id	sid	tid	token	lemma	pos
easat-5356	1	1	edelweiss	edelweiss	PROPN
easat-5356	1	2	applied	apply	VERB
easat-5356	1	3	science	science	NOUN
easat-5356	1	4	and	and	CCONJ
easat-5356	1	5	technology	technology	NOUN
easat-5356	1	6	issn	issn	PROPN
easat-5356	1	7	:	:	PUNCT
easat-5356	1	8	2576	2576	NUM
easat-5356	1	9	-	-	SYM
easat-5356	1	10	8484	8484	NUM
easat-5356	1	11	vol	vol	NOUN
easat-5356	1	12	.	.	PROPN
easat-5356	2	1	9	9	NUM
easat-5356	2	2	,	,	PUNCT
easat-5356	2	3	no	no	INTJ
easat-5356	2	4	.	.	NOUN
easat-5356	2	5	3	3	NUM
easat-5356	2	6	,	,	PUNCT
easat-5356	2	7	840	840	NUM
easat-5356	2	8	-	-	SYM
easat-5356	2	9	855	855	NUM
easat-5356	2	10	2025	2025	NUM
easat-5356	2	11	publisher	publisher	NOUN
easat-5356	2	12	:	:	PUNCT
easat-5356	2	13	learning	learn	VERB
easat-5356	2	14	gate	gate	NOUN
easat-5356	2	15	doi	doi	PROPN
easat-5356	2	16	:	:	PUNCT
easat-5356	2	17	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	2	18	©	©	PROPN
easat-5356	2	19	2025	2025	NUM
easat-5356	2	20	by	by	ADP
easat-5356	2	21	the	the	DET
easat-5356	2	22	authors	author	NOUN
easat-5356	2	23	;	;	PUNCT
easat-5356	2	24	licensee	licensee	PROPN
easat-5356	2	25	learning	learning	NOUN
easat-5356	2	26	gate	gate	NOUN
easat-5356	3	1	©	©	PROPN
easat-5356	3	2	2025	2025	NUM
easat-5356	3	3	by	by	ADP
easat-5356	3	4	the	the	DET
easat-5356	3	5	authors	author	NOUN
easat-5356	3	6	;	;	PUNCT
easat-5356	3	7	licensee	licensee	PROPN
easat-5356	3	8	learning	learning	NOUN
easat-5356	3	9	gate	gate	NOUN
easat-5356	3	10	history	history	NOUN
easat-5356	3	11	:	:	PUNCT
easat-5356	3	12	received	receive	VERB
easat-5356	3	13	:	:	PUNCT
easat-5356	3	14	6	6	NUM
easat-5356	3	15	january	january	NOUN
easat-5356	3	16	2025	2025	NUM
easat-5356	3	17	;	;	PUNCT
easat-5356	3	18	revised	revise	VERB
easat-5356	3	19	:	:	PUNCT
easat-5356	3	20	25	25	NUM
easat-5356	3	21	february	february	NOUN
easat-5356	3	22	2025	2025	NUM
easat-5356	3	23	;	;	PUNCT
easat-5356	3	24	accepted	accept	VERB
easat-5356	3	25	:	:	PUNCT
easat-5356	3	26	27	27	NUM
easat-5356	3	27	february	february	NOUN
easat-5356	3	28	2025	2025	NUM
easat-5356	3	29	;	;	PUNCT
easat-5356	3	30	published	publish	VERB
easat-5356	3	31	:	:	PUNCT
easat-5356	4	1	11	11	NUM
easat-5356	4	2	march	march	NOUN
easat-5356	4	3	2025	2025	NUM
easat-5356	4	4	*	*	PUNCT
easat-5356	4	5	correspondence	correspondence	NOUN
easat-5356	4	6	:	:	PUNCT
easat-5356	4	7	chonlatee.p@msu.ac.th	chonlatee.p@msu.ac.th	ADJ
easat-5356	4	8	gcsem	gcsem	ADJ
easat-5356	4	9	-	-	PUNCT
easat-5356	4	10	yolo	yolo	ADJ
easat-5356	4	11	small	small	ADJ
easat-5356	4	12	scale	scale	NOUN
easat-5356	4	13	enhanced	enhance	VERB
easat-5356	4	14	face	face	NOUN
easat-5356	4	15	detector	detector	NOUN
easat-5356	4	16	based	base	VERB
easat-5356	4	17	on	on	ADP
easat-5356	4	18	yolo	yolo	PROPN
easat-5356	4	19	xuwen	xuwen	PROPN
easat-5356	5	1	zheng1	zheng1	PROPN
easat-5356	5	2	,	,	PUNCT
easat-5356	5	3	zhiwei	zhiwei	PROPN
easat-5356	5	4	zhou2	zhou2	PROPN
easat-5356	5	5	,	,	PUNCT
easat-5356	5	6	chonlatee	chonlatee	PROPN
easat-5356	5	7	photong3	photong3	PROPN
easat-5356	5	8	*	*	PUNCT
easat-5356	5	9	1,3faculty	1,3faculty	NUM
easat-5356	5	10	of	of	ADP
easat-5356	5	11	engineering	engineering	NOUN
easat-5356	5	12	mahasarakham	mahasarakham	PROPN
easat-5356	5	13	university	university	PROPN
easat-5356	5	14	,	,	PUNCT
easat-5356	5	15	maha	maha	PROPN
easat-5356	5	16	sarakham	sarakham	PROPN
easat-5356	5	17	,	,	PUNCT
easat-5356	5	18	44150	44150	NUM
easat-5356	5	19	,	,	PUNCT
easat-5356	5	20	thailand	thailand	PROPN
easat-5356	5	21	;	;	PUNCT
easat-5356	5	22	2240053955@qq.com	2240053955@qq.com	NUM
easat-5356	5	23	(	(	PUNCT
easat-5356	5	24	x.z	x.z	PROPN
easat-5356	5	25	.	.	PUNCT
easat-5356	5	26	)	)	PUNCT
easat-5356	6	1	chonlatee.p@msu.ac.th	chonlatee.p@msu.ac.th	PROPN
easat-5356	6	2	(	(	PUNCT
easat-5356	6	3	c.p	c.p	PROPN
easat-5356	6	4	.	.	PROPN
easat-5356	6	5	)	)	PUNCT
easat-5356	6	6	.	.	PUNCT
easat-5356	7	1	2hunan	2hunan	NUM
easat-5356	7	2	mechanical	mechanical	ADJ
easat-5356	7	3	&	&	CCONJ
easat-5356	7	4	electrical	electrical	ADJ
easat-5356	7	5	polytechnic	polytechnic	ADJ
easat-5356	7	6	,	,	PUNCT
easat-5356	7	7	changsha	changsha	PROPN
easat-5356	7	8	,	,	PUNCT
easat-5356	7	9	410151	410151	NUM
easat-5356	7	10	,	,	PUNCT
easat-5356	7	11	hunan	hunan	PROPN
easat-5356	7	12	,	,	PUNCT
easat-5356	7	13	china	china	PROPN
easat-5356	7	14	;	;	PUNCT
easat-5356	7	15	65010363010@msu.ac.th	65010363010@msu.ac.th	NUM
easat-5356	7	16	(	(	PUNCT
easat-5356	7	17	c.p	c.p	PROPN
easat-5356	7	18	.	.	PROPN
easat-5356	7	19	)	)	PUNCT
easat-5356	7	20	.	.	PUNCT
easat-5356	8	1	abstract	abstract	ADJ
easat-5356	8	2	:	:	PUNCT
easat-5356	8	3	face	face	NOUN
easat-5356	8	4	detection	detection	NOUN
easat-5356	8	5	is	be	AUX
easat-5356	8	6	a	a	DET
easat-5356	8	7	crucial	crucial	ADJ
easat-5356	8	8	aspect	aspect	NOUN
easat-5356	8	9	of	of	ADP
easat-5356	8	10	computer	computer	NOUN
easat-5356	8	11	vision	vision	NOUN
easat-5356	8	12	,	,	PUNCT
easat-5356	8	13	often	often	ADV
easat-5356	8	14	challenged	challenge	VERB
easat-5356	8	15	by	by	ADP
easat-5356	8	16	factors	factor	NOUN
easat-5356	8	17	such	such	ADJ
easat-5356	8	18	as	as	ADP
easat-5356	8	19	varying	vary	VERB
easat-5356	8	20	scales	scale	NOUN
easat-5356	8	21	,	,	PUNCT
easat-5356	8	22	occlusions	occlusion	NOUN
easat-5356	8	23	,	,	PUNCT
easat-5356	8	24	and	and	CCONJ
easat-5356	8	25	diverse	diverse	VERB
easat-5356	8	26	facial	facial	ADJ
easat-5356	8	27	features	feature	NOUN
easat-5356	8	28	.	.	PUNCT
easat-5356	9	1	in	in	ADP
easat-5356	9	2	this	this	DET
easat-5356	9	3	study	study	NOUN
easat-5356	9	4	,	,	PUNCT
easat-5356	9	5	we	we	PRON
easat-5356	9	6	introduce	introduce	VERB
easat-5356	9	7	gcsem	gcsem	ADJ
easat-5356	9	8	-	-	PUNCT
easat-5356	9	9	yolo	yolo	PROPN
easat-5356	9	10	,	,	PUNCT
easat-5356	9	11	an	an	DET
easat-5356	9	12	innovative	innovative	ADJ
easat-5356	9	13	real	real	ADJ
easat-5356	9	14	-	-	PUNCT
easat-5356	9	15	time	time	NOUN
easat-5356	9	16	face	face	NOUN
easat-5356	9	17	detection	detection	NOUN
easat-5356	9	18	method	method	NOUN
easat-5356	9	19	built	build	VERB
easat-5356	9	20	upon	upon	SCONJ
easat-5356	9	21	the	the	DET
easat-5356	9	22	yolov8	yolov8	NOUN
easat-5356	9	23	architecture	architecture	NOUN
easat-5356	9	24	.	.	PUNCT
easat-5356	10	1	this	this	DET
easat-5356	10	2	approach	approach	NOUN
easat-5356	10	3	incorporates	incorporate	VERB
easat-5356	10	4	a	a	DET
easat-5356	10	5	novel	novel	ADJ
easat-5356	10	6	feature	feature	NOUN
easat-5356	10	7	extraction	extraction	NOUN
easat-5356	10	8	module	module	NOUN
easat-5356	10	9	(	(	PUNCT
easat-5356	10	10	gcsem	gcsem	NOUN
easat-5356	10	11	)	)	PUNCT
easat-5356	10	12	alongside	alongside	ADP
easat-5356	10	13	a	a	DET
easat-5356	10	14	specialized	specialized	ADJ
easat-5356	10	15	small	small	ADJ
easat-5356	10	16	-	-	PUNCT
easat-5356	10	17	scale	scale	NOUN
easat-5356	10	18	detection	detection	NOUN
easat-5356	10	19	head	head	NOUN
easat-5356	10	20	,	,	PUNCT
easat-5356	10	21	designed	design	VERB
easat-5356	10	22	to	to	PART
easat-5356	10	23	capture	capture	VERB
easat-5356	10	24	pixel	pixel	ADJ
easat-5356	10	25	information	information	NOUN
easat-5356	10	26	across	across	ADP
easat-5356	10	27	multiple	multiple	ADJ
easat-5356	10	28	levels	level	NOUN
easat-5356	10	29	and	and	CCONJ
easat-5356	10	30	enhance	enhance	VERB
easat-5356	10	31	the	the	DET
easat-5356	10	32	receptive	receptive	ADJ
easat-5356	10	33	field	field	NOUN
easat-5356	10	34	,	,	PUNCT
easat-5356	10	35	thereby	thereby	ADV
easat-5356	10	36	improving	improve	VERB
easat-5356	10	37	the	the	DET
easat-5356	10	38	accuracy	accuracy	NOUN
easat-5356	10	39	of	of	ADP
easat-5356	10	40	small	small	ADJ
easat-5356	10	41	face	face	NOUN
easat-5356	10	42	detection	detection	NOUN
easat-5356	10	43	.	.	PUNCT
easat-5356	11	1	to	to	PART
easat-5356	11	2	address	address	VERB
easat-5356	11	3	the	the	DET
easat-5356	11	4	imbalance	imbalance	NOUN
easat-5356	11	5	between	between	ADP
easat-5356	11	6	easy	easy	ADJ
easat-5356	11	7	and	and	CCONJ
easat-5356	11	8	difficult	difficult	ADJ
easat-5356	11	9	samples	sample	NOUN
easat-5356	11	10	,	,	PUNCT
easat-5356	11	11	we	we	PRON
easat-5356	11	12	employ	employ	VERB
easat-5356	11	13	an	an	DET
easat-5356	11	14	adaptive	adaptive	ADJ
easat-5356	11	15	anchor	anchor	NOUN
easat-5356	11	16	box	box	NOUN
easat-5356	11	17	filtering	filtering	NOUN
easat-5356	11	18	algorithm	algorithm	NOUN
easat-5356	11	19	coupled	couple	VERB
easat-5356	11	20	with	with	ADP
easat-5356	11	21	the	the	DET
easat-5356	11	22	new	new	ADJ
easat-5356	11	23	wiouv3	wiouv3	NOUN
easat-5356	11	24	loss	loss	NOUN
easat-5356	11	25	function	function	NOUN
easat-5356	11	26	.	.	PUNCT
easat-5356	12	1	experimental	experimental	ADJ
easat-5356	12	2	results	result	NOUN
easat-5356	12	3	demonstrate	demonstrate	VERB
easat-5356	12	4	that	that	SCONJ
easat-5356	12	5	gcsem	gcsem	NOUN
easat-5356	12	6	-	-	PUNCT
easat-5356	12	7	yolo	yolo	PROPN
easat-5356	12	8	achieves	achieve	VERB
easat-5356	12	9	outstanding	outstanding	ADJ
easat-5356	12	10	efficiency	efficiency	NOUN
easat-5356	12	11	on	on	ADP
easat-5356	12	12	the	the	DET
easat-5356	12	13	wider	wide	ADJ
easat-5356	12	14	face	face	NOUN
easat-5356	12	15	validation	validation	NOUN
easat-5356	12	16	datasets	dataset	NOUN
easat-5356	12	17	.	.	PUNCT
easat-5356	13	1	keywords	keyword	NOUN
easat-5356	13	2	:	:	PUNCT
easat-5356	13	3	face	face	NOUN
easat-5356	13	4	recognition	recognition	NOUN
easat-5356	13	5	,	,	PUNCT
easat-5356	13	6	gcsem	gcsem	NOUN
easat-5356	13	7	-	-	PUNCT
easat-5356	13	8	yolo	yolo	ADJ
easat-5356	13	9	,	,	PUNCT
easat-5356	13	10	small	small	ADJ
easat-5356	13	11	scale	scale	NOUN
easat-5356	13	12	,	,	PUNCT
easat-5356	13	13	yolo	yolo	ADJ
easat-5356	13	14	.	.	PROPN
easat-5356	14	1	1	1	X
easat-5356	14	2	.	.	X
easat-5356	14	3	introduction	introduction	NOUN
easat-5356	14	4	face	face	NOUN
easat-5356	14	5	detection	detection	NOUN
easat-5356	14	6	serves	serve	VERB
easat-5356	14	7	as	as	ADP
easat-5356	14	8	a	a	DET
easat-5356	14	9	crucial	crucial	ADJ
easat-5356	14	10	foundation	foundation	NOUN
easat-5356	14	11	for	for	ADP
easat-5356	14	12	enabling	enable	VERB
easat-5356	14	13	a	a	DET
easat-5356	14	14	variety	variety	NOUN
easat-5356	14	15	of	of	ADP
easat-5356	14	16	facial	facial	ADJ
easat-5356	14	17	analysis	analysis	NOUN
easat-5356	14	18	tasks	task	NOUN
easat-5356	14	19	,	,	PUNCT
easat-5356	14	20	particularly	particularly	ADV
easat-5356	14	21	facial	facial	ADJ
easat-5356	14	22	alignment	alignment	NOUN
easat-5356	14	23	meher	meher	NOUN
easat-5356	14	24	,	,	PUNCT
easat-5356	14	25	et	et	PROPN
easat-5356	14	26	al	al	PROPN
easat-5356	14	27	.	.	PUNCT
easat-5356	15	1	[	[	X
easat-5356	15	2	1	1	NUM
easat-5356	15	3	]	]	PUNCT
easat-5356	15	4	attribute	attribute	NOUN
easat-5356	15	5	analysis	analysis	NOUN
easat-5356	15	6	dey	dey	PROPN
easat-5356	15	7	,	,	PUNCT
easat-5356	15	8	et	et	PROPN
easat-5356	15	9	al	al	PROPN
easat-5356	15	10	.	.	PUNCT
easat-5356	16	1	[	[	X
easat-5356	16	2	2	2	NUM
easat-5356	16	3	]	]	PUNCT
easat-5356	16	4	recognition	recognition	NOUN
easat-5356	16	5	du	du	PROPN
easat-5356	16	6	,	,	PUNCT
easat-5356	16	7	et	et	PROPN
easat-5356	16	8	al	al	PROPN
easat-5356	16	9	.	.	PUNCT
easat-5356	17	1	[	[	X
easat-5356	17	2	3	3	X
easat-5356	17	3	]	]	PUNCT
easat-5356	17	4	and	and	CCONJ
easat-5356	17	5	verification	verification	NOUN
easat-5356	17	6	[	[	X
easat-5356	17	7	4	4	NUM
easat-5356	17	8	]	]	PUNCT
easat-5356	17	9	.	.	PUNCT
easat-5356	18	1	the	the	DET
easat-5356	18	2	performance	performance	NOUN
easat-5356	18	3	of	of	ADP
easat-5356	18	4	facial	facial	ADJ
easat-5356	18	5	detectors	detector	NOUN
easat-5356	18	6	has	have	AUX
easat-5356	18	7	significantly	significantly	ADV
easat-5356	18	8	increased	increase	VERB
easat-5356	18	9	with	with	ADP
easat-5356	18	10	the	the	DET
easat-5356	18	11	recent	recent	ADJ
easat-5356	18	12	explosion	explosion	NOUN
easat-5356	18	13	in	in	ADP
easat-5356	18	14	deep	deep	ADJ
easat-5356	18	15	convolutional	convolutional	ADJ
easat-5356	18	16	neural	neural	ADJ
easat-5356	18	17	networks	network	NOUN
easat-5356	18	18	.	.	PUNCT
easat-5356	19	1	several	several	ADJ
easat-5356	19	2	deep	deep	ADJ
easat-5356	19	3	learning	learning	NOUN
easat-5356	19	4	-	-	PUNCT
easat-5356	19	5	based	base	VERB
easat-5356	19	6	high	high	ADJ
easat-5356	19	7	-	-	PUNCT
easat-5356	19	8	performance	performance	NOUN
easat-5356	19	9	facial	facial	ADJ
easat-5356	19	10	detection	detection	NOUN
easat-5356	19	11	systems	system	NOUN
easat-5356	19	12	have	have	AUX
easat-5356	19	13	been	be	AUX
easat-5356	19	14	put	put	VERB
easat-5356	19	15	forth	forth	ADV
easat-5356	19	16	.	.	PUNCT
easat-5356	20	1	these	these	DET
easat-5356	20	2	algorithms	algorithm	NOUN
easat-5356	20	3	often	often	ADV
easat-5356	20	4	fall	fall	VERB
easat-5356	20	5	into	into	ADP
easat-5356	20	6	two	two	NUM
easat-5356	20	7	major	major	ADJ
easat-5356	20	8	categories	category	NOUN
easat-5356	20	9	.	.	PUNCT
easat-5356	21	1	one	one	NUM
easat-5356	21	2	category	category	NOUN
easat-5356	21	3	is	be	AUX
easat-5356	21	4	cascaded	cascade	VERB
easat-5356	21	5	neural	neural	ADJ
easat-5356	21	6	nets	net	NOUN
easat-5356	21	7	.	.	PUNCT
easat-5356	22	1	the	the	DET
easat-5356	22	2	face	face	NOUN
easat-5356	22	3	detection	detection	NOUN
easat-5356	22	4	technique	technique	NOUN
easat-5356	22	5	detects	detect	NOUN
easat-5356	22	6	coarse	coarse	ADV
easat-5356	22	7	to	to	ADP
easat-5356	22	8	fine	fine	ADJ
easat-5356	22	9	faces	face	NOUN
easat-5356	22	10	by	by	ADP
easat-5356	22	11	using	use	VERB
easat-5356	22	12	cascaded	cascade	VERB
easat-5356	22	13	neural	neural	ADJ
easat-5356	22	14	networks	network	NOUN
easat-5356	22	15	as	as	ADP
easat-5356	22	16	classifiers	classifier	NOUN
easat-5356	22	17	and	and	CCONJ
easat-5356	22	18	feature	feature	NOUN
easat-5356	22	19	extractors	extractor	NOUN
easat-5356	22	20	.	.	PUNCT
easat-5356	23	1	although	although	SCONJ
easat-5356	23	2	cascade	cascade	NOUN
easat-5356	23	3	detectors	detector	NOUN
easat-5356	23	4	have	have	AUX
easat-5356	23	5	achieved	achieve	VERB
easat-5356	23	6	notable	notable	ADJ
easat-5356	23	7	success	success	NOUN
easat-5356	23	8	,	,	PUNCT
easat-5356	23	9	they	they	PRON
easat-5356	23	10	also	also	ADV
easat-5356	23	11	present	present	VERB
easat-5356	23	12	certain	certain	ADJ
easat-5356	23	13	limitations	limitation	NOUN
easat-5356	23	14	,	,	PUNCT
easat-5356	23	15	including	include	VERB
easat-5356	23	16	challenges	challenge	NOUN
easat-5356	23	17	in	in	ADP
easat-5356	23	18	training	training	NOUN
easat-5356	23	19	and	and	CCONJ
easat-5356	23	20	slower	slow	ADJ
easat-5356	23	21	detection	detection	NOUN
easat-5356	23	22	speeds	speed	NOUN
easat-5356	23	23	.	.	PUNCT
easat-5356	24	1	another	another	DET
easat-5356	24	2	class	class	NOUN
easat-5356	24	3	of	of	ADP
easat-5356	24	4	algorithms	algorithm	NOUN
easat-5356	24	5	stems	stem	VERB
easat-5356	24	6	from	from	ADP
easat-5356	24	7	improvements	improvement	NOUN
easat-5356	24	8	to	to	ADP
easat-5356	24	9	generalized	generalized	ADJ
easat-5356	24	10	target	target	NOUN
easat-5356	24	11	detection	detection	NOUN
