[	[	X
ejpam-6638	503	2	53	53	NUM
ejpam-6638	503	3	]	]	PUNCT
ejpam-6638	503	4	r.	r.	PROPN
ejpam-6638	503	5	dubey	dubey	PROPN
ejpam-6638	503	6	and	and	CCONJ
ejpam-6638	503	7	a.	a.	NOUN
ejpam-6638	503	8	jain	jain	PROPN
ejpam-6638	503	9	.	.	PUNCT
ejpam-6638	504	1	diffgrad	diffgrad	PROPN
ejpam-6638	504	2	:	:	PUNCT
ejpam-6638	504	3	an	an	DET
ejpam-6638	504	4	optimization	optimization	NOUN
ejpam-6638	504	5	method	method	NOUN
ejpam-6638	504	6	for	for	ADP
ejpam-6638	504	7	convolutional	convolutional	ADJ
ejpam-6638	504	8	neural	neural	ADJ
ejpam-6638	504	9	networks	network	NOUN
ejpam-6638	504	10	.	.	PUNCT
ejpam-6638	505	1	arxiv	arxiv	PROPN
ejpam-6638	505	2	preprint	preprint	VERB
ejpam-6638	505	3	arxiv:1909.11015	arxiv:1909.11015	PROPN
ejpam-6638	505	4	,	,	PUNCT
ejpam-6638	505	5	2019	2019	NUM
ejpam-6638	505	6	.	.	PUNCT
ejpam-6638	506	1	m.	m.	NOUN
ejpam-6638	506	2	israr	israr	PROPN
ejpam-6638	506	3	et	et	PROPN
ejpam-6638	506	4	al	al	PROPN
ejpam-6638	506	5	.	.	PUNCT
ejpam-6638	506	6	/	/	SYM
ejpam-6638	506	7	eur	eur	PROPN
ejpam-6638	506	8	.	.	PUNCT
ejpam-6638	507	1	j.	j.	PROPN
ejpam-6638	507	2	pure	pure	PROPN
ejpam-6638	507	3	appl	appl	PROPN
ejpam-6638	507	4	.	.	PROPN
ejpam-6638	507	5	math	math	PROPN
ejpam-6638	507	6	,	,	PUNCT
ejpam-6638	507	7	18	18	NUM
ejpam-6638	507	8	(	(	PUNCT
ejpam-6638	507	9	4	4	NUM
ejpam-6638	507	10	)	)	PUNCT
ejpam-6638	507	11	(	(	PUNCT
ejpam-6638	507	12	2025	2025	NUM
ejpam-6638	507	13	)	)	PUNCT
ejpam-6638	507	14	,	,	PUNCT
ejpam-6638	507	15	6638	6638	NUM
ejpam-6638	507	16	22	22	NUM
ejpam-6638	507	17	of	of	ADP
ejpam-6638	507	18	22	22	NUM
ejpam-6638	507	19	abbreviations	abbreviation	NOUN
ejpam-6638	507	20	abbreviation	abbreviation	NOUN
ejpam-6638	507	21	definition	definition	NOUN
ejpam-6638	507	22	dip	dip	NOUN
ejpam-6638	507	23	deep	deep	ADJ
ejpam-6638	507	24	image	image	NOUN
ejpam-6638	507	25	prior	prior	ADV
ejpam-6638	507	26	svd	svd	PROPN
ejpam-6638	507	27	singular	singular	PROPN
ejpam-6638	507	28	value	value	NOUN
ejpam-6638	507	29	decomposition	decomposition	NOUN
ejpam-6638	507	30	ssim	ssim	NOUN
ejpam-6638	507	31	structural	structural	ADJ
ejpam-6638	507	32	similarity	similarity	NOUN
ejpam-6638	507	33	index	index	NOUN
ejpam-6638	507	34	measure	measure	NOUN
ejpam-6638	507	35	psnr	psnr	PROPN
ejpam-6638	507	36	peak	peak	NOUN
ejpam-6638	507	37	signal	signal	NOUN
ejpam-6638	507	38	-	-	PUNCT
ejpam-6638	507	39	to	to	ADP
ejpam-6638	507	40	-	-	PUNCT
ejpam-6638	507	41	noise	noise	NOUN
ejpam-6638	507	42	ratio	ratio	NOUN
ejpam-6638	507	43	cnn	cnn	PROPN
ejpam-6638	507	44	convolutional	convolutional	ADJ
ejpam-6638	507	45	neural	neural	ADJ
ejpam-6638	507	46	network	network	NOUN
ejpam-6638	507	47	gan	gan	PROPN
ejpam-6638	507	48	generative	generative	ADJ
ejpam-6638	507	49	adversarial	adversarial	ADJ
ejpam-6638	507	50	network	network	NOUN
ejpam-6638	507	51	awgn	awgn	NOUN
ejpam-6638	507	52	additive	additive	ADJ
ejpam-6638	507	53	white	white	ADJ
ejpam-6638	507	54	gaussian	gaussian	PROPN
ejpam-6638	507	55	noise	noise	NOUN
ejpam-6638	507	56	tv	tv	NOUN
ejpam-6638	507	57	total	total	NOUN
ejpam-6638	507	58	variation	variation	NOUN
ejpam-6638	507	59	lr	lr	NOUN
ejpam-6638	507	60	learning	learning	NOUN
ejpam-6638	507	61	rate	rate	NOUN
ejpam-6638	507	62	mse	mse	PROPN
ejpam-6638	507	63	mean	mean	VERB
ejpam-6638	507	64	squared	square	VERB
ejpam-6638	507	65	error	error	NOUN