easat-5356	24	12	algorithms	algorithm	NOUN
easat-5356	24	13	.	.	PUNCT
easat-5356	25	1	with	with	ADP
easat-5356	25	2	the	the	DET
easat-5356	25	3	use	use	NOUN
easat-5356	25	4	of	of	ADP
easat-5356	25	5	specialized	specialized	ADJ
easat-5356	25	6	architectural	architectural	ADJ
easat-5356	25	7	designs	design	NOUN
easat-5356	25	8	,	,	PUNCT
easat-5356	25	9	task	task	NOUN
easat-5356	25	10	-	-	PUNCT
easat-5356	25	11	specific	specific	ADJ
easat-5356	25	12	models	model	NOUN
easat-5356	25	13	can	can	AUX
easat-5356	25	14	improve	improve	VERB
easat-5356	25	15	performance	performance	NOUN
easat-5356	25	16	by	by	ADP
easat-5356	25	17	utilizing	utilize	VERB
easat-5356	25	18	the	the	DET
easat-5356	25	19	more	more	ADV
easat-5356	25	20	universal	universal	ADJ
easat-5356	25	21	object	object	NOUN
easat-5356	25	22	properties	property	NOUN
easat-5356	25	23	that	that	PRON
easat-5356	25	24	these	these	DET
easat-5356	25	25	detectors	detector	NOUN
easat-5356	25	26	are	be	AUX
easat-5356	25	27	made	make	VERB
easat-5356	25	28	to	to	PART
easat-5356	25	29	capture	capture	VERB
easat-5356	25	30	.	.	PUNCT
easat-5356	26	1	prominent	prominent	ADJ
easat-5356	26	2	face	face	NOUN
easat-5356	26	3	detectors	detector	NOUN
easat-5356	26	4	like	like	ADP
easat-5356	26	5	yolo	yolo	PRON
easat-5356	27	1	[	[	X
easat-5356	27	2	5	5	NUM
easat-5356	27	3	]	]	PUNCT
easat-5356	27	4	.	.	PUNCT
easat-5356	28	1	faster	fast	ADV
easat-5356	28	2	r	r	VERB
easat-5356	28	3	-	-	PUNCT
easat-5356	28	4	cnn	cnn	PROPN
easat-5356	28	5	nogales	nogale	NOUN
easat-5356	28	6	,	,	PUNCT
easat-5356	28	7	et	et	PROPN
easat-5356	28	8	al	al	PROPN
easat-5356	28	9	.	.	PUNCT
easat-5356	29	1	[	[	X
easat-5356	29	2	6	6	NUM
easat-5356	29	3	]	]	PUNCT
easat-5356	29	4	and	and	CCONJ
easat-5356	29	5	retinanet	retinanet	PROPN
easat-5356	29	6	exemplify	exemplify	VERB
easat-5356	29	7	this	this	DET
easat-5356	29	8	approach	approach	NOUN
easat-5356	29	9	.	.	PUNCT
easat-5356	30	1	in	in	ADP
easat-5356	30	2	the	the	DET
easat-5356	30	3	research	research	NOUN
easat-5356	30	4	,	,	PUNCT
easat-5356	30	5	inspired	inspire	VERB
easat-5356	30	6	by	by	ADP
easat-5356	30	7	yolov5	yolov5	NOUN
easat-5356	30	8	[	[	X
easat-5356	30	9	7	7	NUM
easat-5356	30	10	]	]	PUNCT
easat-5356	30	11	.	.	PUNCT
easat-5356	31	1	they	they	PRON
easat-5356	31	2	present	present	VERB
easat-5356	31	3	gcsem	gcsem	PROPN
easat-5356	31	4	-	-	PUNCT
easat-5356	31	5	yolo	yolo	PROPN
easat-5356	31	6	,	,	PUNCT
easat-5356	31	7	a	a	DET
easat-5356	31	8	revolutionary	revolutionary	ADJ
easat-5356	31	9	face	face	NOUN
easat-5356	31	10	detector	detector	NOUN
easat-5356	31	11	that	that	PRON
easat-5356	31	12	reaches	reach	VERB
easat-5356	31	13	cutting	cut	VERB
easat-5356	31	14	-	-	PUNCT
easat-5356	31	15	edge	edge	NOUN
easat-5356	31	16	results	result	NOUN
easat-5356	31	17	in	in	ADP
easat-5356	31	18	single	single	ADJ
easat-5356	31	19	-	-	PUNCT
easat-5356	31	20	stage	stage	NOUN
easat-5356	31	21	face	face	NOUN
easat-5356	31	22	detection	detection	NOUN
easat-5356	31	23	.	.	PUNCT
easat-5356	32	1	even	even	ADV
easat-5356	32	2	though	though	SCONJ
easat-5356	32	3	face	face	NOUN
easat-5356	32	4	detection	detection	NOUN
easat-5356	32	5	performance	performance	NOUN
easat-5356	32	6	has	have	AUX
easat-5356	32	7	significantly	significantly	ADV
easat-5356	32	8	improved	improve	VERB
easat-5356	32	9	thanks	thank	NOUN
easat-5356	32	10	to	to	ADP
easat-5356	32	11	deep	deep	ADJ
easat-5356	32	12	convolutional	convolutional	ADJ
easat-5356	32	13	networks	network	NOUN
easat-5356	32	14	,	,	PUNCT
easat-5356	32	15	it	it	PRON
easat-5356	32	16	is	be	AUX
easat-5356	32	17	still	still	ADV
easat-5356	32	18	difficult	difficult	ADJ
easat-5356	32	19	to	to	PART
easat-5356	32	20	distinguish	distinguish	VERB
easat-5356	32	21	faces	face	NOUN
easat-5356	32	22	with	with	ADP
easat-5356	32	23	large	large	ADJ
easat-5356	32	24	scale	scale	NOUN
easat-5356	32	25	differences	difference	NOUN
easat-5356	32	26	in	in	ADP
easat-5356	32	27	real	real	ADJ
easat-5356	32	28	-	-	PUNCT
easat-5356	32	29	world	world	NOUN
easat-5356	32	30	situations	situation	NOUN
easat-5356	32	31	.	.	PUNCT
easat-5356	33	1	unlike	unlike	ADP
easat-5356	33	2	general	general	ADJ
easat-5356	33	3	-	-	PUNCT
easat-5356	33	4	purpose	purpose	NOUN
easat-5356	33	5	object	object	NOUN
easat-5356	33	6	detection	detection	NOUN
easat-5356	33	7	,	,	PUNCT
easat-5356	33	8	face	face	NOUN
easat-5356	33	9	detection	detection	NOUN
easat-5356	33	10	involves	involve	VERB
easat-5356	33	11	relatively	relatively	ADV
easat-5356	33	12	small	small	ADJ
easat-5356	33	13	aspect	aspect	NOUN
easat-5356	33	14	ratio	ratio	NOUN
easat-5356	33	15	variations	variation	NOUN
easat-5356	33	16	(	(	PUNCT
easat-5356	33	17	typically	typically	ADV
easat-5356	33	18	between	between	ADP
easat-5356	33	19	1:1	1:1	NUM
easat-5356	33	20	and	and	CCONJ
easat-5356	33	21	1:1.5	1:1.5	NUM
easat-5356	33	22	)	)	PUNCT
easat-5356	33	23	but	but	CCONJ
easat-5356	33	24	substantial	substantial	ADJ
easat-5356	33	25	scale	scale	NOUN
easat-5356	33	26	variations	variation	NOUN
easat-5356	33	27	,	,	PUNCT
easat-5356	33	28	with	with	ADP
easat-5356	33	29	face	face	NOUN
easat-5356	33	30	sizes	size	NOUN
easat-5356	33	31	ranging	range	VERB
easat-5356	33	32	from	from	ADP
easat-5356	33	33	a	a	DET
easat-5356	33	34	few	few	ADJ
easat-5356	33	35	pixels	pixel	NOUN
easat-5356	33	36	to	to	ADP
easat-5356	33	37	thousands	thousand	NOUN
easat-5356	33	38	of	of	ADP
easat-5356	33	39	pixels	pixel	NOUN
easat-5356	33	40	[	[	X
easat-5356	33	41	8	8	NUM
easat-5356	33	42	]	]	PUNCT
easat-5356	33	43	.	.	PUNCT
easat-5356	34	1	this	this	DET
easat-5356	34	2	continuous	continuous	ADJ
easat-5356	34	3	variation	variation	NOUN
easat-5356	34	4	in	in	ADP
easat-5356	34	5	scale	scale	NOUN
easat-5356	34	6	complicates	complicate	VERB
easat-5356	34	7	adaptation	adaptation	NOUN
easat-5356	34	8	for	for	ADP
easat-5356	34	9	discrete	discrete	ADJ
easat-5356	34	10	anchor	anchor	NOUN
easat-5356	34	11	-	-	PUNCT
easat-5356	34	12	based	base	VERB
easat-5356	34	13	algorithms	algorithm	NOUN
easat-5356	34	14	,	,	PUNCT
easat-5356	34	15	making	make	VERB
easat-5356	34	16	it	it	PRON
easat-5356	34	17	difficult	difficult	ADJ
easat-5356	34	18	to	to	PART
easat-5356	34	19	accurately	accurately	ADV
easat-5356	34	20	detect	detect	VERB
easat-5356	34	21	faces	face	NOUN
easat-5356	34	22	across	across	ADP
easat-5356	34	23	diverse	diverse	ADJ
easat-5356	34	24	scales	scale	NOUN
easat-5356	34	25	.	.	PUNCT
easat-5356	35	1	to	to	PART
easat-5356	35	2	tackle	tackle	VERB
easat-5356	35	3	this	this	DET
easat-5356	35	4	issue	issue	NOUN
easat-5356	35	5	,	,	PUNCT
easat-5356	35	6	scholars	scholar	NOUN
easat-5356	35	7	have	have	AUX
easat-5356	35	8	suggested	suggest	VERB
easat-5356	35	9	various	various	ADJ
easat-5356	35	10	approaches.du	approaches.du	NUM
easat-5356	35	11	et	et	PROPN
easat-5356	35	12	al	al	PROPN
easat-5356	35	13	.	.	PROPN
easat-5356	35	14	proposed	propose	VERB
easat-5356	35	15	yolo	yolo	PROPN
easat-5356	35	16	-	-	PUNCT
easat-5356	35	17	face	face	NOUN
easat-5356	35	18	du	du	PROPN
easat-5356	35	19	,	,	PUNCT
easat-5356	35	20	et	et	PROPN
easat-5356	35	21	al	al	PROPN
easat-5356	35	22	.	.	PUNCT
easat-5356	36	1	[	[	X
easat-5356	36	2	3	3	X
easat-5356	36	3	]	]	PUNCT
easat-5356	36	4	an	an	DET
easat-5356	36	5	enhanced	enhanced	ADJ
easat-5356	36	6	face	face	NOUN
easat-5356	36	7	detector	detector	NOUN
easat-5356	36	8	based	base	VERB
easat-5356	36	9	on	on	ADP
easat-5356	36	10	yolov3	yolov3	PROPN
easat-5356	36	11	that	that	PRON
easat-5356	36	12	focuses	focus	VERB
easat-5356	36	13	on	on	ADP
easat-5356	36	14	scale	scale	NOUN
easat-5356	36	15	variation	variation	NOUN
easat-5356	36	16	.	.	PUNCT
easat-5356	37	1	the	the	DET
easat-5356	37	2	method	method	NOUN
easat-5356	37	3	involves	involve	VERB
easat-5356	37	4	using	use	VERB
easat-5356	37	5	a	a	DET
easat-5356	37	6	more	more	ADV
easat-5356	37	7	accurate	accurate	ADJ
easat-5356	37	8	regression	regression	NOUN
easat-5356	37	9	loss	loss	NOUN
easat-5356	37	10	function	function	NOUN
easat-5356	37	11	and	and	CCONJ
easat-5356	37	12	creating	create	VERB
easat-5356	37	13	anchor	anchor	NOUN
easat-5356	37	14	scales	scale	NOUN
easat-5356	37	15	that	that	PRON
easat-5356	37	16	are	be	AUX
easat-5356	37	17	especially	especially	ADV
easat-5356	37	18	suited	suit	VERB
easat-5356	37	19	for	for	ADP
easat-5356	37	20	human	human	ADJ
easat-5356	37	21	841	841	NUM
easat-5356	37	22	edelweiss	edelweiss	PROPN
easat-5356	37	23	applied	apply	VERB
easat-5356	37	24	science	science	NOUN
easat-5356	37	25	and	and	CCONJ
easat-5356	37	26	technology	technology	NOUN
easat-5356	37	27	issn	issn	PROPN
easat-5356	37	28	:	:	PUNCT
easat-5356	37	29	2576	2576	NUM
easat-5356	37	30	-	-	SYM
easat-5356	37	31	8484	8484	NUM
easat-5356	37	32	vol	vol	NOUN
easat-5356	37	33	.	.	PROPN
easat-5356	38	1	9	9	NUM
easat-5356	38	2	,	,	PUNCT
easat-5356	38	3	no	no	INTJ
easat-5356	38	4	.	.	NOUN
easat-5356	39	1	3	3	NUM
easat-5356	39	2	:	:	PUNCT
easat-5356	39	3	840	840	NUM
easat-5356	39	4	-	-	SYM
easat-5356	39	5	855	855	NUM
easat-5356	39	6	,	,	PUNCT
easat-5356	39	7	2025	2025	NUM
easat-5356	39	8	doi	doi	NOUN
easat-5356	39	9	:	:	PUNCT
easat-5356	39	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	39	11	©	©	PROPN
easat-5356	39	12	2025	2025	NUM
easat-5356	39	13	by	by	ADP
easat-5356	39	14	the	the	DET
easat-5356	39	15	authors	author	NOUN
easat-5356	39	16	;	;	PUNCT
easat-5356	39	17	licensee	licensee	PROPN
easat-5356	39	18	learning	learn	VERB
easat-5356	39	19	gate	gate	NOUN
easat-5356	39	20	faces	face	NOUN
easat-5356	39	21	.	.	PUNCT
easat-5356	40	1	for	for	ADP
easat-5356	40	2	example	example	NOUN
easat-5356	40	3	,	,	PUNCT
easat-5356	40	4	hossain	hossain	PROPN
easat-5356	40	5	et	et	PROPN
easat-5356	40	6	al	al	PROPN
easat-5356	40	7	.	.	PUNCT
easat-5356	41	1	[	[	X
easat-5356	41	2	9	9	NUM
easat-5356	41	3	]	]	PUNCT
easat-5356	41	4	optimized	optimize	VERB
easat-5356	41	5	anchor	anchor	NOUN
easat-5356	41	6	sizes	size	NOUN
easat-5356	41	7	based	base	VERB
easat-5356	41	8	on	on	ADP
easat-5356	41	9	the	the	DET
easat-5356	41	10	effective	effective	ADJ
easat-5356	41	11	receptive	receptive	ADJ
easat-5356	41	12	field	field	NOUN
easat-5356	41	13	,	,	PUNCT
easat-5356	41	14	allowing	allow	VERB
easat-5356	41	15	the	the	DET
easat-5356	41	16	model	model	NOUN
easat-5356	41	17	to	to	PART
easat-5356	41	18	better	well	ADV
easat-5356	41	19	match	match	VERB
easat-5356	41	20	the	the	DET
easat-5356	41	21	natural	natural	ADJ
easat-5356	41	22	variation	variation	NOUN
easat-5356	41	23	in	in	ADP
easat-5356	41	24	face	face	NOUN
easat-5356	41	25	sizes	size	NOUN
easat-5356	41	26	.	.	PUNCT
easat-5356	42	1	similarly	similarly	ADV
easat-5356	42	2	,	,	PUNCT
easat-5356	42	3	faceboxes	faceboxes	PROPN
easat-5356	42	4	qi	qi	PROPN
easat-5356	42	5	,	,	PUNCT
easat-5356	42	6	et	et	PROPN
easat-5356	42	7	al	al	PROPN
easat-5356	42	8	.	.	PUNCT
easat-5356	43	1	[	[	X
easat-5356	43	2	10	10	NUM
easat-5356	43	3	]	]	PUNCT
easat-5356	43	4	introduced	introduce	VERB
easat-5356	43	5	multi	multi	ADJ
easat-5356	43	6	-	-	ADJ
easat-5356	43	7	scale	scale	ADJ
easat-5356	43	8	anchors	anchor	NOUN
easat-5356	43	9	,	,	PUNCT
easat-5356	43	10	which	which	PRON
easat-5356	43	11	enhance	enhance	VERB
easat-5356	43	12	the	the	DET
easat-5356	43	13	receptive	receptive	ADJ
easat-5356	43	14	field	field	NOUN
easat-5356	43	15	by	by	ADP
easat-5356	43	16	assigning	assign	VERB
easat-5356	43	17	different	different	ADJ
easat-5356	43	18	anchor	anchor	NOUN
easat-5356	43	19	sizes	size	NOUN
easat-5356	43	20	to	to	ADP
easat-5356	43	21	various	various	ADJ
easat-5356	43	22	layers	layer	NOUN
easat-5356	43	23	,	,	PUNCT
easat-5356	43	24	thereby	thereby	ADV
easat-5356	43	25	improving	improve	VERB
easat-5356	43	26	the	the	DET
easat-5356	43	27	model	model	NOUN
easat-5356	43	28	’s	’s	PART
easat-5356	43	29	ability	ability	NOUN
easat-5356	43	30	to	to	PART
easat-5356	43	31	detect	detect	VERB
easat-5356	43	32	faces	face	NOUN
easat-5356	43	33	at	at	ADP
easat-5356	43	34	a	a	DET
easat-5356	43	35	wide	wide	ADJ
easat-5356	43	36	range	range	NOUN
easat-5356	43	37	of	of	ADP
easat-5356	43	38	scales	scale	NOUN
easat-5356	43	39	.	.	PUNCT
easat-5356	44	1	these	these	DET
easat-5356	44	2	adaptations	adaptation	NOUN
easat-5356	44	3	help	help	VERB
easat-5356	44	4	ensure	ensure	VERB
easat-5356	44	5	that	that	SCONJ
easat-5356	44	6	the	the	DET
easat-5356	44	7	network	network	NOUN
easat-5356	44	8	can	can	AUX
easat-5356	44	9	effectively	effectively	ADV
easat-5356	44	10	capture	capture	VERB
easat-5356	44	11	faces	face	NOUN
easat-5356	44	12	of	of	ADP
easat-5356	44	13	varying	vary	VERB
easat-5356	44	14	sizes	size	NOUN
easat-5356	44	15	,	,	PUNCT
easat-5356	44	16	enhancing	enhance	VERB
easat-5356	44	17	detection	detection	NOUN
easat-5356	44	18	accuracy	accuracy	NOUN
easat-5356	44	19	and	and	CCONJ
easat-5356	44	20	robustness	robustness	NOUN
easat-5356	44	21	in	in	ADP
easat-5356	44	22	complex	complex	ADJ
easat-5356	44	23	,	,	PUNCT
easat-5356	44	24	real	real	ADJ
easat-5356	44	25	-	-	PUNCT
easat-5356	44	26	world	world	NOUN
easat-5356	44	27	scenarios	scenario	NOUN
easat-5356	44	28	.	.	PUNCT
easat-5356	45	1	motivated	motivate	VERB
easat-5356	45	2	by	by	ADP
easat-5356	45	3	the	the	DET
easat-5356	45	4	idea	idea	NOUN
easat-5356	45	5	of	of	ADP
easat-5356	45	6	receptive	receptive	ADJ
easat-5356	45	7	fields	field	NOUN
easat-5356	45	8	,	,	PUNCT
easat-5356	45	9	we	we	PRON
easat-5356	45	10	created	create	VERB
easat-5356	45	11	a	a	DET
easat-5356	45	12	gcsem	gcsem	NOUN
easat-5356	45	13	module	module	NOUN
easat-5356	45	14	that	that	PRON
easat-5356	45	15	offers	offer	VERB
easat-5356	45	16	a	a	DET
easat-5356	45	17	greater	great	ADJ
easat-5356	45	18	variety	variety	NOUN
easat-5356	45	19	of	of	ADP
easat-5356	45	20	receptive	receptive	ADJ
easat-5356	45	21	fields	field	NOUN
easat-5356	45	22	,	,	PUNCT
easat-5356	45	23	improving	improve	VERB
easat-5356	45	24	the	the	DET
easat-5356	45	25	model	model	NOUN
easat-5356	45	26	's	's	PART
easat-5356	45	27	capacity	capacity	NOUN
easat-5356	45	28	to	to	PART
easat-5356	45	29	identify	identify	VERB
easat-5356	45	30	faces	face	NOUN
easat-5356	45	31	at	at	ADP
easat-5356	45	32	various	various	ADJ
easat-5356	45	33	scales	scale	NOUN
easat-5356	45	34	.	.	PUNCT
easat-5356	46	1	numerous	numerous	ADJ
easat-5356	46	2	studies	study	NOUN
easat-5356	46	3	have	have	AUX
easat-5356	46	4	aimed	aim	VERB
easat-5356	46	5	to	to	PART
easat-5356	46	6	advance	advance	VERB
easat-5356	46	7	face	face	NOUN
easat-5356	46	8	detection	detection	NOUN
easat-5356	46	9	architectures	architecture	NOUN
easat-5356	46	10	,	,	PUNCT
easat-5356	46	11	primarily	primarily	ADV
easat-5356	46	12	divided	divide	VERB
easat-5356	46	13	into	into	ADP
easat-5356	46	14	single	single	ADJ
easat-5356	46	15	-	-	PUNCT
easat-5356	46	16	stage	stage	NOUN
easat-5356	46	17	zhang	zhang	PROPN
easat-5356	46	18	,	,	PUNCT
easat-5356	46	19	et	et	PROPN
easat-5356	46	20	al	al	PROPN
easat-5356	46	21	.	.	PUNCT
easat-5356	47	1	[	[	X
easat-5356	47	2	11	11	NUM
easat-5356	47	3	]	]	PUNCT
easat-5356	47	4	and	and	CCONJ
easat-5356	47	5	two	two	NUM
easat-5356	47	6	-	-	PUNCT
easat-5356	47	7	stage	stage	NOUN
easat-5356	47	8	wu	wu	PROPN
easat-5356	47	9	,	,	PUNCT
easat-5356	47	10	et	et	PROPN
easat-5356	47	11	al	al	PROPN
easat-5356	47	12	.	.	PUNCT
easat-5356	48	1	[	[	X
easat-5356	48	2	12	12	NUM
easat-5356	48	3	]	]	PUNCT
easat-5356	48	4	detectors	detector	NOUN
easat-5356	48	5	.	.	PUNCT
easat-5356	49	1	densely	densely	ADV
easat-5356	49	2	packed	pack	VERB
easat-5356	49	3	anchor	anchor	NOUN
easat-5356	49	4	points	point	NOUN
easat-5356	49	5	are	be	AUX
easat-5356	49	6	deployed	deploy	VERB
easat-5356	49	7	throughout	throughout	ADP
easat-5356	49	8	all	all	DET
easat-5356	49	9	locations	location	NOUN
easat-5356	49	10	of	of	ADP
easat-5356	49	11	numerous	numerous	ADJ
easat-5356	49	12	multi	multi	ADJ
easat-5356	49	13	-	-	ADJ
easat-5356	49	14	scale	scale	ADJ
easat-5356	49	15	feature	feature	NOUN
easat-5356	49	16	maps	map	NOUN
easat-5356	49	17	with	with	ADP
easat-5356	49	18	different	different	ADJ
easat-5356	49	19	sizes	size	NOUN
easat-5356	49	20	and	and	CCONJ
easat-5356	49	21	aspect	aspect	NOUN
easat-5356	49	22	ratios	ratio	NOUN
easat-5356	49	23	by	by	ADP
easat-5356	49	24	single	single	ADJ
easat-5356	49	25	-	-	PUNCT
easat-5356	49	26	stage	stage	NOUN
easat-5356	49	27	detectors	detector	NOUN
easat-5356	49	28	,	,	PUNCT
easat-5356	49	29	which	which	PRON
easat-5356	49	30	are	be	AUX
easat-5356	49	31	domainand	domainand	ADJ
easat-5356	49	32	anchorpoint	anchorpoint	NOUN
easat-5356	49	33	-	-	PUNCT
easat-5356	49	34	based	base	VERB
easat-5356	49	35	.	.	PUNCT
easat-5356	50	1	by	by	ADP
easat-5356	50	2	utilizing	utilize	VERB
easat-5356	50	3	the	the	DET
easat-5356	50	4	flattened	flatten	VERB
easat-5356	50	5	anchor	anchor	NOUN
easat-5356	50	6	points	point	NOUN
easat-5356	50	7	,	,	PUNCT
easat-5356	50	8	this	this	DET
easat-5356	50	9	method	method	NOUN
easat-5356	50	10	enables	enable	VERB
easat-5356	50	11	the	the	DET
easat-5356	50	12	model	model	NOUN
easat-5356	50	13	to	to	PART
easat-5356	50	14	effectively	effectively	ADV
easat-5356	50	15	recognize	recognize	VERB
easat-5356	50	16	faces	face	NOUN
easat-5356	50	17	in	in	ADP
easat-5356	50	18	a	a	DET
easat-5356	50	19	single	single	ADJ
easat-5356	50	20	pass	pass	NOUN
easat-5356	50	21	,	,	PUNCT
easat-5356	50	22	providing	provide	VERB
easat-5356	50	23	a	a	DET
easat-5356	50	24	simplified	simplified	ADJ
easat-5356	50	25	and	and	CCONJ
easat-5356	50	26	effective	effective	ADJ
easat-5356	50	27	solution	solution	NOUN
easat-5356	50	28	that	that	PRON
easat-5356	50	29	is	be	AUX
easat-5356	50	30	especially	especially	ADV
easat-5356	50	31	advantageous	advantageous	ADJ
easat-5356	50	32	for	for	ADP
easat-5356	50	33	real	real	ADJ
easat-5356	50	34	-	-	PUNCT
easat-5356	50	35	time	time	NOUN
easat-5356	50	36	applications	application	NOUN
easat-5356	50	37	.	.	PUNCT
easat-5356	51	1	two	two	NUM
easat-5356	51	2	-	-	PUNCT
easat-5356	51	3	stage	stage	NOUN
easat-5356	51	4	detectors	detector	NOUN
easat-5356	51	5	,	,	PUNCT
easat-5356	51	6	on	on	ADP
easat-5356	51	7	the	the	DET
easat-5356	51	8	other	other	ADJ
easat-5356	51	9	hand	hand	NOUN
easat-5356	51	10	,	,	PUNCT
easat-5356	51	11	usually	usually	ADV
easat-5356	51	12	use	use	VERB
easat-5356	51	13	a	a	DET
easat-5356	51	14	region	region	NOUN
easat-5356	51	15	proposal	proposal	NOUN
easat-5356	51	16	network	network	NOUN
easat-5356	51	17	to	to	PART
easat-5356	51	18	produce	produce	VERB
easat-5356	51	19	candidate	candidate	NOUN
easat-5356	51	20	regions	region	NOUN
easat-5356	51	21	,	,	PUNCT
easat-5356	51	22	which	which	PRON
easat-5356	51	23	are	be	AUX
easat-5356	51	24	then	then	ADV
easat-5356	51	25	refined	refine	VERB
easat-5356	51	26	in	in	ADP
easat-5356	51	27	a	a	DET
easat-5356	51	28	second	second	ADJ
easat-5356	51	29	stage	stage	NOUN
easat-5356	51	30	.	.	PUNCT
easat-5356	52	1	this	this	PRON
easat-5356	52	2	increases	increase	VERB
easat-5356	52	3	accuracy	accuracy	NOUN
easat-5356	52	4	but	but	CCONJ
easat-5356	52	5	frequently	frequently	ADV
easat-5356	52	6	comes	come	VERB
easat-5356	52	7	at	at	ADP
easat-5356	52	8	the	the	DET
easat-5356	52	9	expense	expense	NOUN
easat-5356	52	10	of	of	ADP
easat-5356	52	11	more	more	ADJ
easat-5356	52	12	computing	computing	NOUN
easat-5356	52	13	complexity	complexity	NOUN
easat-5356	52	14	.	.	PUNCT
easat-5356	53	1	the	the	DET
easat-5356	53	2	head	head	NOUN
easat-5356	53	3	network	network	NOUN
easat-5356	53	4	,	,	PUNCT
easat-5356	53	5	the	the	DET
easat-5356	53	6	feature	feature	NOUN
easat-5356	53	7	module	module	NOUN
easat-5356	53	8	,	,	PUNCT
easat-5356	53	9	the	the	DET
easat-5356	53	10	multitask	multitask	ADJ
easat-5356	53	11	loss	loss	NOUN
easat-5356	53	12	,	,	PUNCT
easat-5356	53	13	and	and	CCONJ
easat-5356	53	14	the	the	DET
easat-5356	53	15	backbone	backbone	NOUN
easat-5356	53	16	are	be	AUX
easat-5356	53	17	the	the	DET
easat-5356	53	18	four	four	NUM
easat-5356	53	19	main	main	ADJ
easat-5356	53	20	parts	part	NOUN
easat-5356	53	21	of	of	ADP
easat-5356	53	22	this	this	DET
easat-5356	53	23	system	system	NOUN
easat-5356	53	24	.	.	PUNCT
easat-5356	54	1	the	the	DET
easat-5356	54	2	feature	feature	NOUN
easat-5356	54	3	extraction	extraction	NOUN
easat-5356	54	4	module	module	NOUN
easat-5356	54	5	employs	employ	VERB
easat-5356	54	6	a	a	DET
easat-5356	54	7	feature	feature	NOUN
easat-5356	54	8	pyramid	pyramid	NOUN
easat-5356	54	9	network	network	NOUN
easat-5356	54	10	(	(	PUNCT
easat-5356	54	11	fpn	fpn	PROPN
easat-5356	54	12	)	)	PUNCT
easat-5356	54	13	huang	huang	PROPN
easat-5356	54	14	,	,	PUNCT
easat-5356	54	15	et	et	PROPN
easat-5356	54	16	al	al	PROPN
easat-5356	54	17	.	.	PUNCT
easat-5356	55	1	[	[	X
easat-5356	55	2	13	13	NUM
easat-5356	55	3	]	]	PUNCT
easat-5356	55	4	to	to	PART
easat-5356	55	5	effectively	effectively	ADV
easat-5356	55	6	aggregate	aggregate	VERB
easat-5356	55	7	highand	highand	ADJ
easat-5356	55	8	low	low	ADJ
easat-5356	55	9	-	-	PUNCT
easat-5356	55	10	level	level	NOUN
easat-5356	55	11	features	feature	NOUN
easat-5356	55	12	from	from	ADP
easat-5356	55	13	the	the	DET
easat-5356	55	14	backbone	backbone	NOUN
easat-5356	55	15	network	network	NOUN
easat-5356	55	16	,	,	PUNCT
easat-5356	55	17	enabling	enable	VERB
easat-5356	55	18	the	the	DET
easat-5356	55	19	model	model	NOUN
easat-5356	55	20	to	to	PART
easat-5356	55	21	capture	capture	VERB
easat-5356	55	22	fine	fine	ADV
easat-5356	55	23	-	-	PUNCT
easat-5356	55	24	grained	grain	VERB
easat-5356	55	25	and	and	CCONJ
easat-5356	55	26	contextually	contextually	ADV
easat-5356	55	27	rich	rich	ADJ
easat-5356	55	28	details	detail	NOUN
easat-5356	55	29	across	across	ADP
easat-5356	55	30	scales	scale	NOUN
easat-5356	55	31	.	.	PUNCT
easat-5356	56	1	additionally	additionally	ADV
easat-5356	56	2	,	,	PUNCT
easat-5356	56	3	modules	module	NOUN
easat-5356	56	4	for	for	ADP
easat-5356	56	5	refining	refine	VERB
easat-5356	56	6	the	the	DET
easat-5356	56	7	receptive	receptive	ADJ
easat-5356	56	8	field	field	NOUN
easat-5356	56	9	,	,	PUNCT
easat-5356	56	10	such	such	ADJ
easat-5356	56	11	as	as	ADP
easat-5356	56	12	receptive	receptive	ADJ
easat-5356	56	13	field	field	NOUN
easat-5356	56	14	blocks	block	NOUN
easat-5356	56	15	(	(	PUNCT
easat-5356	56	16	rfbs	rfbs	NOUN
easat-5356	56	17	)	)	PUNCT
easat-5356	56	18	terven	terven	VERB
easat-5356	56	19	,	,	PUNCT
easat-5356	56	20	et	et	PROPN
easat-5356	56	21	al	al	PROPN
easat-5356	56	22	.	.	PUNCT
easat-5356	57	1	[	[	X
easat-5356	57	2	14	14	NUM
easat-5356	57	3	]	]	PUNCT
easat-5356	57	4	are	be	AUX
easat-5356	57	5	incorporated	incorporate	VERB
easat-5356	57	6	to	to	PART
easat-5356	57	7	enhance	enhance	VERB
easat-5356	57	8	contextual	contextual	ADJ
easat-5356	57	9	information	information	NOUN
easat-5356	57	10	,	,	PUNCT
easat-5356	57	11	especially	especially	ADV
easat-5356	57	12	for	for	ADP
easat-5356	57	13	challenging	challenge	VERB
easat-5356	57	14	face	face	NOUN
easat-5356	57	15	detections	detection	NOUN
easat-5356	57	16	in	in	ADP
easat-5356	57	17	complex	complex	ADJ
easat-5356	57	18	environments	environment	NOUN
easat-5356	57	19	.	.	PUNCT
easat-5356	58	1	the	the	DET
easat-5356	58	2	multitask	multitask	ADJ
easat-5356	58	3	loss	loss	NOUN
easat-5356	58	4	function	function	NOUN
easat-5356	58	5	used	use	VERB
easat-5356	58	6	in	in	ADP
easat-5356	58	7	this	this	DET
easat-5356	58	8	model	model	NOUN
easat-5356	58	9	includes	include	VERB
easat-5356	58	10	both	both	PRON
easat-5356	58	11	binary	binary	ADJ
easat-5356	58	12	classification	classification	NOUN
easat-5356	58	13	and	and	CCONJ
easat-5356	58	14	bounding	bound	VERB
easat-5356	58	15	box	box	NOUN
easat-5356	58	16	regression	regression	NOUN
easat-5356	58	17	.	.	PUNCT
easat-5356	59	1	the	the	DET
easat-5356	59	2	binary	binary	PROPN
easat-5356	59	3	classification	classification	NOUN
easat-5356	59	4	loss	loss	NOUN
easat-5356	59	5	categorizes	categorize	VERB
easat-5356	59	6	predefined	predefine	VERB
easat-5356	59	7	anchors	anchor	NOUN
easat-5356	59	8	as	as	ADP
easat-5356	59	9	either	either	DET
easat-5356	59	10	faces	face	NOUN
easat-5356	59	11	or	or	CCONJ
easat-5356	59	12	backgrounds	background	NOUN
easat-5356	59	13	,	,	PUNCT
easat-5356	59	14	while	while	SCONJ
easat-5356	59	15	the	the	DET
easat-5356	59	16	bounding	bounding	NOUN
easat-5356	59	17	box	box	NOUN
easat-5356	59	18	regression	regression	NOUN
easat-5356	59	19	loss	loss	NOUN
easat-5356	59	20	enhances	enhance	VERB
easat-5356	59	21	the	the	DET
easat-5356	59	22	predicted	predict	VERB
easat-5356	59	23	locations	location	NOUN
easat-5356	59	24	of	of	ADP
easat-5356	59	25	detected	detect	VERB
easat-5356	59	26	faces	face	NOUN
easat-5356	59	27	,	,	PUNCT
easat-5356	59	28	ensuring	ensure	VERB
easat-5356	59	29	precise	precise	ADJ
easat-5356	59	30	localization	localization	NOUN
easat-5356	59	31	.	.	PUNCT
easat-5356	60	1	together	together	ADV
easat-5356	60	2	,	,	PUNCT
easat-5356	60	3	these	these	DET
easat-5356	60	4	components	component	NOUN
easat-5356	60	5	contribute	contribute	VERB
easat-5356	60	6	to	to	ADP
easat-5356	60	7	a	a	DET
easat-5356	60	8	more	more	ADV
easat-5356	60	9	robust	robust	ADJ
easat-5356	60	10	face	face	NOUN
easat-5356	60	11	detection	detection	NOUN
easat-5356	60	12	system	system	NOUN
easat-5356	60	13	that	that	PRON
easat-5356	60	14	performs	perform	VERB
easat-5356	60	15	well	well	ADV
easat-5356	60	16	across	across	ADP
easat-5356	60	17	varied	varied	ADJ
easat-5356	60	18	face	face	NOUN
easat-5356	60	19	sizes	size	NOUN
easat-5356	60	20	and	and	CCONJ
easat-5356	60	21	challenging	challenging	ADJ
easat-5356	60	22	visual	visual	ADJ
easat-5356	60	23	contexts	contexts	NOUN
easat-5356	60	24	.	.	PUNCT
easat-5356	61	1	however	however	ADV
easat-5356	61	2	,	,	PUNCT
easat-5356	61	3	difficult	difficult	ADJ
easat-5356	61	4	samples	sample	NOUN
easat-5356	61	5	and	and	CCONJ
easat-5356	61	6	small	small	ADJ
easat-5356	61	7	-	-	PUNCT
easat-5356	61	8	size	size	NOUN
easat-5356	61	9	face	face	NOUN
easat-5356	61	10	detection	detection	NOUN
easat-5356	61	11	issues	issue	NOUN
easat-5356	61	12	remain	remain	VERB
easat-5356	61	13	a	a	DET
easat-5356	61	14	problem	problem	NOUN
easat-5356	61	15	for	for	ADP
easat-5356	61	16	many	many	ADJ
easat-5356	61	17	detection	detection	NOUN
easat-5356	61	18	methods	method	NOUN
easat-5356	61	19	.	.	PUNCT
easat-5356	62	1	to	to	PART
easat-5356	62	2	address	address	VERB
easat-5356	62	3	this	this	DET
easat-5356	62	4	problem	problem	NOUN
easat-5356	62	5	,	,	PUNCT
easat-5356	62	6	we	we	PRON
easat-5356	62	7	designed	design	VERB
easat-5356	62	8	a	a	DET
easat-5356	62	9	novel	novel	ADJ
easat-5356	62	10	feature	feature	NOUN
easat-5356	62	11	extraction	extraction	NOUN
easat-5356	62	12	module	module	NOUN
easat-5356	62	13	that	that	PRON
easat-5356	62	14	improves	improve	VERB
easat-5356	62	15	smallsize	smallsize	NOUN
easat-5356	62	16	and	and	CCONJ
easat-5356	62	17	difficult	difficult	ADJ
easat-5356	62	18	sample	sample	NOUN
easat-5356	62	19	detection	detection	NOUN
easat-5356	62	20	while	while	SCONJ
easat-5356	62	21	minimizing	minimize	VERB
easat-5356	62	22	the	the	DET
easat-5356	62	23	loss	loss	NOUN
easat-5356	62	24	of	of	ADP
easat-5356	62	25	detail	detail	NOUN
easat-5356	62	26	information	information	NOUN
easat-5356	62	27	.	.	PUNCT
easat-5356	63	1	however	however	ADV
easat-5356	63	2	,	,	PUNCT
easat-5356	63	3	faces	face	NOUN
easat-5356	63	4	may	may	AUX
easat-5356	63	5	appear	appear	VERB
easat-5356	63	6	extremely	extremely	ADV
easat-5356	63	7	small	small	ADJ
easat-5356	63	8	in	in	ADP
easat-5356	63	9	certain	certain	ADJ
easat-5356	63	10	situations	situation	NOUN
easat-5356	63	11	,	,	PUNCT
easat-5356	63	12	such	such	ADJ
easat-5356	63	13	as	as	ADP
easat-5356	63	14	surveillance	surveillance	NOUN
easat-5356	63	15	settings	setting	NOUN
easat-5356	63	16	.	.	PUNCT
easat-5356	64	1	ignoring	ignore	VERB
easat-5356	64	2	the	the	DET
easat-5356	64	3	identification	identification	NOUN
easat-5356	64	4	of	of	ADP
easat-5356	64	5	these	these	DET
easat-5356	64	6	targets	target	NOUN
easat-5356	64	7	could	could	AUX
easat-5356	64	8	cause	cause	VERB
easat-5356	64	9	an	an	DET
easat-5356	64	10	intelligent	intelligent	ADJ
easat-5356	64	11	surveillance	surveillance	NOUN
easat-5356	64	12	system	system	NOUN
easat-5356	64	13	to	to	ADP
easat-5356	64	14	malfunction	malfunction	NOUN
easat-5356	64	15	.	.	PUNCT
easat-5356	65	1	tiny	tiny	ADJ
easat-5356	65	2	faces	face	NOUN
easat-5356	65	3	,	,	PUNCT
easat-5356	65	4	particularly	particularly	ADV
easat-5356	65	5	those	those	DET
easat-5356	65	6	smaller	small	ADJ
easat-5356	65	7	than	than	ADP
easat-5356	65	8	16x16	16x16	NUM
easat-5356	65	9	pixels	pixel	NOUN
easat-5356	65	10	,	,	PUNCT
easat-5356	65	11	are	be	AUX
easat-5356	65	12	usually	usually	ADV
easat-5356	65	13	detected	detect	VERB
easat-5356	65	14	by	by	ADP
easat-5356	65	15	densely	densely	ADV
easat-5356	65	16	tiling	tile	VERB
easat-5356	65	17	a	a	DET
easat-5356	65	18	large	large	ADJ
easat-5356	65	19	number	number	NOUN
easat-5356	65	20	of	of	ADP
easat-5356	65	21	tiny	tiny	ADJ
easat-5356	65	22	anchor	anchor	NOUN
easat-5356	65	23	points	point	NOUN
easat-5356	65	24	throughout	throughout	ADP
easat-5356	65	25	the	the	DET
easat-5356	65	26	image	image	NOUN
easat-5356	65	27	in	in	ADP
easat-5356	65	28	modern	modern	ADJ
easat-5356	65	29	face	face	NOUN
easat-5356	65	30	identification	identification	NOUN
easat-5356	65	31	techniques	technique	NOUN
easat-5356	65	32	.	.	PUNCT
easat-5356	66	1	although	although	SCONJ
easat-5356	66	2	this	this	DET
easat-5356	66	3	method	method	NOUN
easat-5356	66	4	increases	increase	VERB
easat-5356	66	5	recall	recall	VERB
easat-5356	66	6	,	,	PUNCT
easat-5356	66	7	it	it	PRON
easat-5356	66	8	exacerbates	exacerbate	VERB
easat-5356	66	9	the	the	DET
easat-5356	66	10	issue	issue	NOUN
easat-5356	66	11	of	of	ADP
easat-5356	66	12	high	high	ADJ
easat-5356	66	13	category	category	NOUN
easat-5356	66	14	imbalance	imbalance	NOUN
easat-5356	66	15	,	,	PUNCT
easat-5356	66	16	a	a	DET
easat-5356	66	17	significant	significant	ADJ
easat-5356	66	18	cause	cause	NOUN
easat-5356	66	19	of	of	ADP
easat-5356	66	20	excessive	excessive	ADJ
easat-5356	66	21	false	false	ADJ
easat-5356	66	22	positives	positive	NOUN
easat-5356	66	23	[	[	X
easat-5356	66	24	15	15	NUM
easat-5356	66	25	]	]	PUNCT
easat-5356	66	26	.	.	PUNCT
easat-5356	67	1	the	the	DET
easat-5356	67	2	cumulative	cumulative	ADJ
easat-5356	67	3	effect	effect	NOUN
easat-5356	67	4	of	of	ADP
easat-5356	67	5	these	these	DET
easat-5356	67	6	false	false	ADJ
easat-5356	67	7	positives	positive	NOUN
easat-5356	67	8	during	during	ADP
easat-5356	67	9	training	training	NOUN
easat-5356	67	10	significantly	significantly	ADV
easat-5356	67	11	affects	affect	VERB
easat-5356	67	12	the	the	DET
easat-5356	67	13	model	model	NOUN
easat-5356	67	14	's	's	PART
easat-5356	67	15	updates	update	NOUN
easat-5356	67	16	,	,	PUNCT
easat-5356	67	17	resulting	result	VERB
easat-5356	67	18	in	in	ADP
easat-5356	67	19	a	a	DET
easat-5356	67	20	notable	notable	ADJ
easat-5356	67	21	sample	sample	NOUN
easat-5356	67	22	imbalance	imbalance	NOUN
easat-5356	67	23	that	that	PRON
easat-5356	67	24	compromises	compromise	VERB
easat-5356	67	25	the	the	DET
easat-5356	67	26	training	training	NOUN
easat-5356	67	27	process	process	NOUN
easat-5356	67	28	'	'	PART
easat-5356	67	29	effectiveness	effectiveness	NOUN
easat-5356	67	30	.	.	PUNCT
easat-5356	68	1	this	this	DET
easat-5356	68	2	imbalance	imbalance	NOUN
easat-5356	68	3	often	often	ADV
easat-5356	68	4	causes	cause	VERB
easat-5356	68	5	the	the	DET
easat-5356	68	6	model	model	NOUN
easat-5356	68	7	to	to	PART
easat-5356	68	8	focus	focus	VERB
easat-5356	68	9	disproportionately	disproportionately	ADV
easat-5356	68	10	on	on	ADP
easat-5356	68	11	background	background	NOUN
easat-5356	68	12	regions	region	NOUN
easat-5356	68	13	,	,	PUNCT
easat-5356	68	14	reducing	reduce	VERB
easat-5356	68	15	its	its	PRON
easat-5356	68	16	ability	ability	NOUN
easat-5356	68	17	to	to	PART
easat-5356	68	18	effectively	effectively	ADV
easat-5356	68	19	learn	learn	VERB
easat-5356	68	20	distinctive	distinctive	ADJ
easat-5356	68	21	features	feature	NOUN
easat-5356	68	22	for	for	ADP
easat-5356	68	23	accurate	accurate	ADJ
easat-5356	68	24	face	face	NOUN
easat-5356	68	25	detection	detection	NOUN
easat-5356	68	26	,	,	PUNCT
easat-5356	68	27	especially	especially	ADV
easat-5356	68	28	for	for	ADP
easat-5356	68	29	small	small	ADJ
easat-5356	68	30	faces	face	NOUN
easat-5356	68	31	in	in	ADP
easat-5356	68	32	cluttered	cluttered	ADJ
easat-5356	68	33	or	or	CCONJ
easat-5356	68	34	complex	complex	ADJ
easat-5356	68	35	scenes	scene	NOUN
easat-5356	68	36	.	.	PUNCT
easat-5356	69	1	to	to	PART
easat-5356	69	2	deal	deal	VERB
easat-5356	69	3	with	with	ADP
easat-5356	69	4	the	the	DET
easat-5356	69	5	problem	problem	NOUN
easat-5356	69	6	,	,	PUNCT
easat-5356	69	7	lin	lin	PROPN
easat-5356	69	8	et	et	PROPN
easat-5356	69	9	al	al	PROPN
easat-5356	69	10	.	.	PUNCT
easat-5356	70	1	[	[	X
easat-5356	70	2	16	16	NUM
easat-5356	70	3	]	]	PUNCT
easat-5356	70	4	introduced	introduce	VERB
easat-5356	70	5	focal	focal	ADJ
easat-5356	70	6	loss	loss	NOUN
easat-5356	70	7	,	,	PUNCT
easat-5356	70	8	which	which	PRON
easat-5356	70	9	dynamically	dynamically	ADV
easat-5356	70	10	assigns	assign	VERB
easat-5356	70	11	greater	great	ADJ
easat-5356	70	12	weight	weight	NOUN
easat-5356	70	13	to	to	ADP
easat-5356	70	14	difficult	difficult	ADJ
easat-5356	70	15	samples	sample	NOUN
easat-5356	70	16	,	,	PUNCT
easat-5356	70	17	reducing	reduce	VERB
easat-5356	70	18	the	the	DET
easat-5356	70	19	impact	impact	NOUN
easat-5356	70	20	of	of	ADP
easat-5356	70	21	easily	easily	ADV
easat-5356	70	22	classified	classified	ADJ
easat-5356	70	23	examples	example	NOUN
easat-5356	70	24	and	and	CCONJ
easat-5356	70	25	thus	thus	ADV
easat-5356	70	26	helping	help	VERB
easat-5356	70	27	to	to	PART
easat-5356	70	28	alleviate	alleviate	VERB
easat-5356	70	29	class	class	NOUN
easat-5356	70	30	imbalance	imbalance	NOUN
easat-5356	70	31	.	.	PUNCT
easat-5356	71	1	correspondingly	correspondingly	ADV
easat-5356	71	2	,	,	PUNCT
easat-5356	71	3	the	the	DET
easat-5356	71	4	gradient	gradient	NOUN
easat-5356	71	5	harmonizing	harmonize	VERB
easat-5356	71	6	mechanism	mechanism	NOUN
easat-5356	71	7	(	(	PUNCT
easat-5356	71	8	ghm	ghm	NOUN
easat-5356	71	9	)	)	PUNCT
easat-5356	71	10	meher	meher	PROPN
easat-5356	71	11	,	,	PUNCT
easat-5356	71	12	et	et	PROPN
easat-5356	71	13	al	al	PROPN
easat-5356	71	14	.	.	PUNCT
easat-5356	72	1	[	[	X
easat-5356	72	2	1	1	X
easat-5356	72	3	]	]	PUNCT
easat-5356	72	4	reduces	reduce	VERB
easat-5356	72	5	the	the	DET
easat-5356	72	6	gradient	gradient	ADJ
easat-5356	72	7	contributions	contribution	NOUN
easat-5356	72	8	of	of	ADP
easat-5356	72	9	simple	simple	ADJ
easat-5356	72	10	samples	sample	NOUN
easat-5356	72	11	,	,	PUNCT
easat-5356	72	12	allowing	allow	VERB
easat-5356	72	13	the	the	DET
easat-5356	72	14	model	model	NOUN
easat-5356	72	15	to	to	PART
easat-5356	72	16	prioritize	prioritize	VERB
easat-5356	72	17	learning	learn	VERB
easat-5356	72	18	from	from	ADP
easat-5356	72	19	harder	hard	ADJ
easat-5356	72	20	examples	example	NOUN
easat-5356	72	21	.	.	PUNCT
easat-5356	73	1	another	another	DET
easat-5356	73	2	approach	approach	NOUN
easat-5356	73	3	,	,	PUNCT
easat-5356	73	4	prime	prime	ADJ
easat-5356	73	5	sample	sample	NOUN
easat-5356	73	6	attention	attention	NOUN
easat-5356	73	7	(	(	PUNCT
easat-5356	73	8	pisa	pisa	ADJ
easat-5356	73	9	)	)	PUNCT
easat-5356	73	10	song	song	NOUN
easat-5356	73	11	,	,	PUNCT
easat-5356	73	12	et	et	PROPN
easat-5356	73	13	al	al	PROPN
easat-5356	73	14	.	.	PUNCT
easat-5356	74	1	[	[	X
easat-5356	74	2	17	17	NUM
easat-5356	74	3	]	]	PUNCT
easat-5356	74	4	assigns	assign	VERB
easat-5356	74	5	weights	weight	NOUN
easat-5356	74	6	to	to	ADP
easat-5356	74	7	both	both	CCONJ
easat-5356	74	8	positive	positive	ADJ
easat-5356	74	9	and	and	CCONJ
easat-5356	74	10	negative	negative	ADJ
easat-5356	74	11	samples	sample	NOUN
easat-5356	74	12	according	accord	VERB
easat-5356	74	13	to	to	ADP
easat-5356	74	14	distinct	distinct	ADJ
easat-5356	74	15	criteria	criterion	NOUN
easat-5356	74	16	,	,	PUNCT
easat-5356	74	17	further	far	ADV
easat-5356	74	18	enhancing	enhance	VERB
easat-5356	74	19	the	the	DET
easat-5356	74	20	model	model	NOUN
easat-5356	74	21	’s	’s	PART
easat-5356	74	22	focus	focus	NOUN
easat-5356	74	23	on	on	ADP
easat-5356	74	24	challenging	challenging	ADJ
easat-5356	74	25	samples	sample	NOUN
easat-5356	74	26	.	.	PUNCT
easat-5356	75	1	these	these	DET
easat-5356	75	2	strategies	strategy	NOUN
easat-5356	75	3	collectively	collectively	ADV
easat-5356	75	4	improve	improve	VERB
easat-5356	75	5	training	training	NOUN
easat-5356	75	6	efficiency	efficiency	NOUN
easat-5356	75	7	and	and	CCONJ
easat-5356	75	8	detection	detection	NOUN
easat-5356	75	9	accuracy	accuracy	NOUN
easat-5356	75	10	by	by	ADP
easat-5356	75	11	mitigating	mitigate	VERB
easat-5356	75	12	the	the	DET
easat-5356	75	13	effect	effect	NOUN
easat-5356	75	14	of	of	ADP
easat-5356	75	15	sample	sample	NOUN
easat-5356	75	16	imbalance	imbalance	NOUN
easat-5356	75	17	and	and	CCONJ
easat-5356	75	18	reducing	reduce	VERB
easat-5356	75	19	false	false	ADJ
easat-5356	75	20	positives	positive	NOUN
easat-5356	75	21	.	.	PUNCT
easat-5356	76	1	842	842	NUM
easat-5356	76	2	edelweiss	edelweiss	PROPN
easat-5356	76	3	applied	apply	VERB
easat-5356	76	4	science	science	NOUN
easat-5356	76	5	and	and	CCONJ
easat-5356	76	6	technology	technology	NOUN
easat-5356	76	7	issn	issn	PROPN
easat-5356	76	8	:	:	PUNCT
easat-5356	76	9	2576	2576	NUM
easat-5356	76	10	-	-	SYM
easat-5356	76	11	8484	8484	NUM
easat-5356	76	12	vol	vol	NOUN
easat-5356	76	13	.	.	PROPN
easat-5356	77	1	9	9	NUM
easat-5356	77	2	,	,	PUNCT
easat-5356	77	3	no	no	INTJ
easat-5356	77	4	.	.	NOUN
easat-5356	78	1	3	3	NUM
easat-5356	78	2	:	:	PUNCT
easat-5356	78	3	840	840	NUM
easat-5356	78	4	-	-	SYM
easat-5356	78	5	855	855	NUM
easat-5356	78	6	,	,	PUNCT
easat-5356	78	7	2025	2025	NUM
easat-5356	78	8	doi	doi	NOUN
easat-5356	78	9	:	:	PUNCT
easat-5356	78	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	78	11	©	©	PROPN
easat-5356	78	12	2025	2025	NUM
easat-5356	78	13	by	by	ADP
easat-5356	78	14	the	the	DET
easat-5356	78	15	authors	author	NOUN
easat-5356	78	16	;	;	PUNCT
easat-5356	78	17	licensee	licensee	PROPN
easat-5356	78	18	learning	learning	NOUN
easat-5356	78	19	gate	gate	NOUN
easat-5356	78	20	however	however	ADV
easat-5356	78	21	,	,	PUNCT
easat-5356	78	22	current	current	ADJ
easat-5356	78	23	hard	hard	ADJ
easat-5356	78	24	sample	sample	NOUN
easat-5356	78	25	mining	mining	NOUN
easat-5356	78	26	techniques	technique	NOUN
easat-5356	78	27	are	be	AUX
easat-5356	78	28	difficult	difficult	ADJ
easat-5356	78	29	to	to	PART
easat-5356	78	30	put	put	VERB
easat-5356	78	31	into	into	ADP
easat-5356	78	32	reality	reality	NOUN
easat-5356	78	33	because	because	SCONJ
easat-5356	78	34	they	they	PRON
easat-5356	78	35	frequently	frequently	ADV
easat-5356	78	36	call	call	VERB
easat-5356	78	37	for	for	ADP
easat-5356	78	38	adjusting	adjust	VERB
easat-5356	78	39	a	a	DET
easat-5356	78	40	lot	lot	NOUN
easat-5356	78	41	of	of	ADP
easat-5356	78	42	hyperparameters	hyperparameter	NOUN
easat-5356	78	43	.	.	PUNCT
easat-5356	79	1	furthermore	furthermore	ADV
easat-5356	79	2	,	,	PUNCT
easat-5356	79	3	accurately	accurately	ADV
easat-5356	79	4	detecting	detect	VERB
easat-5356	79	5	faces	face	NOUN
easat-5356	79	6	with	with	ADP
easat-5356	79	7	substantial	substantial	ADJ
easat-5356	79	8	variations	variation	NOUN
easat-5356	79	9	in	in	ADP
easat-5356	79	10	scale	scale	NOUN
easat-5356	79	11	,	,	PUNCT
easat-5356	79	12	pose	pose	VERB
easat-5356	79	13	,	,	PUNCT
easat-5356	79	14	occlusion	occlusion	NOUN
easat-5356	79	15	,	,	PUNCT
easat-5356	79	16	expression	expression	NOUN
easat-5356	79	17	,	,	PUNCT
easat-5356	79	18	appearance	appearance	NOUN
easat-5356	79	19	,	,	PUNCT
easat-5356	79	20	and	and	CCONJ
easat-5356	79	21	illumination	illumination	NOUN
easat-5356	79	22	remains	remain	VERB
easat-5356	79	23	a	a	DET
easat-5356	79	24	significant	significant	ADJ
easat-5356	79	25	challenge	challenge	NOUN
easat-5356	79	26	[	[	X
easat-5356	79	27	18	18	NUM
easat-5356	79	28	]	]	PUNCT
easat-5356	79	29	.	.	PUNCT
easat-5356	80	1	in	in	ADP
easat-5356	80	2	addition	addition	NOUN
easat-5356	80	3	,	,	PUNCT
easat-5356	80	4	the	the	DET
easat-5356	80	5	use	use	NOUN
easat-5356	80	6	of	of	ADP
easat-5356	80	7	large	large	ADJ
easat-5356	80	8	models	model	NOUN
easat-5356	80	9	and	and	CCONJ
easat-5356	80	10	high	high	ADJ
easat-5356	80	11	computational	computational	ADJ
easat-5356	80	12	costs	cost	NOUN
easat-5356	80	13	can	can	AUX
easat-5356	80	14	limit	limit	VERB
easat-5356	80	15	the	the	DET
easat-5356	80	16	applicability	applicability	NOUN
easat-5356	80	17	of	of	ADP
easat-5356	80	18	these	these	DET
easat-5356	80	19	detectors	detector	NOUN
easat-5356	80	20	in	in	ADP
easat-5356	80	21	real	real	ADJ
easat-5356	80	22	-	-	PUNCT
easat-5356	80	23	world	world	NOUN
easat-5356	80	24	scenarios	scenario	NOUN
easat-5356	80	25	,	,	PUNCT
easat-5356	80	26	particularly	particularly	ADV
easat-5356	80	27	those	those	PRON
easat-5356	80	28	with	with	ADP
easat-5356	80	29	stringent	stringent	ADJ
easat-5356	80	30	memory	memory	NOUN
easat-5356	80	31	and	and	CCONJ
easat-5356	80	32	latency	latency	NOUN
easat-5356	80	33	constraints	constraint	NOUN
easat-5356	80	34	.	.	PUNCT
easat-5356	81	1	to	to	PART
easat-5356	81	2	handle	handle	VERB
easat-5356	81	3	these	these	DET
easat-5356	81	4	problems	problem	NOUN
easat-5356	81	5	,	,	PUNCT
easat-5356	81	6	we	we	PRON
easat-5356	81	7	propose	propose	VERB
easat-5356	81	8	a	a	DET
easat-5356	81	9	novel	novel	ADJ
easat-5356	81	10	model	model	NOUN
easat-5356	81	11	called	call	VERB
easat-5356	81	12	atw	atw	PROPN
easat-5356	81	13	,	,	PUNCT
easat-5356	81	14	designed	design	VERB
easat-5356	81	15	to	to	PART
easat-5356	81	16	alleviate	alleviate	VERB
easat-5356	81	17	the	the	DET
easat-5356	81	18	problem	problem	NOUN
easat-5356	81	19	of	of	ADP
easat-5356	81	20	unbalanced	unbalanced	ADJ
easat-5356	81	21	samples	sample	NOUN
easat-5356	81	22	.	.	PUNCT
easat-5356	82	1	atw	atw	PROPN
easat-5356	82	2	enables	enable	VERB
easat-5356	82	3	the	the	DET
easat-5356	82	4	model	model	NOUN
easat-5356	82	5	to	to	PART
easat-5356	82	6	adaptively	adaptively	ADV
easat-5356	82	7	learn	learn	VERB
easat-5356	82	8	the	the	DET
easat-5356	82	9	threshold	threshold	NOUN
easat-5356	82	10	parameters	parameter	NOUN
easat-5356	82	11	for	for	ADP
easat-5356	82	12	positive	positive	ADJ
easat-5356	82	13	and	and	CCONJ
easat-5356	82	14	negative	negative	ADJ
easat-5356	82	15	samples	sample	NOUN
easat-5356	82	16	,	,	PUNCT
easat-5356	82	17	reducing	reduce	VERB
easat-5356	82	18	reliance	reliance	NOUN
easat-5356	82	19	on	on	ADP
easat-5356	82	20	manual	manual	ADJ
easat-5356	82	21	tuning	tuning	NOUN
easat-5356	82	22	.	.	PUNCT
easat-5356	83	1	additionally	additionally	ADV
easat-5356	83	2	,	,	PUNCT
easat-5356	83	3	we	we	PRON
easat-5356	83	4	incorporate	incorporate	VERB
easat-5356	83	5	the	the	DET
easat-5356	83	6	wiouv3	wiouv3	NOUN
easat-5356	83	7	tong	tong	PROPN
easat-5356	83	8	,	,	PUNCT
easat-5356	83	9	et	et	PROPN
easat-5356	83	10	al	al	PROPN
easat-5356	83	11	.	.	PUNCT
easat-5356	84	1	[	[	X
easat-5356	84	2	19	19	NUM
easat-5356	84	3	]	]	PUNCT
easat-5356	84	4	loss	loss	NOUN
easat-5356	84	5	to	to	PART
easat-5356	84	6	address	address	VERB
easat-5356	84	7	the	the	DET
easat-5356	84	8	issue	issue	NOUN
easat-5356	84	9	of	of	ADP
easat-5356	84	10	slight	slight	ADJ
easat-5356	84	11	intersection	intersection	NOUN
easat-5356	84	12	over	over	ADP
easat-5356	84	13	union	union	NOUN
easat-5356	84	14	(	(	PUNCT
easat-5356	84	15	iou	iou	NOUN
easat-5356	84	16	)	)	PUNCT
easat-5356	84	17	bias	bias	NOUN
easat-5356	84	18	,	,	PUNCT
easat-5356	84	19	which	which	PRON
easat-5356	84	20	is	be	AUX
easat-5356	84	21	especially	especially	ADV
easat-5356	84	22	problematic	problematic	ADJ
easat-5356	84	23	in	in	ADP
easat-5356	84	24	small	small	ADJ
easat-5356	84	25	-	-	PUNCT
easat-5356	84	26	scale	scale	NOUN
easat-5356	84	27	face	face	NOUN
easat-5356	84	28	detection	detection	NOUN
easat-5356	84	29	.	.	PUNCT
easat-5356	85	1	these	these	DET
easat-5356	85	2	innovations	innovation	NOUN
easat-5356	85	3	improve	improve	VERB
easat-5356	85	4	the	the	DET
easat-5356	85	5	efficiency	efficiency	NOUN
easat-5356	85	6	and	and	CCONJ
easat-5356	85	7	accuracy	accuracy	NOUN
easat-5356	85	8	of	of	ADP
easat-5356	85	9	face	face	NOUN
easat-5356	85	10	detection	detection	NOUN
easat-5356	85	11	while	while	SCONJ
easat-5356	85	12	ensuring	ensure	VERB
easat-5356	85	13	that	that	SCONJ
easat-5356	85	14	the	the	DET
easat-5356	85	15	model	model	NOUN
easat-5356	85	16	can	can	AUX
easat-5356	85	17	be	be	AUX
easat-5356	85	18	deployed	deploy	VERB
easat-5356	85	19	in	in	ADP
easat-5356	85	20	resourceconstrained	resourceconstraine	VERB
easat-5356	85	21	environments	environment	NOUN
easat-5356	85	22	.	.	PUNCT
easat-5356	86	1	all	all	ADV
easat-5356	86	2	in	in	ADP
easat-5356	86	3	all	all	PRON
easat-5356	86	4	,	,	PUNCT
easat-5356	86	5	our	our	PRON
easat-5356	86	6	main	main	ADJ
easat-5356	86	7	efforts	effort	NOUN
easat-5356	86	8	are	be	AUX
easat-5356	86	9	:	:	PUNCT
easat-5356	86	10	1.a	1.a	NUM
easat-5356	86	11	new	new	ADJ
easat-5356	86	12	feature	feature	NOUN
easat-5356	86	13	extraction	extraction	NOUN
easat-5356	86	14	module	module	NOUN
easat-5356	86	15	which	which	PRON
easat-5356	86	16	leverages	leverage	VERB
easat-5356	86	17	se	se	NOUN
easat-5356	86	18	mechanisms	mechanism	NOUN
easat-5356	86	19	and	and	CCONJ
easat-5356	86	20	spatial	spatial	ADJ
easat-5356	86	21	attention	attention	NOUN
easat-5356	86	22	.	.	PUNCT
easat-5356	87	1	2.introduced	2.introduce	VERB
easat-5356	87	2	the	the	DET
easat-5356	87	3	wiouv3	wiouv3	NOUN
easat-5356	87	4	loss	loss	NOUN
easat-5356	87	5	function	function	NOUN
easat-5356	87	6	to	to	PART
easat-5356	87	7	enhance	enhance	VERB
easat-5356	87	8	bounding	bound	VERB
easat-5356	87	9	box	box	NOUN
easat-5356	87	10	regression	regression	NOUN
easat-5356	87	11	and	and	CCONJ
easat-5356	87	12	applied	apply	VERB
easat-5356	87	13	an	an	DET
easat-5356	87	14	adaptive	adaptive	ADJ
easat-5356	87	15	anchor	anchor	NOUN
easat-5356	87	16	frame	frame	NOUN
easat-5356	87	17	filtering	filter	VERB
easat-5356	87	18	algorithm	algorithm	NOUN
easat-5356	87	19	to	to	PART
easat-5356	87	20	address	address	VERB
easat-5356	87	21	sample	sample	NOUN
easat-5356	87	22	imbalance	imbalance	NOUN
easat-5356	87	23	,	,	PUNCT
easat-5356	87	24	improving	improve	VERB
easat-5356	87	25	the	the	DET
easat-5356	87	26	model	model	NOUN
easat-5356	87	27	's	's	PART
easat-5356	87	28	performance	performance	NOUN
easat-5356	87	29	on	on	ADP
easat-5356	87	30	challenging	challenge	VERB
easat-5356	87	31	detection	detection	NOUN
easat-5356	87	32	tasks	task	NOUN
easat-5356	87	33	.	.	PUNCT
easat-5356	88	1	3.to	3.to	NUM
easat-5356	88	2	enhance	enhance	VERB
easat-5356	88	3	small	small	ADJ
easat-5356	88	4	item	item	NOUN
easat-5356	88	5	identification	identification	NOUN
easat-5356	88	6	,	,	PUNCT
easat-5356	88	7	a	a	DET
easat-5356	88	8	tiny	tiny	ADJ
easat-5356	88	9	detection	detection	NOUN
easat-5356	88	10	head	head	NOUN
easat-5356	88	11	is	be	AUX
easat-5356	88	12	added	add	VERB
easat-5356	88	13	.	.	PUNCT
easat-5356	89	1	2	2	X
easat-5356	89	2	.	.	X
easat-5356	89	3	related	relate	VERB
easat-5356	89	4	work	work	NOUN
easat-5356	89	5	for	for	ADP
easat-5356	89	6	many	many	ADJ
easat-5356	89	7	years	year	NOUN
easat-5356	89	8	,	,	PUNCT
easat-5356	89	9	computer	computer	NOUN
easat-5356	89	10	vision	vision	NOUN
easat-5356	89	11	research	research	NOUN
easat-5356	89	12	has	have	AUX
easat-5356	89	13	focused	focus	VERB
easat-5356	89	14	heavily	heavily	ADV
easat-5356	89	15	on	on	ADP
easat-5356	89	16	facial	facial	ADJ
easat-5356	89	17	detection	detection	NOUN
easat-5356	89	18	.	.	PUNCT
easat-5356	90	1	handcrafted	handcraft	VERB
easat-5356	90	2	features	feature	NOUN
easat-5356	90	3	,	,	PUNCT
easat-5356	90	4	like	like	ADP
easat-5356	90	5	haar	haar	PROPN
easat-5356	90	6	features	feature	NOUN
easat-5356	90	7	,	,	PUNCT
easat-5356	90	8	were	be	AUX
easat-5356	90	9	the	the	DET
easat-5356	90	10	mainstay	mainstay	NOUN
easat-5356	90	11	of	of	ADP
easat-5356	90	12	early	early	ADJ
easat-5356	90	13	facial	facial	ADJ
easat-5356	90	14	detection	detection	NOUN
easat-5356	90	15	techniques	technique	NOUN
easat-5356	90	16	alrammahi	alrammahi	ADV
easat-5356	90	17	and	and	CCONJ
easat-5356	90	18	jabbar	jabbar	ADJ
easat-5356	90	19	[	[	X
easat-5356	90	20	20	20	NUM
easat-5356	90	21	]	]	PUNCT
easat-5356	90	22	control	control	NOUN
easat-5356	90	23	point	point	NOUN
easat-5356	90	24	sets	set	NOUN
easat-5356	90	25	,	,	PUNCT
easat-5356	90	26	and	and	CCONJ
easat-5356	90	27	edge	edge	NOUN
easat-5356	90	28	direction	direction	NOUN
easat-5356	90	29	histograms	histogram	NOUN
easat-5356	90	30	.	.	PUNCT
easat-5356	91	1	face	face	NOUN
easat-5356	91	2	detection	detection	NOUN
easat-5356	91	3	methods	method	NOUN
easat-5356	91	4	started	start	VERB
easat-5356	91	5	using	use	VERB
easat-5356	91	6	neural	neural	ADJ
easat-5356	91	7	networks	network	NOUN
easat-5356	91	8	for	for	ADP
easat-5356	91	9	automatic	automatic	ADJ
easat-5356	91	10	feature	feature	NOUN
easat-5356	91	11	extraction	extraction	NOUN
easat-5356	91	12	and	and	CCONJ
easat-5356	91	13	classification	classification	NOUN
easat-5356	91	14	with	with	ADP
easat-5356	91	15	the	the	DET
easat-5356	91	16	introduction	introduction	NOUN
easat-5356	91	17	of	of	ADP
easat-5356	91	18	deep	deep	ADJ
easat-5356	91	19	learning	learning	NOUN
easat-5356	91	20	.	.	PUNCT
easat-5356	92	1	for	for	ADP
easat-5356	92	2	instance	instance	NOUN
easat-5356	92	3	,	,	PUNCT
easat-5356	92	4	cascadecnn	cascadecnn	VERB
easat-5356	92	5	venieris	venieris	PROPN
easat-5356	92	6	,	,	PUNCT
easat-5356	92	7	et	et	PROPN
easat-5356	92	8	al	al	PROPN
easat-5356	92	9	.	.	PUNCT
easat-5356	93	1	[	[	X
easat-5356	93	2	21	21	NUM
easat-5356	93	3	]	]	PUNCT
easat-5356	93	4	presented	present	VERB
easat-5356	93	5	a	a	DET
easat-5356	93	6	cascade	cascade	NOUN
easat-5356	93	7	structure	structure	NOUN
easat-5356	93	8	made	make	VERB
easat-5356	93	9	up	up	ADP
easat-5356	93	10	of	of	ADP
easat-5356	93	11	three	three	NUM
easat-5356	93	12	meticulously	meticulously	ADV
easat-5356	93	13	crafted	craft	VERB
easat-5356	93	14	deep	deep	ADJ
easat-5356	93	15	convolutional	convolutional	ADJ
easat-5356	93	16	networks	network	NOUN
easat-5356	93	17	that	that	PRON
easat-5356	93	18	made	make	VERB
easat-5356	93	19	coarse	coarse	NOUN
easat-5356	93	20	-	-	PUNCT
easat-5356	93	21	to	to	ADP
easat-5356	93	22	-	-	PUNCT
easat-5356	93	23	fine	fine	ADJ
easat-5356	93	24	predictions	prediction	NOUN
easat-5356	93	25	about	about	ADP
easat-5356	93	26	landmark	landmark	NOUN
easat-5356	93	27	and	and	CCONJ
easat-5356	93	28	facial	facial	ADJ
easat-5356	93	29	positions	position	NOUN
easat-5356	93	30	.	.	PUNCT
easat-5356	94	1	similarly	similarly	ADV
easat-5356	94	2	,	,	PUNCT
easat-5356	94	3	mtcnn	mtcnn	PROPN
easat-5356	94	4	khan	khan	PROPN
easat-5356	94	5	,	,	PUNCT
easat-5356	94	6	et	et	PROPN
easat-5356	94	7	al	al	PROPN
easat-5356	94	8	.	.	PUNCT
easat-5356	95	1	[	[	X
easat-5356	95	2	22	22	NUM
easat-5356	95	3	]	]	PUNCT
easat-5356	95	4	employed	employ	VERB
easat-5356	95	5	a	a	DET
easat-5356	95	6	cascading	cascade	VERB
easat-5356	95	7	architecture	architecture	NOUN
easat-5356	95	8	that	that	PRON
easat-5356	95	9	enabled	enable	VERB
easat-5356	95	10	joint	joint	ADJ
easat-5356	95	11	alignment	alignment	NOUN
easat-5356	95	12	of	of	ADP
easat-5356	95	13	facial	facial	ADJ
easat-5356	95	14	landmarks	landmark	NOUN
easat-5356	95	15	and	and	CCONJ
easat-5356	95	16	detection	detection	NOUN
easat-5356	95	17	of	of	ADP
easat-5356	95	18	facial	facial	ADJ
easat-5356	95	19	positions	position	NOUN
easat-5356	95	20	.	.	PUNCT
easat-5356	96	1	the	the	DET
easat-5356	96	2	pcn	pcn	PROPN
easat-5356	96	3	kaewkorn	kaewkorn	PROPN
easat-5356	96	4	,	,	PUNCT
easat-5356	96	5	et	et	PROPN
easat-5356	96	6	al	al	PROPN
easat-5356	96	7	.	.	PUNCT
easat-5356	97	1	[	[	X
easat-5356	97	2	23	23	NUM
easat-5356	97	3	]	]	PUNCT
easat-5356	97	4	on	on	ADP
easat-5356	97	5	the	the	DET
easat-5356	97	6	other	other	ADJ
easat-5356	97	7	hand	hand	NOUN
easat-5356	97	8	,	,	PUNCT
easat-5356	97	9	used	use	VERB
easat-5356	97	10	a	a	DET
easat-5356	97	11	direction	direction	NOUN
easat-5356	97	12	prediction	prediction	NOUN
easat-5356	97	13	network	network	NOUN
easat-5356	97	14	to	to	PART
easat-5356	97	15	correct	correct	VERB
easat-5356	97	16	facial	facial	ADJ
easat-5356	97	17	orientations	orientation	NOUN
easat-5356	97	18	,	,	PUNCT
easat-5356	97	19	further	far	ADV
easat-5356	97	20	improving	improve	VERB
easat-5356	97	21	detection	detection	NOUN
easat-5356	97	22	accuracy	accuracy	NOUN
easat-5356	97	23	.	.	PUNCT
easat-5356	98	1	these	these	DET
easat-5356	98	2	early	early	ADJ
easat-5356	98	3	advancements	advancement	NOUN
easat-5356	98	4	laid	lay	VERB
easat-5356	98	5	the	the	DET
easat-5356	98	6	groundwork	groundwork	NOUN
easat-5356	98	7	for	for	ADP
easat-5356	98	8	modern	modern	ADJ
easat-5356	98	9	,	,	PUNCT
easat-5356	98	10	more	more	ADV
easat-5356	98	11	robust	robust	ADJ
easat-5356	98	12	face	face	NOUN
easat-5356	98	13	detection	detection	NOUN
easat-5356	98	14	methods	method	NOUN
easat-5356	98	15	.	.	PUNCT
easat-5356	99	1	early	early	ADJ
easat-5356	99	2	deep	deep	ADJ
easat-5356	99	3	learning	learning	NOUN
easat-5356	99	4	-	-	PUNCT
easat-5356	99	5	based	base	VERB
easat-5356	99	6	facial	facial	ADJ
easat-5356	99	7	detection	detection	NOUN
easat-5356	99	8	techniques	technique	NOUN
easat-5356	99	9	,	,	PUNCT
easat-5356	99	10	however	however	ADV
easat-5356	99	11	,	,	PUNCT
easat-5356	99	12	had	have	VERB
easat-5356	99	13	a	a	DET
easat-5356	99	14	number	number	NOUN
easat-5356	99	15	of	of	ADP
easat-5356	99	16	shortcomings	shortcoming	NOUN
easat-5356	99	17	,	,	PUNCT
easat-5356	99	18	for	for	ADP
easat-5356	99	19	instance	instance	NOUN
easat-5356	99	20	laborious	laborious	ADJ
easat-5356	99	21	training	training	NOUN
easat-5356	99	22	procedures	procedure	NOUN
easat-5356	99	23	,	,	PUNCT
easat-5356	99	24	a	a	DET
easat-5356	99	25	vulnerability	vulnerability	NOUN
easat-5356	99	26	to	to	ADP
easat-5356	99	27	local	local	ADJ
easat-5356	99	28	optima	optima	NOUN
easat-5356	99	29	,	,	PUNCT
easat-5356	99	30	sluggish	sluggish	ADJ
easat-5356	99	31	detection	detection	NOUN
easat-5356	99	32	speeds	speed	NOUN
easat-5356	99	33	,	,	PUNCT
easat-5356	99	34	and	and	CCONJ
easat-5356	99	35	poor	poor	ADJ
easat-5356	99	36	accuracy	accuracy	NOUN
easat-5356	99	37	.	.	PUNCT
easat-5356	100	1	the	the	DET
easat-5356	100	2	sliding	slide	VERB
easat-5356	100	3	window	window	NOUN
easat-5356	100	4	paradigm	paradigm	NOUN
easat-5356	100	5	,	,	PUNCT
easat-5356	100	6	which	which	PRON
easat-5356	100	7	applies	apply	VERB
easat-5356	100	8	classifiers	classifier	NOUN
easat-5356	100	9	to	to	ADP
easat-5356	100	10	dense	dense	ADJ
easat-5356	100	11	grids	grid	NOUN
easat-5356	100	12	of	of	ADP
easat-5356	100	13	images	image	NOUN
easat-5356	100	14	,	,	PUNCT
easat-5356	100	15	has	have	VERB
easat-5356	100	16	roots	root	NOUN
easat-5356	100	17	extending	extend	VERB
easat-5356	100	18	back	back	ADV
easat-5356	100	19	several	several	ADJ
easat-5356	100	20	decades	decade	NOUN
easat-5356	100	21	.	.	PUNCT
easat-5356	101	1	the	the	DET
easat-5356	101	2	landmark	landmark	PROPN
easat-5356	101	3	study	study	NOUN
easat-5356	101	4	by	by	ADP
easat-5356	101	5	viola	viola	PROPN
easat-5356	101	6	-	-	PUNCT
easat-5356	101	7	jones	jones	PROPN
easat-5356	101	8	[	[	X
easat-5356	101	9	24	24	NUM
easat-5356	101	10	]	]	PUNCT
easat-5356	101	11	introduced	introduce	VERB
easat-5356	101	12	a	a	DET
easat-5356	101	13	cascade	cascade	NOUN
easat-5356	101	14	framework	framework	NOUN
easat-5356	101	15	that	that	PRON
easat-5356	101	16	effectively	effectively	ADV
easat-5356	101	17	eliminated	eliminate	VERB
easat-5356	101	18	erroneous	erroneous	ADJ
easat-5356	101	19	facial	facial	ADJ
easat-5356	101	20	regions	region	NOUN
easat-5356	101	21	within	within	ADP
easat-5356	101	22	image	image	NOUN
easat-5356	101	23	pyramids	pyramid	NOUN
easat-5356	101	24	in	in	ADP
easat-5356	101	25	real	real	ADJ
easat-5356	101	26	-	-	PUNCT
easat-5356	101	27	time	time	NOUN
easat-5356	101	28	,	,	PUNCT
easat-5356	101	29	leading	lead	VERB
easat-5356	101	30	to	to	ADP
easat-5356	101	31	the	the	DET
easat-5356	101	32	widespread	widespread	ADJ
easat-5356	101	33	adoption	adoption	NOUN
easat-5356	101	34	of	of	ADP
easat-5356	101	35	this	this	DET
easat-5356	101	36	scale	scale	NOUN
easat-5356	101	37	-	-	PUNCT
easat-5356	101	38	invariant	invariant	ADJ
easat-5356	101	39	face	face	NOUN
easat-5356	101	40	detection	detection	NOUN
easat-5356	101	41	method	method	NOUN
easat-5356	101	42	.	.	PUNCT
easat-5356	102	1	while	while	SCONJ
easat-5356	102	2	the	the	DET
easat-5356	102	3	sliding	slide	VERB
easat-5356	102	4	window	window	NOUN
easat-5356	102	5	approach	approach	NOUN
easat-5356	102	6	aggarwal	aggarwal	NOUN
easat-5356	102	7	,	,	PUNCT
easat-5356	102	8	et	et	PROPN
easat-5356	102	9	al	al	PROPN
easat-5356	102	10	.	.	PUNCT
easat-5356	103	1	[	[	X
easat-5356	103	2	25	25	NUM
easat-5356	103	3	]	]	PUNCT
easat-5356	103	4	on	on	ADP
easat-5356	103	5	image	image	NOUN
easat-5356	103	6	pyramids	pyramid	NOUN
easat-5356	103	7	had	have	AUX
easat-5356	103	8	long	long	ADV
easat-5356	103	9	been	be	AUX
easat-5356	103	10	the	the	DET
easat-5356	103	11	dominant	dominant	ADJ
easat-5356	103	12	paradigm	paradigm	NOUN
easat-5356	103	13	for	for	ADP
easat-5356	103	14	detection	detection	NOUN
easat-5356	103	15	,	,	PUNCT
easat-5356	103	16	the	the	DET
easat-5356	103	17	introduction	introduction	NOUN
easat-5356	103	18	of	of	ADP
easat-5356	103	19	feature	feature	NOUN
easat-5356	103	20	pyramids	pyramid	NOUN
easat-5356	103	21	zhao	zhao	NOUN
easat-5356	103	22	,	,	PUNCT
easat-5356	103	23	et	et	PROPN
easat-5356	103	24	al	al	PROPN
easat-5356	103	25	.	.	PUNCT
easat-5356	104	1	[	[	X
easat-5356	104	2	26	26	NUM
easat-5356	104	3	]	]	PUNCT
easat-5356	104	4	has	have	AUX
easat-5356	104	5	rapidly	rapidly	ADV
easat-5356	104	6	shifted	shift	VERB
easat-5356	104	7	the	the	DET
easat-5356	104	8	focus	focus	NOUN
easat-5356	104	9	towards	towards	ADP
easat-5356	104	10	sliding	slide	VERB
easat-5356	104	11	anchors	anchor	NOUN
easat-5356	104	12	on	on	ADP
easat-5356	104	13	multi	multi	ADJ
easat-5356	104	14	-	-	ADJ
easat-5356	104	15	scale	scale	ADJ
easat-5356	104	16	feature	feature	NOUN
easat-5356	104	17	maps	map	NOUN
easat-5356	104	18	.	.	PUNCT
easat-5356	105	1	this	this	DET
easat-5356	105	2	transition	transition	NOUN
easat-5356	105	3	allows	allow	VERB
easat-5356	105	4	for	for	ADP
easat-5356	105	5	more	more	ADV
easat-5356	105	6	efficient	efficient	ADJ
easat-5356	105	7	and	and	CCONJ
easat-5356	105	8	accurate	accurate	ADJ
easat-5356	105	9	face	face	NOUN
easat-5356	105	10	detection	detection	NOUN
easat-5356	105	11	by	by	ADP
easat-5356	105	12	leveraging	leverage	VERB
easat-5356	105	13	rich	rich	ADJ
easat-5356	105	14	hierarchical	hierarchical	ADJ
easat-5356	105	15	features	feature	NOUN
easat-5356	105	16	,	,	PUNCT
easat-5356	105	17	enabling	enable	VERB
easat-5356	105	18	better	well	ADJ
easat-5356	105	19	handling	handling	NOUN
easat-5356	105	20	of	of	ADP
easat-5356	105	21	faces	face	NOUN
easat-5356	105	22	at	at	ADP
easat-5356	105	23	various	various	ADJ
easat-5356	105	24	scales	scale	NOUN
easat-5356	105	25	and	and	CCONJ
easat-5356	105	26	improving	improve	VERB
easat-5356	105	27	overall	overall	ADJ
easat-5356	105	28	detection	detection	NOUN
easat-5356	105	29	performance	performance	NOUN
easat-5356	105	30	.	.	PUNCT
easat-5356	106	1	current	current	ADJ
easat-5356	106	2	facial	facial	ADJ
easat-5356	106	3	detection	detection	NOUN
easat-5356	106	4	methods	method	NOUN
easat-5356	106	5	build	build	VERB
easat-5356	106	6	upon	upon	SCONJ
easat-5356	106	7	advancements	advancement	NOUN
easat-5356	106	8	from	from	ADP
easat-5356	106	9	general	general	ADJ
easat-5356	106	10	object	object	NOUN
easat-5356	106	11	detection	detection	NOUN
easat-5356	106	12	techniques	technique	NOUN
easat-5356	106	13	and	and	CCONJ
easat-5356	106	14	are	be	AUX
easat-5356	106	15	typically	typically	ADV
easat-5356	106	16	categorized	categorize	VERB
easat-5356	106	17	into	into	ADP
easat-5356	106	18	two	two	NUM
easat-5356	106	19	main	main	ADJ
easat-5356	106	20	classes	class	NOUN
easat-5356	106	21	:	:	PUNCT
easat-5356	106	22	two	two	NUM
easat-5356	106	23	-	-	PUNCT
easat-5356	106	24	stage	stage	NOUN
easat-5356	106	25	methods	method	NOUN
easat-5356	106	26	(	(	PUNCT
easat-5356	106	27	e.g.	e.g.	ADV
easat-5356	106	28	,	,	PUNCT
easat-5356	106	29	faster	fast	ADJ
easat-5356	106	30	r	r	NOUN
easat-5356	106	31	-	-	PUNCT
easat-5356	106	32	cnn	cnn	NOUN
easat-5356	106	33	)	)	PUNCT
easat-5356	106	34	and	and	CCONJ
easat-5356	106	35	singlestage	singlestage	NOUN
easat-5356	106	36	methods	method	NOUN
easat-5356	106	37	(	(	PUNCT
easat-5356	106	38	e.g.	e.g.	ADV
easat-5356	106	39	,	,	PUNCT
easat-5356	106	40	ssd	ssd	NOUN
easat-5356	106	41	and	and	CCONJ
easat-5356	106	42	retinanet	retinanet	NOUN
easat-5356	106	43	)	)	PUNCT
easat-5356	106	44	.	.	PUNCT
easat-5356	107	1	two	two	NUM
easat-5356	107	2	-	-	PUNCT
easat-5356	107	3	stage	stage	NOUN
easat-5356	107	4	methods	method	NOUN
easat-5356	107	5	employ	employ	VERB
easat-5356	107	6	a	a	DET
easat-5356	107	7	"	"	PUNCT
easat-5356	107	8	proposal	proposal	NOUN
easat-5356	107	9	and	and	CCONJ
easat-5356	107	10	refinement	refinement	NOUN
easat-5356	107	11	"	"	PUNCT
easat-5356	107	12	mechanism	mechanism	NOUN
easat-5356	107	13	,	,	PUNCT
easat-5356	107	14	providing	provide	VERB
easat-5356	107	15	higher	high	ADJ
easat-5356	107	16	localization	localization	NOUN
easat-5356	107	17	accuracy	accuracy	NOUN
easat-5356	107	18	[	[	X
easat-5356	107	19	27	27	NUM
easat-5356	107	20	]	]	PUNCT
easat-5356	107	21	.	.	PUNCT
easat-5356	108	1	single	single	ADJ
easat-5356	108	2	-	-	PUNCT
easat-5356	108	3	stage	stage	NOUN
easat-5356	108	4	approaches	approach	NOUN
easat-5356	108	5	,	,	PUNCT
easat-5356	108	6	on	on	ADP
easat-5356	108	7	the	the	DET
easat-5356	108	8	other	other	ADJ
easat-5356	108	9	hand	hand	NOUN
easat-5356	108	10	,	,	PUNCT
easat-5356	108	11	sample	sample	VERB
easat-5356	108	12	facial	facial	ADJ
easat-5356	108	13	positions	position	NOUN
easat-5356	108	14	and	and	CCONJ
easat-5356	108	15	scales	scale	NOUN
easat-5356	108	16	densely	densely	ADV
easat-5356	108	17	,	,	PUNCT
easat-5356	108	18	which	which	PRON
easat-5356	108	19	frequently	frequently	ADV
easat-5356	108	20	leads	lead	VERB
easat-5356	108	21	to	to	ADP
easat-5356	108	22	a	a	DET
easat-5356	108	23	notable	notable	ADJ
easat-5356	108	24	imbalance	imbalance	NOUN
easat-5356	108	25	between	between	ADP
easat-5356	108	26	positive	positive	ADJ
easat-5356	108	27	and	and	CCONJ
easat-5356	108	28	negative	negative	ADJ
easat-5356	108	29	data	datum	NOUN
easat-5356	108	30	during	during	ADP
easat-5356	108	31	training	training	NOUN
easat-5356	108	32	.	.	PUNCT
easat-5356	109	1	this	this	DET
easat-5356	109	2	imbalance	imbalance	NOUN
easat-5356	109	3	arises	arise	VERB
easat-5356	109	4	because	because	SCONJ
easat-5356	109	5	the	the	DET
easat-5356	109	6	majority	majority	NOUN
easat-5356	109	7	of	of	ADP
easat-5356	109	8	anchor	anchor	NOUN
easat-5356	109	9	points	point	NOUN
easat-5356	109	10	correspond	correspond	VERB
easat-5356	109	11	to	to	ADP
easat-5356	109	12	background	background	NOUN
easat-5356	109	13	regions	region	NOUN
easat-5356	109	14	,	,	PUNCT
easat-5356	109	15	while	while	SCONJ
easat-5356	109	16	only	only	ADV
easat-5356	109	17	a	a	DET
easat-5356	109	18	small	small	ADJ
easat-5356	109	19	fraction	fraction	NOUN
easat-5356	109	20	corresponds	correspond	VERB
easat-5356	109	21	to	to	ADP
easat-5356	109	22	actual	actual	ADJ
easat-5356	109	23	faces	face	NOUN
easat-5356	109	24	,	,	PUNCT
easat-5356	109	25	leading	lead	VERB
easat-5356	109	26	to	to	ADP
easat-5356	109	27	a	a	DET
easat-5356	109	28	dominance	dominance	NOUN
easat-5356	109	29	of	of	ADP
easat-5356	109	30	easy	easy	ADJ
easat-5356	109	31	negative	negative	ADJ
easat-5356	109	32	samples	sample	NOUN
easat-5356	109	33	.	.	PUNCT
easat-5356	110	1	this	this	DET
easat-5356	110	2	issue	issue	NOUN
easat-5356	110	3	can	can	AUX
easat-5356	110	4	hinder	hinder	VERB
easat-5356	110	5	the	the	DET
easat-5356	110	6	model	model	NOUN
easat-5356	110	7	's	's	PART
easat-5356	110	8	ability	ability	NOUN
easat-5356	110	9	to	to	PART
easat-5356	110	10	learn	learn	VERB
easat-5356	110	11	effectively	effectively	ADV
easat-5356	110	12	,	,	PUNCT
easat-5356	110	13	843	843	NUM
easat-5356	110	14	edelweiss	edelweiss	PROPN
easat-5356	110	15	applied	apply	VERB
easat-5356	110	16	science	science	NOUN
easat-5356	110	17	and	and	CCONJ
easat-5356	110	18	technology	technology	NOUN
easat-5356	110	19	issn	issn	PROPN
easat-5356	110	20	:	:	PUNCT
easat-5356	110	21	2576	2576	NUM
easat-5356	110	22	-	-	SYM
easat-5356	110	23	8484	8484	NUM
easat-5356	110	24	vol	vol	NOUN
easat-5356	110	25	.	.	PROPN
easat-5356	111	1	9	9	NUM
easat-5356	111	2	,	,	PUNCT
easat-5356	111	3	no	no	INTJ
easat-5356	111	4	.	.	NOUN
easat-5356	112	1	3	3	NUM
easat-5356	112	2	:	:	PUNCT
easat-5356	112	3	840	840	NUM
easat-5356	112	4	-	-	SYM
easat-5356	112	5	855	855	NUM
easat-5356	112	6	,	,	PUNCT
easat-5356	112	7	2025	2025	NUM
easat-5356	112	8	doi	doi	NOUN
easat-5356	112	9	:	:	PUNCT
easat-5356	112	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	112	11	©	©	PROPN
easat-5356	112	12	2025	2025	NUM
easat-5356	112	13	by	by	ADP
easat-5356	112	14	the	the	DET
easat-5356	112	15	authors	author	NOUN
easat-5356	112	16	;	;	PUNCT
easat-5356	112	17	licensee	licensee	PROPN
easat-5356	112	18	learning	learning	NOUN
easat-5356	112	19	gate	gate	NOUN
easat-5356	112	20	particularly	particularly	ADV
easat-5356	112	21	for	for	ADP
easat-5356	112	22	detecting	detect	VERB
easat-5356	112	23	smaller	small	ADJ
easat-5356	112	24	or	or	CCONJ
easat-5356	112	25	more	more	ADV
easat-5356	112	26	difficult	difficult	ADJ
easat-5356	112	27	faces	face	NOUN
easat-5356	112	28	,	,	PUNCT
easat-5356	112	29	and	and	CCONJ
easat-5356	112	30	may	may	AUX
easat-5356	112	31	require	require	VERB
easat-5356	112	32	additional	additional	ADJ
easat-5356	112	33	strategies	strategy	NOUN
easat-5356	112	34	,	,	PUNCT
easat-5356	112	35	such	such	ADJ
easat-5356	112	36	as	as	ADP
easat-5356	112	37	loss	loss	NOUN
easat-5356	112	38	weighting	weighting	NOUN
easat-5356	112	39	or	or	CCONJ
easat-5356	112	40	hard	hard	ADJ
easat-5356	112	41	sample	sample	NOUN
easat-5356	112	42	mining	mining	NOUN
easat-5356	112	43	,	,	PUNCT
easat-5356	112	44	to	to	PART
easat-5356	112	45	improve	improve	VERB
easat-5356	112	46	performance	performance	NOUN
easat-5356	112	47	.	.	PUNCT
easat-5356	113	1	to	to	PART
easat-5356	113	2	address	address	VERB
easat-5356	113	3	this	this	DET
easat-5356	113	4	imbalance	imbalance	NOUN
easat-5356	113	5	,	,	PUNCT
easat-5356	113	6	sampling	sample	VERB
easat-5356	113	7	and	and	CCONJ
easat-5356	113	8	weighting	weight	VERB
easat-5356	113	9	techniques	technique	NOUN
easat-5356	113	10	have	have	AUX
easat-5356	113	11	been	be	AUX
easat-5356	113	12	widely	widely	ADV
easat-5356	113	13	adopted	adopt	VERB
easat-5356	113	14	.	.	PUNCT
easat-5356	114	1	compared	compare	VERB
easat-5356	114	2	to	to	ADP
easat-5356	114	3	two	two	NUM
easat-5356	114	4	-	-	PUNCT
easat-5356	114	5	stage	stage	NOUN
easat-5356	114	6	methods	method	NOUN
easat-5356	114	7	,	,	PUNCT
easat-5356	114	8	single	single	ADJ
easat-5356	114	9	-	-	PUNCT
easat-5356	114	10	stage	stage	NOUN
easat-5356	114	11	approaches	approach	NOUN
easat-5356	114	12	are	be	AUX
easat-5356	114	13	more	more	ADV
easat-5356	114	14	efficient	efficient	ADJ
easat-5356	114	15	and	and	CCONJ
easat-5356	114	16	typically	typically	ADV
easat-5356	114	17	offer	offer	VERB
easat-5356	114	18	higher	high	ADJ
easat-5356	114	19	recall	recall	NOUN
easat-5356	114	20	rates	rate	NOUN
easat-5356	114	21	,	,	PUNCT
easat-5356	114	22	as	as	SCONJ
easat-5356	114	23	they	they	PRON
easat-5356	114	24	perform	perform	VERB
easat-5356	114	25	detection	detection	NOUN
easat-5356	114	26	in	in	ADP
easat-5356	114	27	a	a	DET
easat-5356	114	28	single	single	ADJ
easat-5356	114	29	pass	pass	NOUN
easat-5356	114	30	,	,	PUNCT
easat-5356	114	31	making	make	VERB
easat-5356	114	32	them	they	PRON
easat-5356	114	33	well	well	ADV
easat-5356	114	34	-	-	PUNCT
easat-5356	114	35	suited	suit	VERB
easat-5356	114	36	for	for	ADP
easat-5356	114	37	real	real	ADJ
easat-5356	114	38	-	-	PUNCT
easat-5356	114	39	time	time	NOUN
easat-5356	114	40	applications	application	NOUN
easat-5356	114	41	.	.	PUNCT
easat-5356	115	1	but	but	CCONJ
easat-5356	115	2	this	this	DET
easat-5356	115	3	effectiveness	effectiveness	NOUN
easat-5356	115	4	may	may	AUX
easat-5356	115	5	come	come	VERB
easat-5356	115	6	at	at	ADP
easat-5356	115	7	the	the	DET
easat-5356	115	8	expense	expense	NOUN
easat-5356	115	9	of	of	ADP
easat-5356	115	10	decreased	decrease	VERB
easat-5356	115	11	localization	localization	NOUN
easat-5356	115	12	accuracy	accuracy	NOUN
easat-5356	115	13	and	and	CCONJ
easat-5356	115	14	possibly	possibly	ADV
easat-5356	115	15	increased	increase	VERB
easat-5356	115	16	false	false	ADJ
easat-5356	115	17	positive	positive	ADJ
easat-5356	115	18	rates	rate	NOUN
easat-5356	115	19	.	.	PUNCT
easat-5356	116	1	the	the	DET
easat-5356	116	2	dense	dense	ADJ
easat-5356	116	3	sampling	sampling	NOUN
easat-5356	116	4	of	of	ADP
easat-5356	116	5	anchor	anchor	NOUN
easat-5356	116	6	points	point	NOUN
easat-5356	116	7	in	in	ADP
easat-5356	116	8	single	single	ADJ
easat-5356	116	9	-	-	PUNCT
easat-5356	116	10	stage	stage	NOUN
easat-5356	116	11	methods	method	NOUN
easat-5356	116	12	can	can	AUX
easat-5356	116	13	result	result	VERB
easat-5356	116	14	in	in	ADP
easat-5356	116	15	more	more	ADJ
easat-5356	116	16	background	background	NOUN
easat-5356	116	17	regions	region	NOUN
easat-5356	116	18	being	be	AUX
easat-5356	116	19	misclassified	misclassifie	VERB
easat-5356	116	20	as	as	ADP
easat-5356	116	21	faces	face	NOUN
easat-5356	116	22	,	,	PUNCT
easat-5356	116	23	and	and	CCONJ
easat-5356	116	24	the	the	DET
easat-5356	116	25	lack	lack	NOUN
easat-5356	116	26	of	of	ADP
easat-5356	116	27	a	a	DET
easat-5356	116	28	refinement	refinement	NOUN
easat-5356	116	29	step	step	NOUN
easat-5356	116	30	,	,	PUNCT
easat-5356	116	31	common	common	ADJ
easat-5356	116	32	in	in	ADP
easat-5356	116	33	two	two	NUM
easat-5356	116	34	-	-	PUNCT
easat-5356	116	35	stage	stage	NOUN
easat-5356	116	36	methods	method	NOUN
easat-5356	116	37	,	,	PUNCT
easat-5356	116	38	can	can	AUX
easat-5356	116	39	lead	lead	VERB
easat-5356	116	40	to	to	ADP
easat-5356	116	41	less	less	ADV
easat-5356	116	42	precise	precise	ADJ
easat-5356	116	43	bounding	bounding	NOUN
easat-5356	116	44	box	box	NOUN
easat-5356	116	45	predictions	prediction	NOUN
easat-5356	116	46	.	.	PUNCT
easat-5356	117	1	balancing	balance	VERB
easat-5356	117	2	these	these	DET
easat-5356	117	3	trade	trade	NOUN
easat-5356	117	4	-	-	PUNCT
easat-5356	117	5	offs	off	NOUN
easat-5356	117	6	remains	remain	VERB
easat-5356	117	7	a	a	DET
easat-5356	117	8	key	key	ADJ
easat-5356	117	9	challenge	challenge	NOUN
easat-5356	117	10	in	in	ADP
easat-5356	117	11	optimizing	optimize	VERB
easat-5356	117	12	single	single	ADJ
easat-5356	117	13	-	-	PUNCT
easat-5356	117	14	stage	stage	NOUN
easat-5356	117	15	face	face	NOUN
easat-5356	117	16	detection	detection	NOUN
easat-5356	117	17	models	model	NOUN
easat-5356	117	18	.	.	PUNCT
easat-5356	118	1	the	the	DET
easat-5356	118	2	benefits	benefit	NOUN
easat-5356	118	3	of	of	ADP
easat-5356	118	4	general	general	ADJ
easat-5356	118	5	object	object	NOUN
easat-5356	118	6	detection	detection	NOUN
easat-5356	118	7	methods	method	NOUN
easat-5356	118	8	,	,	PUNCT
easat-5356	118	9	like	like	ADP
easat-5356	118	10	ssd	ssd	NOUN
easat-5356	118	11	,	,	PUNCT
easat-5356	118	12	have	have	AUX
easat-5356	118	13	had	have	VERB
easat-5356	118	14	a	a	DET
easat-5356	118	15	major	major	ADJ
easat-5356	118	16	impact	impact	NOUN
easat-5356	118	17	on	on	ADP
easat-5356	118	18	recent	recent	ADJ
easat-5356	118	19	developments	development	NOUN
easat-5356	118	20	in	in	ADP
easat-5356	118	21	facial	facial	ADJ
easat-5356	118	22	detection	detection	NOUN
easat-5356	118	23	algorithms	algorithm	NOUN
easat-5356	118	24	[	[	X
easat-5356	118	25	28	28	NUM
easat-5356	118	26	]	]	PUNCT
easat-5356	118	27	.	.	PUNCT
easat-5356	119	1	faster	fast	ADV
easat-5356	119	2	r	r	VERB
easat-5356	119	3	-	-	PUNCT
easat-5356	119	4	cnn	cnn	PROPN
easat-5356	119	5	han	han	PROPN
easat-5356	119	6	,	,	PUNCT
easat-5356	119	7	et	et	PROPN
easat-5356	119	8	al	al	PROPN
easat-5356	119	9	.	.	PUNCT
easat-5356	120	1	[	[	X
easat-5356	120	2	29	29	NUM
easat-5356	120	3	]	]	PUNCT
easat-5356	120	4	and	and	CCONJ
easat-5356	120	5	retinanet	retinanet	NOUN
easat-5356	120	6	[	[	X
easat-5356	120	7	30	30	NUM
easat-5356	120	8	]	]	PUNCT
easat-5356	120	9	.	.	PUNCT
easat-5356	121	1	faceboxes	faceboxes	PROPN
easat-5356	121	2	qi	qi	PROPN
easat-5356	121	3	,	,	PUNCT
easat-5356	121	4	et	et	PROPN
easat-5356	121	5	al	al	PROPN
easat-5356	121	6	.	.	PUNCT
easat-5356	122	1	[	[	X
easat-5356	122	2	10	10	NUM
easat-5356	122	3	]	]	X
easat-5356	122	4	a	a	DET
easat-5356	122	5	lightweight	lightweight	ADJ
easat-5356	122	6	network	network	NOUN
easat-5356	122	7	built	build	VERB
easat-5356	122	8	on	on	ADP
easat-5356	122	9	the	the	DET
easat-5356	122	10	ssd	ssd	NOUN
easat-5356	122	11	architecture	architecture	NOUN
easat-5356	122	12	,	,	PUNCT
easat-5356	122	13	was	be	AUX
easat-5356	122	14	created	create	VERB
easat-5356	122	15	by	by	ADP
easat-5356	122	16	zhang	zhang	PROPN
easat-5356	122	17	et	et	PROPN
easat-5356	122	18	al	al	PROPN
easat-5356	122	19	.	.	PUNCT
easat-5356	123	1	this	this	DET
easat-5356	123	2	algorithm	algorithm	NOUN
easat-5356	123	3	efficiently	efficiently	ADV
easat-5356	123	4	reduces	reduce	VERB
easat-5356	123	5	the	the	DET
easat-5356	123	6	feature	feature	NOUN
easat-5356	123	7	size	size	NOUN
easat-5356	123	8	by	by	ADP
easat-5356	123	9	a	a	DET
easat-5356	123	10	factor	factor	NOUN
easat-5356	123	11	of	of	ADP
easat-5356	123	12	32	32	NUM
easat-5356	123	13	through	through	ADP
easat-5356	123	14	down	down	ADP
easat-5356	123	15	sampling	sampling	NOUN
easat-5356	123	16	,	,	PUNCT
easat-5356	123	17	while	while	SCONJ
easat-5356	123	18	simultaneously	simultaneously	ADV
easat-5356	123	19	utilizing	utilize	VERB
easat-5356	123	20	multi	multi	ADJ
easat-5356	123	21	-	-	ADJ
easat-5356	123	22	scale	scale	ADJ
easat-5356	123	23	network	network	NOUN
easat-5356	123	24	modules	module	NOUN
easat-5356	123	25	to	to	PART
easat-5356	123	26	enrich	enrich	VERB
easat-5356	123	27	feature	feature	NOUN
easat-5356	123	28	dimensions	dimension	NOUN
easat-5356	123	29	in	in	ADP
easat-5356	123	30	both	both	CCONJ
easat-5356	123	31	width	width	ADJ
easat-5356	123	32	and	and	CCONJ
easat-5356	123	33	depth	depth	NOUN
easat-5356	123	34	.	.	PUNCT
easat-5356	124	1	this	this	DET
easat-5356	124	2	method	method	NOUN
easat-5356	124	3	enables	enable	VERB
easat-5356	124	4	the	the	DET
easat-5356	124	5	model	model	NOUN
easat-5356	124	6	to	to	PART
easat-5356	124	7	capture	capture	VERB
easat-5356	124	8	a	a	DET
easat-5356	124	9	broader	broad	ADJ
easat-5356	124	10	spectrum	spectrum	NOUN
easat-5356	124	11	of	of	ADP
easat-5356	124	12	spatial	spatial	ADJ
easat-5356	124	13	information	information	NOUN
easat-5356	124	14	and	and	CCONJ
easat-5356	124	15	scale	scale	NOUN
easat-5356	124	16	variations	variation	NOUN
easat-5356	124	17	,	,	PUNCT
easat-5356	124	18	improving	improve	VERB
easat-5356	124	19	its	its	PRON
easat-5356	124	20	ability	ability	NOUN
easat-5356	124	21	to	to	PART
easat-5356	124	22	detect	detect	VERB
easat-5356	124	23	faces	face	NOUN
easat-5356	124	24	at	at	ADP
easat-5356	124	25	different	different	ADJ
easat-5356	124	26	sizes	size	NOUN
easat-5356	124	27	and	and	CCONJ
easat-5356	124	28	under	under	ADP
easat-5356	124	29	varying	vary	VERB
easat-5356	124	30	conditions	condition	NOUN
easat-5356	124	31	.	.	PUNCT
easat-5356	125	1	by	by	ADP
easat-5356	125	2	balancing	balance	VERB
easat-5356	125	3	dimensionality	dimensionality	NOUN
easat-5356	125	4	reduction	reduction	NOUN
easat-5356	125	5	with	with	ADP
easat-5356	125	6	enhanced	enhanced	ADJ
easat-5356	125	7	feature	feature	NOUN
easat-5356	125	8	representation	representation	NOUN
easat-5356	125	9	,	,	PUNCT
easat-5356	125	10	the	the	DET
easat-5356	125	11	algorithm	algorithm	NOUN
easat-5356	125	12	achieves	achieve	VERB
easat-5356	125	13	both	both	DET
easat-5356	125	14	computational	computational	ADJ
easat-5356	125	15	efficiency	efficiency	NOUN
easat-5356	125	16	and	and	CCONJ
easat-5356	125	17	improved	improved	ADJ
easat-5356	125	18	detection	detection	NOUN
easat-5356	125	19	accuracy	accuracy	NOUN
easat-5356	125	20	.	.	PUNCT
easat-5356	126	1	the	the	DET
easat-5356	126	2	srn	srn	PROPN
easat-5356	126	3	yang	yang	PROPN
easat-5356	126	4	,	,	PUNCT
easat-5356	126	5	et	et	PROPN
easat-5356	126	6	al	al	PROPN
easat-5356	126	7	.	.	PUNCT
easat-5356	127	1	[	[	X
easat-5356	127	2	31	31	NUM
easat-5356	127	3	]	]	PUNCT
easat-5356	127	4	is	be	AUX
easat-5356	127	5	an	an	DET
easat-5356	127	6	improved	improved	ADJ
easat-5356	127	7	version	version	NOUN
easat-5356	127	8	of	of	ADP
easat-5356	127	9	the	the	DET
easat-5356	127	10	refinedet	refinedet	NOUN
easat-5356	127	11	and	and	CCONJ
easat-5356	127	12	retinanet	retinanet	NOUN
easat-5356	127	13	object	object	NOUN
easat-5356	127	14	identification	identification	NOUN
easat-5356	127	15	algorithms	algorithm	NOUN
easat-5356	127	16	that	that	PRON
easat-5356	127	17	achieves	achieve	VERB
easat-5356	127	18	great	great	ADJ
easat-5356	127	19	performance	performance	NOUN
easat-5356	127	20	by	by	ADP
easat-5356	127	21	adding	add	VERB
easat-5356	127	22	multi	multi	ADJ
easat-5356	127	23	-	-	ADJ
easat-5356	127	24	branch	branch	ADJ
easat-5356	127	25	modules	module	NOUN
easat-5356	127	26	to	to	PART
easat-5356	127	27	increase	increase	VERB
easat-5356	127	28	the	the	DET
easat-5356	127	29	impact	impact	NOUN
easat-5356	127	30	of	of	ADP
easat-5356	127	31	the	the	DET
easat-5356	127	32	receptive	receptive	ADJ
easat-5356	127	33	field	field	NOUN
easat-5356	127	34	and	and	CCONJ
easat-5356	127	35	introducing	introduce	VERB
easat-5356	127	36	two	two	NUM
easat-5356	127	37	-	-	PUNCT
easat-5356	127	38	stage	stage	NOUN
easat-5356	127	39	classification	classification	NOUN
easat-5356	127	40	and	and	CCONJ
easat-5356	127	41	regression	regression	NOUN
easat-5356	127	42	procedures	procedure	NOUN
easat-5356	127	43	.	.	PUNCT
easat-5356	128	1	furthermore	furthermore	ADV
easat-5356	128	2	,	,	PUNCT
easat-5356	128	3	by	by	ADP
easat-5356	128	4	combining	combine	VERB
easat-5356	128	5	facial	facial	ADJ
easat-5356	128	6	alignment	alignment	NOUN
easat-5356	128	7	and	and	CCONJ
easat-5356	128	8	facial	facial	ADJ
easat-5356	128	9	detection	detection	NOUN
easat-5356	128	10	,	,	PUNCT
easat-5356	128	11	mtcnet	mtcnet	NOUN
easat-5356	128	12	utilizes	utilize	VERB
easat-5356	128	13	facial	facial	ADJ
easat-5356	128	14	landmarks	landmark	NOUN
easat-5356	128	15	to	to	PART
easat-5356	128	16	improve	improve	VERB
easat-5356	128	17	detection	detection	NOUN
easat-5356	128	18	performance	performance	NOUN
easat-5356	128	19	.	.	PUNCT
easat-5356	129	1	facial	facial	ADJ
easat-5356	129	2	identification	identification	NOUN
easat-5356	129	3	algorithms	algorithm	NOUN
easat-5356	129	4	face	face	VERB
easat-5356	129	5	significant	significant	ADJ
easat-5356	129	6	hurdles	hurdle	NOUN
easat-5356	129	7	due	due	ADP
easat-5356	129	8	to	to	ADP
easat-5356	129	9	significant	significant	ADJ
easat-5356	129	10	fluctuations	fluctuation	NOUN
easat-5356	129	11	in	in	ADP
easat-5356	129	12	facial	facial	ADJ
easat-5356	129	13	scale	scale	NOUN
easat-5356	129	14	within	within	ADP
easat-5356	129	15	complex	complex	ADJ
easat-5356	129	16	scenarios	scenario	NOUN
easat-5356	129	17	,	,	PUNCT
easat-5356	129	18	which	which	PRON
easat-5356	129	19	negatively	negatively	ADV
easat-5356	129	20	impacts	impact	VERB
easat-5356	129	21	their	their	PRON
easat-5356	129	22	performance	performance	NOUN
easat-5356	129	23	.	.	PUNCT
easat-5356	130	1	multi	multi	ADJ
easat-5356	130	2	-	-	ADJ
easat-5356	130	3	scale	scale	ADJ
easat-5356	130	4	detection	detection	NOUN
easat-5356	130	5	capabilities	capability	NOUN
easat-5356	130	6	are	be	AUX
easat-5356	130	7	largely	largely	ADV
easat-5356	130	8	driven	drive	VERB
easat-5356	130	9	by	by	ADP
easat-5356	130	10	scale	scale	NOUN
easat-5356	130	11	-	-	PUNCT
easat-5356	130	12	invariant	invariant	ADJ
easat-5356	130	13	features	feature	NOUN
easat-5356	130	14	,	,	PUNCT
easat-5356	130	15	which	which	PRON
easat-5356	130	16	allow	allow	VERB
easat-5356	130	17	the	the	DET
easat-5356	130	18	model	model	NOUN
easat-5356	130	19	to	to	PART
easat-5356	130	20	detect	detect	VERB
easat-5356	130	21	objects	object	NOUN
easat-5356	130	22	,	,	PUNCT
easat-5356	130	23	such	such	ADJ
easat-5356	130	24	as	as	ADP
easat-5356	130	25	faces	face	NOUN
easat-5356	130	26	,	,	PUNCT
easat-5356	130	27	across	across	ADP
easat-5356	130	28	a	a	DET
easat-5356	130	29	wide	wide	ADJ
easat-5356	130	30	range	range	NOUN
easat-5356	130	31	of	of	ADP
easat-5356	130	32	sizes	size	NOUN
easat-5356	130	33	.	.	PUNCT
easat-5356	131	1	this	this	PRON
easat-5356	131	2	has	have	AUX
easat-5356	131	3	led	lead	VERB
easat-5356	131	4	to	to	ADP
easat-5356	131	5	extensive	extensive	ADJ
easat-5356	131	6	research	research	NOUN
easat-5356	131	7	focused	focus	VERB
easat-5356	131	8	on	on	ADP
easat-5356	131	9	improving	improve	VERB
easat-5356	131	10	the	the	DET
easat-5356	131	11	accuracy	accuracy	NOUN
easat-5356	131	12	and	and	CCONJ
easat-5356	131	13	efficiency	efficiency	NOUN
easat-5356	131	14	of	of	ADP
easat-5356	131	15	feature	feature	NOUN
easat-5356	131	16	extraction	extraction	NOUN
easat-5356	131	17	.	.	PUNCT
easat-5356	132	1	by	by	ADP
easat-5356	132	2	enhancing	enhance	VERB
easat-5356	132	3	feature	feature	NOUN
easat-5356	132	4	representations	representation	NOUN
easat-5356	132	5	that	that	PRON
easat-5356	132	6	are	be	AUX
easat-5356	132	7	robust	robust	ADJ
easat-5356	132	8	to	to	PART
easat-5356	132	9	scale	scale	VERB
easat-5356	132	10	variations	variation	NOUN
easat-5356	132	11	,	,	PUNCT
easat-5356	132	12	researchers	researcher	NOUN
easat-5356	132	13	aim	aim	VERB
easat-5356	132	14	to	to	PART
easat-5356	132	15	create	create	VERB
easat-5356	132	16	models	model	NOUN
easat-5356	132	17	that	that	PRON
easat-5356	132	18	can	can	AUX
easat-5356	132	19	effectively	effectively	ADV
easat-5356	132	20	detect	detect	VERB
easat-5356	132	21	faces	face	NOUN
easat-5356	132	22	at	at	ADP
easat-5356	132	23	both	both	CCONJ
easat-5356	132	24	small	small	ADJ
easat-5356	132	25	and	and	CCONJ
easat-5356	132	26	large	large	ADJ
easat-5356	132	27	scales	scale	NOUN
easat-5356	132	28	while	while	SCONJ
easat-5356	132	29	maintaining	maintain	VERB
easat-5356	132	30	high	high	ADJ
easat-5356	132	31	precision	precision	NOUN
easat-5356	132	32	and	and	CCONJ
easat-5356	132	33	computational	computational	ADJ
easat-5356	132	34	efficiency	efficiency	NOUN
easat-5356	132	35	.	.	PUNCT
easat-5356	133	1	these	these	DET
easat-5356	133	2	advancements	advancement	NOUN
easat-5356	133	3	often	often	ADV
easat-5356	133	4	involve	involve	VERB
easat-5356	133	5	the	the	DET
easat-5356	133	6	use	use	NOUN
easat-5356	133	7	of	of	ADP
easat-5356	133	8	multi	multi	ADJ
easat-5356	133	9	-	-	ADJ
easat-5356	133	10	scale	scale	ADJ
easat-5356	133	11	feature	feature	NOUN
easat-5356	133	12	pyramids	pyramid	NOUN
easat-5356	133	13	,	,	PUNCT
easat-5356	133	14	adaptive	adaptive	ADJ
easat-5356	133	15	sampling	sampling	NOUN
easat-5356	133	16	strategies	strategy	NOUN
easat-5356	133	17	,	,	PUNCT
easat-5356	133	18	and	and	CCONJ
easat-5356	133	19	more	more	ADV
easat-5356	133	20	sophisticated	sophisticated	ADJ
easat-5356	133	21	network	network	NOUN
easat-5356	133	22	architectures	architecture	NOUN
easat-5356	133	23	that	that	PRON
easat-5356	133	24	better	well	ADJ
easat-5356	133	25	capture	capture	VERB
easat-5356	133	26	scale	scale	NOUN
easat-5356	133	27	-	-	PUNCT
easat-5356	133	28	invariant	invariant	ADJ
easat-5356	133	29	patterns	pattern	NOUN
easat-5356	133	30	.	.	PUNCT
easat-5356	134	1	using	use	VERB
easat-5356	134	2	dilated	dilated	ADJ
easat-5356	134	3	convolutions	convolution	NOUN
easat-5356	134	4	and	and	CCONJ
easat-5356	134	5	lowering	lower	VERB
easat-5356	134	6	the	the	DET
easat-5356	134	7	number	number	NOUN
easat-5356	134	8	of	of	ADP
easat-5356	134	9	down	down	ADV
easat-5356	134	10	sampling	sample	VERB
easat-5356	134	11	layers	layer	NOUN
easat-5356	134	12	can	can	AUX
easat-5356	134	13	greatly	greatly	ADV
easat-5356	134	14	improve	improve	VERB
easat-5356	134	15	detection	detection	NOUN
easat-5356	134	16	performance	performance	NOUN
easat-5356	134	17	for	for	ADP
easat-5356	134	18	small	small	ADJ
easat-5356	134	19	item	item	NOUN
easat-5356	134	20	detection	detection	NOUN
easat-5356	134	21	.	.	PUNCT
easat-5356	135	1	using	use	VERB
easat-5356	135	2	more	more	ADJ
easat-5356	135	3	anchors	anchor	NOUN
easat-5356	135	4	,	,	PUNCT
easat-5356	135	5	which	which	PRON
easat-5356	135	6	offer	offer	VERB
easat-5356	135	7	useful	useful	ADJ
easat-5356	135	8	prior	prior	ADJ
easat-5356	135	9	knowledge	knowledge	NOUN
easat-5356	135	10	,	,	PUNCT
easat-5356	135	11	is	be	AUX
easat-5356	135	12	another	another	DET
easat-5356	135	13	way	way	NOUN
easat-5356	135	14	to	to	PART
easat-5356	135	15	address	address	VERB
easat-5356	135	16	this	this	DET
easat-5356	135	17	issue	issue	NOUN
easat-5356	135	18	;	;	PUNCT
easat-5356	135	19	using	use	VERB
easat-5356	135	20	denser	dense	ADJ
easat-5356	135	21	anchors	anchor	NOUN
easat-5356	135	22	in	in	ADP
easat-5356	135	23	conjunction	conjunction	NOUN
easat-5356	135	24	with	with	ADP
easat-5356	135	25	suitable	suitable	ADJ
easat-5356	135	26	matching	matching	NOUN
easat-5356	135	27	techniques	technique	NOUN
easat-5356	135	28	can	can	AUX
easat-5356	135	29	significantly	significantly	ADV
easat-5356	135	30	raise	raise	VERB
easat-5356	135	31	the	the	DET
easat-5356	135	32	calibre	calibre	NOUN
easat-5356	135	33	of	of	ADP
easat-5356	135	34	object	object	NOUN
easat-5356	135	35	proposals	proposal	NOUN
easat-5356	135	36	.	.	PUNCT
easat-5356	136	1	multi	multi	ADJ
easat-5356	136	2	-	-	ADJ
easat-5356	136	3	scale	scale	ADJ
easat-5356	136	4	training	training	NOUN
easat-5356	136	5	provides	provide	VERB
easat-5356	136	6	a	a	DET
easat-5356	136	7	simple	simple	ADJ
easat-5356	136	8	yet	yet	CCONJ
easat-5356	136	9	efficient	efficient	ADJ
easat-5356	136	10	method	method	NOUN
easat-5356	136	11	to	to	PART
easat-5356	136	12	improve	improve	VERB
easat-5356	136	13	multi	multi	ADJ
easat-5356	136	14	-	-	ADJ
easat-5356	136	15	scale	scale	ADJ
easat-5356	136	16	object	object	NOUN
easat-5356	136	17	detection	detection	NOUN
easat-5356	136	18	performance	performance	NOUN
easat-5356	136	19	,	,	PUNCT
easat-5356	136	20	and	and	CCONJ
easat-5356	136	21	it	it	PRON
easat-5356	136	22	is	be	AUX
easat-5356	136	23	essential	essential	ADJ
easat-5356	136	24	for	for	ADP
easat-5356	136	25	building	build	VERB
easat-5356	136	26	image	image	NOUN
easat-5356	136	27	pyramids	pyramid	NOUN
easat-5356	136	28	and	and	CCONJ
easat-5356	136	29	expanding	expand	VERB
easat-5356	136	30	sample	sample	NOUN
easat-5356	136	31	diversity	diversity	NOUN
easat-5356	136	32	[	[	X
easat-5356	136	33	32	32	NUM
easat-5356	136	34	]	]	PUNCT
easat-5356	136	35	.	.	PUNCT
easat-5356	137	1	by	by	ADP
easat-5356	137	2	training	train	VERB
easat-5356	137	3	on	on	ADP
easat-5356	137	4	images	image	NOUN
easat-5356	137	5	at	at	ADP
easat-5356	137	6	various	various	ADJ
easat-5356	137	7	scales	scale	NOUN
easat-5356	137	8	,	,	PUNCT
easat-5356	137	9	this	this	DET
easat-5356	137	10	technique	technique	NOUN
easat-5356	137	11	allows	allow	VERB
easat-5356	137	12	the	the	DET
easat-5356	137	13	model	model	NOUN
easat-5356	137	14	to	to	PART
easat-5356	137	15	learn	learn	VERB
easat-5356	137	16	features	feature	NOUN
easat-5356	137	17	that	that	PRON
easat-5356	137	18	are	be	AUX
easat-5356	137	19	robust	robust	ADJ
easat-5356	137	20	to	to	PART
easat-5356	137	21	scale	scale	VERB
easat-5356	137	22	variations	variation	NOUN
easat-5356	137	23	,	,	PUNCT
easat-5356	137	24	improving	improve	VERB
easat-5356	137	25	detection	detection	NOUN
easat-5356	137	26	across	across	ADP
easat-5356	137	27	different	different	ADJ
easat-5356	137	28	object	object	NOUN
easat-5356	137	29	sizes	size	NOUN
easat-5356	137	30	.	.	PUNCT
easat-5356	138	1	however	however	ADV
easat-5356	138	2	,	,	PUNCT
easat-5356	138	3	as	as	SCONJ
easat-5356	138	4	the	the	DET
easat-5356	138	5	receptive	receptive	ADJ
easat-5356	138	6	field	field	NOUN
easat-5356	138	7	expands	expand	VERB
easat-5356	138	8	and	and	CCONJ
easat-5356	138	9	deeper	deep	ADJ
easat-5356	138	10	semantic	semantic	ADJ
easat-5356	138	11	information	information	NOUN
easat-5356	138	12	is	be	AUX
easat-5356	138	13	integrated	integrate	VERB
easat-5356	138	14	,	,	PUNCT
easat-5356	138	15	there	there	PRON
easat-5356	138	16	is	be	VERB
easat-5356	138	17	a	a	DET
easat-5356	138	18	risk	risk	NOUN
easat-5356	138	19	of	of	ADP
easat-5356	138	20	losing	lose	VERB
easat-5356	138	21	finegrained	finegraine	VERB
easat-5356	138	22	spatial	spatial	ADJ
easat-5356	138	23	details	detail	NOUN
easat-5356	138	24	.	.	PUNCT
easat-5356	139	1	a	a	DET
easat-5356	139	2	natural	natural	ADJ
easat-5356	139	3	solution	solution	NOUN
easat-5356	139	4	to	to	ADP
easat-5356	139	5	this	this	DET
easat-5356	139	6	challenge	challenge	NOUN
easat-5356	139	7	is	be	AUX
easat-5356	139	8	to	to	PART
easat-5356	139	9	combine	combine	VERB
easat-5356	139	10	deep	deep	ADJ
easat-5356	139	11	semantic	semantic	ADJ
easat-5356	139	12	features	feature	NOUN
easat-5356	139	13	with	with	ADP
easat-5356	139	14	shallow	shallow	ADJ
easat-5356	139	15	features	feature	NOUN
easat-5356	139	16	kavitha	kavitha	PROPN
easat-5356	139	17	and	and	CCONJ
easat-5356	139	18	rajivkannan	rajivkannan	PROPN
easat-5356	140	1	[	[	X
easat-5356	140	2	33	33	NUM
easat-5356	140	3	]	]	PUNCT
easat-5356	140	4	thereby	thereby	ADV
easat-5356	140	5	retaining	retain	VERB
easat-5356	140	6	spatial	spatial	ADJ
easat-5356	140	7	information	information	NOUN
easat-5356	140	8	while	while	SCONJ
easat-5356	140	9	benefiting	benefit	VERB
easat-5356	140	10	from	from	ADP
easat-5356	140	11	richer	rich	ADJ
easat-5356	140	12	context	context	NOUN
easat-5356	140	13	.	.	PUNCT
easat-5356	141	1	this	this	DET
easat-5356	141	2	fusion	fusion	NOUN
easat-5356	141	3	enables	enable	VERB
easat-5356	141	4	the	the	DET
easat-5356	141	5	model	model	NOUN
easat-5356	141	6	to	to	PART
easat-5356	141	7	maintain	maintain	VERB
easat-5356	141	8	both	both	PRON
easat-5356	141	9	high	high	ADJ
easat-5356	141	10	-	-	PUNCT
easat-5356	141	11	level	level	NOUN
easat-5356	141	12	semantic	semantic	ADJ
easat-5356	141	13	understanding	understanding	NOUN
easat-5356	141	14	and	and	CCONJ
easat-5356	141	15	precise	precise	ADJ
easat-5356	141	16	spatial	spatial	ADJ
easat-5356	141	17	localization	localization	NOUN
easat-5356	141	18	,	,	PUNCT
easat-5356	141	19	improving	improve	VERB
easat-5356	141	20	detection	detection	NOUN
easat-5356	141	21	accuracy	accuracy	NOUN
easat-5356	141	22	in	in	ADP
easat-5356	141	23	complex	complex	ADJ
easat-5356	141	24	scenarios	scenario	NOUN
easat-5356	141	25	.	.	PUNCT
easat-5356	142	1	in	in	ADP
easat-5356	142	2	densely	densely	ADV
easat-5356	142	3	populated	populated	ADJ
easat-5356	142	4	scenes	scene	NOUN
easat-5356	142	5	,	,	PUNCT
easat-5356	142	6	occlusion	occlusion	NOUN
easat-5356	142	7	poses	pose	VERB
easat-5356	142	8	a	a	DET
easat-5356	142	9	significant	significant	ADJ
easat-5356	142	10	challenge	challenge	NOUN
easat-5356	142	11	by	by	ADP
easat-5356	142	12	causing	cause	VERB
easat-5356	142	13	incomplete	incomplete	ADJ
easat-5356	142	14	data	datum	NOUN
easat-5356	142	15	regarding	regard	VERB
easat-5356	142	16	occluded	occlude	VERB
easat-5356	142	17	faces	face	NOUN
easat-5356	142	18	.	.	PUNCT
easat-5356	143	1	as	as	SCONJ
easat-5356	143	2	certain	certain	ADJ
easat-5356	143	3	facial	facial	ADJ
easat-5356	143	4	regions	region	NOUN
easat-5356	143	5	become	become	VERB
easat-5356	143	6	hidden	hidden	ADJ
easat-5356	143	7	or	or	CCONJ
easat-5356	143	8	boundaries	boundary	NOUN
easat-5356	143	9	become	become	VERB
easat-5356	143	10	ambiguous	ambiguous	ADJ
easat-5356	143	11	,	,	PUNCT
easat-5356	143	12	the	the	DET
easat-5356	143	13	model	model	NOUN
easat-5356	143	14	may	may	AUX
easat-5356	143	15	fail	fail	VERB
easat-5356	143	16	to	to	PART
easat-5356	143	17	detect	detect	VERB
easat-5356	143	18	these	these	DET
easat-5356	143	19	faces	face	NOUN
easat-5356	143	20	,	,	PUNCT
easat-5356	143	21	leading	lead	VERB
easat-5356	143	22	to	to	ADP
easat-5356	143	23	missed	miss	VERB
easat-5356	143	24	detections	detection	NOUN
easat-5356	143	25	and	and	CCONJ
easat-5356	143	26	low	low	ADJ
easat-5356	143	27	recall	recall	NOUN
easat-5356	143	28	rates	rate	NOUN
easat-5356	143	29	.	.	PUNCT
easat-5356	144	1	occlusion	occlusion	NOUN
easat-5356	144	2	can	can	AUX
easat-5356	144	3	obscure	obscure	VERB
easat-5356	144	4	key	key	ADJ
easat-5356	144	5	facial	facial	ADJ
easat-5356	144	6	features	feature	NOUN
easat-5356	144	7	or	or	CCONJ
easat-5356	144	8	cause	cause	VERB
easat-5356	144	9	overlapping	overlap	VERB
easat-5356	144	10	objects	object	NOUN
easat-5356	144	11	to	to	PART
easat-5356	144	12	be	be	AUX
easat-5356	144	13	misclassified	misclassifie	VERB
easat-5356	144	14	,	,	PUNCT
easat-5356	144	15	reducing	reduce	VERB
easat-5356	144	16	the	the	DET
easat-5356	144	17	effectiveness	effectiveness	NOUN
easat-5356	144	18	of	of	ADP
easat-5356	144	19	traditional	traditional	ADJ
easat-5356	144	20	detection	detection	NOUN
easat-5356	144	21	methods	method	NOUN
easat-5356	144	22	.	.	PUNCT
easat-5356	145	1	to	to	PART
easat-5356	145	2	address	address	VERB
easat-5356	145	3	this	this	PRON
easat-5356	145	4	,	,	PUNCT
easat-5356	145	5	more	more	ADV
easat-5356	145	6	advanced	advanced	ADJ
easat-5356	145	7	strategies	strategy	NOUN
easat-5356	145	8	,	,	PUNCT
easat-5356	145	9	such	such	ADJ
easat-5356	145	10	as	as	ADP
easat-5356	145	11	leveraging	leverage	VERB
easat-5356	145	12	contextual	contextual	ADJ
easat-5356	145	13	information	information	NOUN
easat-5356	145	14	,	,	PUNCT
easat-5356	145	15	multi	multi	ADJ
easat-5356	145	16	-	-	ADJ
easat-5356	145	17	view	view	ADJ
easat-5356	145	18	detection	detection	NOUN
easat-5356	145	19	,	,	PUNCT
easat-5356	145	20	or	or	CCONJ
easat-5356	145	21	incorporating	incorporate	VERB
easat-5356	145	22	occlusion	occlusion	NOUN
easat-5356	145	23	-	-	PUNCT
easat-5356	145	24	aware	aware	ADJ
easat-5356	145	25	models	model	NOUN
easat-5356	145	26	,	,	PUNCT
easat-5356	145	27	are	be	AUX
easat-5356	145	28	needed	need	VERB
easat-5356	145	29	to	to	PART
easat-5356	145	30	improve	improve	VERB
easat-5356	145	31	844	844	NUM
easat-5356	145	32	edelweiss	edelweiss	PROPN
easat-5356	145	33	applied	apply	VERB
easat-5356	145	34	science	science	NOUN
easat-5356	145	35	and	and	CCONJ
easat-5356	145	36	technology	technology	NOUN
easat-5356	145	37	issn	issn	PROPN
easat-5356	145	38	:	:	PUNCT
easat-5356	145	39	2576	2576	NUM
easat-5356	145	40	-	-	SYM
easat-5356	145	41	8484	8484	NUM
easat-5356	145	42	vol	vol	NOUN
easat-5356	145	43	.	.	PROPN
easat-5356	146	1	9	9	NUM
easat-5356	146	2	,	,	PUNCT
easat-5356	146	3	no	no	INTJ
easat-5356	146	4	.	.	NOUN
easat-5356	147	1	3	3	NUM
easat-5356	147	2	:	:	PUNCT
easat-5356	147	3	840	840	NUM
easat-5356	147	4	-	-	SYM
easat-5356	147	5	855	855	NUM
easat-5356	147	6	,	,	PUNCT
easat-5356	147	7	2025	2025	NUM
easat-5356	147	8	doi	doi	NOUN
easat-5356	147	9	:	:	PUNCT
easat-5356	147	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	147	11	©	©	PROPN
easat-5356	147	12	2025	2025	NUM
easat-5356	147	13	by	by	ADP
easat-5356	147	14	the	the	DET
easat-5356	147	15	authors	author	NOUN
easat-5356	147	16	;	;	PUNCT
easat-5356	147	17	licensee	licensee	PROPN
easat-5356	147	18	learning	learn	VERB
easat-5356	147	19	gate	gate	PROPN
easat-5356	147	20	detection	detection	NOUN
easat-5356	147	21	accuracy	accuracy	NOUN
easat-5356	147	22	in	in	ADP
easat-5356	147	23	such	such	ADJ
easat-5356	147	24	complex	complex	ADJ
easat-5356	147	25	scenarios	scenario	NOUN
easat-5356	147	26	.	.	PUNCT
easat-5356	148	1	contextual	contextual	ADJ
easat-5356	148	2	information	information	NOUN
easat-5356	148	3	is	be	AUX
easat-5356	148	4	crucial	crucial	ADJ
easat-5356	148	5	for	for	ADP
easat-5356	148	6	mitigating	mitigate	VERB
easat-5356	148	7	occlusion	occlusion	NOUN
easat-5356	148	8	problems	problem	NOUN
easat-5356	148	9	in	in	ADP
easat-5356	148	10	facial	facial	ADJ
easat-5356	148	11	detection	detection	NOUN
easat-5356	148	12	,	,	PUNCT
easat-5356	148	13	according	accord	VERB
easat-5356	148	14	to	to	ADP
easat-5356	148	15	earlier	early	ADJ
easat-5356	148	16	research	research	NOUN
easat-5356	148	17	.	.	PUNCT
easat-5356	149	1	ssh	ssh	PROPN
easat-5356	149	2	zhang	zhang	PROPN
easat-5356	149	3	,	,	PUNCT
easat-5356	149	4	et	et	PROPN
easat-5356	149	5	al	al	PROPN
easat-5356	149	6	.	.	PUNCT
easat-5356	150	1	[	[	X
easat-5356	150	2	11	11	NUM
easat-5356	150	3	]	]	PUNCT
easat-5356	150	4	utilizes	utilize	VERB
easat-5356	150	5	simple	simple	ADJ
easat-5356	150	6	convolutional	convolutional	ADJ
easat-5356	150	7	layers	layer	NOUN
easat-5356	150	8	to	to	PART
easat-5356	150	9	extend	extend	VERB
easat-5356	150	10	the	the	DET
easat-5356	150	11	windows	window	NOUN
easat-5356	150	12	around	around	ADP
easat-5356	150	13	candidate	candidate	NOUN
easat-5356	150	14	regions	region	NOUN
easat-5356	150	15	,	,	PUNCT
easat-5356	150	16	effectively	effectively	ADV
easat-5356	150	17	incorporating	incorporate	VERB
easat-5356	150	18	contextual	contextual	ADJ
easat-5356	150	19	information	information	NOUN
easat-5356	150	20	to	to	PART
easat-5356	150	21	help	help	VERB
easat-5356	150	22	detect	detect	VERB
easat-5356	150	23	faces	face	NOUN
easat-5356	150	24	that	that	PRON
easat-5356	150	25	may	may	AUX
easat-5356	150	26	be	be	AUX
easat-5356	150	27	partially	partially	ADV
easat-5356	150	28	occluded	occlude	VERB
easat-5356	150	29	.	.	PUNCT
easat-5356	151	1	fan	fan	PROPN
easat-5356	151	2	wang	wang	PROPN
easat-5356	151	3	,	,	PUNCT
easat-5356	151	4	et	et	PROPN
easat-5356	151	5	al	al	PROPN
easat-5356	151	6	.	.	PUNCT
easat-5356	152	1	[	[	X
easat-5356	152	2	34	34	NUM
easat-5356	152	3	]	]	PUNCT
easat-5356	152	4	introduces	introduce	NOUN
easat-5356	152	5	anchor	anchor	NOUN
easat-5356	152	6	attention	attention	NOUN
easat-5356	152	7	,	,	PUNCT
easat-5356	152	8	which	which	PRON
easat-5356	152	9	prioritizes	prioritize	VERB
easat-5356	152	10	facial	facial	ADJ
easat-5356	152	11	features	feature	NOUN
easat-5356	152	12	,	,	PUNCT
easat-5356	152	13	allowing	allow	VERB
easat-5356	152	14	the	the	DET
easat-5356	152	15	model	model	NOUN
easat-5356	152	16	to	to	PART
easat-5356	152	17	focus	focus	VERB
easat-5356	152	18	on	on	ADP
easat-5356	152	19	key	key	ADJ
easat-5356	152	20	areas	area	NOUN
easat-5356	152	21	even	even	ADV
easat-5356	152	22	when	when	SCONJ
easat-5356	152	23	faces	face	NOUN
easat-5356	152	24	are	be	AUX
easat-5356	152	25	partially	partially	ADV
easat-5356	152	26	hidden	hide	VERB
easat-5356	152	27	.	.	PUNCT
easat-5356	153	1	pyramidbox	pyramidbox	PROPN
easat-5356	153	2	cao	cao	PROPN
easat-5356	153	3	,	,	PUNCT
easat-5356	153	4	et	et	PROPN
easat-5356	153	5	al	al	PROPN
easat-5356	153	6	.	.	PUNCT
easat-5356	154	1	[	[	X
easat-5356	154	2	35	35	NUM
easat-5356	154	3	]	]	PUNCT
easat-5356	154	4	enhances	enhance	VERB
easat-5356	154	5	this	this	DET
easat-5356	154	6	approach	approach	NOUN
easat-5356	154	7	by	by	ADP
easat-5356	154	8	implementing	implement	VERB
easat-5356	154	9	a	a	DET
easat-5356	154	10	context	context	NOUN
easat-5356	154	11	-	-	PUNCT
easat-5356	154	12	sensitive	sensitive	ADJ
easat-5356	154	13	prediction	prediction	NOUN
easat-5356	154	14	module	module	NOUN
easat-5356	154	15	,	,	PUNCT
easat-5356	154	16	replacing	replace	VERB
easat-5356	154	17	ssh	ssh	NOUN
easat-5356	154	18	's	's	PART
easat-5356	154	19	convolutional	convolutional	ADJ
easat-5356	154	20	layers	layer	NOUN
easat-5356	154	21	with	with	ADP
easat-5356	154	22	dssd	dssd	NOUN
easat-5356	154	23	's	's	PART
easat-5356	154	24	residual	residual	ADJ
easat-5356	154	25	prediction	prediction	NOUN
easat-5356	154	26	module	module	NOUN
easat-5356	154	27	,	,	PUNCT
easat-5356	154	28	further	far	ADV
easat-5356	154	29	improving	improve	VERB
easat-5356	154	30	its	its	PRON
easat-5356	154	31	ability	ability	NOUN
easat-5356	154	32	to	to	PART
easat-5356	154	33	handle	handle	VERB
easat-5356	154	34	occlusions	occlusion	NOUN
easat-5356	154	35	.	.	PUNCT
easat-5356	155	1	additionally	additionally	ADV
easat-5356	155	2	,	,	PUNCT
easat-5356	155	3	retinaface	retinaface	PROPN
easat-5356	155	4	hao	hao	PROPN
easat-5356	155	5	,	,	PUNCT
easat-5356	155	6	et	et	PROPN
easat-5356	155	7	al	al	PROPN
easat-5356	155	8	.	.	PUNCT
easat-5356	156	1	[	[	X
easat-5356	156	2	36	36	NUM
easat-5356	156	3	]	]	PUNCT
easat-5356	156	4	applies	apply	VERB
easat-5356	156	5	independent	independent	ADJ
easat-5356	156	6	contextual	contextual	ADJ
easat-5356	156	7	modules	module	NOUN
easat-5356	156	8	across	across	ADP
easat-5356	156	9	five	five	NUM
easat-5356	156	10	feature	feature	NOUN
easat-5356	156	11	pyramid	pyramid	NOUN
easat-5356	156	12	layers	layer	NOUN
easat-5356	156	13	,	,	PUNCT
easat-5356	156	14	expanding	expand	VERB
easat-5356	156	15	the	the	DET
easat-5356	156	16	receptive	receptive	ADJ
easat-5356	156	17	field	field	NOUN
easat-5356	156	18	and	and	CCONJ
easat-5356	156	19	strengthening	strengthen	VERB
easat-5356	156	20	the	the	DET
easat-5356	156	21	model	model	NOUN
easat-5356	156	22	’s	’s	PART
easat-5356	156	23	capacity	capacity	NOUN
easat-5356	156	24	to	to	PART
easat-5356	156	25	model	model	VERB
easat-5356	156	26	rigid	rigid	ADJ
easat-5356	156	27	context	context	NOUN
easat-5356	156	28	.	.	PUNCT
easat-5356	157	1	these	these	DET
easat-5356	157	2	methods	method	NOUN
easat-5356	157	3	have	have	AUX
easat-5356	157	4	shown	show	VERB
easat-5356	157	5	strong	strong	ADJ
easat-5356	157	6	performance	performance	NOUN
easat-5356	157	7	in	in	ADP
easat-5356	157	8	mitigating	mitigate	VERB
easat-5356	157	9	occlusion	occlusion	NOUN
easat-5356	157	10	issues	issue	NOUN
easat-5356	157	11	,	,	PUNCT
easat-5356	157	12	demonstrating	demonstrate	VERB
easat-5356	157	13	that	that	SCONJ
easat-5356	157	14	leveraging	leverage	VERB
easat-5356	157	15	contextual	contextual	ADJ
easat-5356	157	16	information	information	NOUN
easat-5356	157	17	is	be	AUX
easat-5356	157	18	an	an	DET
easat-5356	157	19	effective	effective	ADJ
easat-5356	157	20	strategy	strategy	NOUN
easat-5356	157	21	for	for	ADP
easat-5356	157	22	improving	improve	VERB
easat-5356	157	23	detection	detection	NOUN
easat-5356	157	24	in	in	ADP
easat-5356	157	25	occluded	occluded	ADJ
easat-5356	157	26	areas	area	NOUN
easat-5356	157	27	.	.	PUNCT
easat-5356	158	1	this	this	DET
easat-5356	158	2	line	line	NOUN
easat-5356	158	3	of	of	ADP
easat-5356	158	4	research	research	NOUN
easat-5356	158	5	offers	offer	VERB
easat-5356	158	6	promising	promise	VERB
easat-5356	158	7	potential	potential	NOUN
easat-5356	158	8	and	and	CCONJ
easat-5356	158	9	warrants	warrant	NOUN
easat-5356	158	10	further	further	ADJ
easat-5356	158	11	exploration	exploration	NOUN
easat-5356	158	12	to	to	PART
easat-5356	158	13	enhance	enhance	VERB
easat-5356	158	14	robustness	robustness	NOUN
easat-5356	158	15	in	in	ADP
easat-5356	158	16	challenging	challenge	VERB
easat-5356	158	17	detection	detection	NOUN
easat-5356	158	18	environments	environment	NOUN
easat-5356	158	19	.	.	PUNCT
easat-5356	159	1	the	the	DET
easat-5356	159	2	total	total	ADJ
easat-5356	159	3	loss	loss	NOUN
easat-5356	159	4	function	function	NOUN
easat-5356	159	5	is	be	AUX
easat-5356	159	6	dominated	dominate	VERB
easat-5356	159	7	by	by	ADP
easat-5356	159	8	easy	easy	ADJ
easat-5356	159	9	samples	sample	NOUN
easat-5356	159	10	,	,	PUNCT
easat-5356	159	11	which	which	PRON
easat-5356	159	12	make	make	VERB
easat-5356	159	13	up	up	ADP
easat-5356	159	14	a	a	DET
easat-5356	159	15	sizable	sizable	ADJ
easat-5356	159	16	part	part	NOUN
easat-5356	159	17	of	of	ADP
easat-5356	159	18	samples	sample	NOUN
easat-5356	159	19	in	in	ADP
easat-5356	159	20	the	the	DET
easat-5356	159	21	single	single	ADJ
easat-5356	159	22	-	-	PUNCT
easat-5356	159	23	stage	stage	NOUN
easat-5356	159	24	facial	facial	ADJ
easat-5356	159	25	detection	detection	NOUN
easat-5356	159	26	domain	domain	NOUN
easat-5356	159	27	.	.	PUNCT
easat-5356	160	1	consequently	consequently	ADV
easat-5356	160	2	,	,	PUNCT
easat-5356	160	3	learning	learn	VERB
easat-5356	160	4	features	feature	NOUN
easat-5356	160	5	from	from	ADP
easat-5356	160	6	simple	simple	ADJ
easat-5356	160	7	samples	sample	NOUN
easat-5356	160	8	is	be	AUX
easat-5356	160	9	frequently	frequently	ADV
easat-5356	160	10	given	give	VERB
easat-5356	160	11	priority	priority	NOUN
easat-5356	160	12	in	in	ADP
easat-5356	160	13	models	model	NOUN
easat-5356	160	14	,	,	PUNCT
easat-5356	160	15	whereas	whereas	SCONJ
easat-5356	160	16	the	the	DET
easat-5356	160	17	more	more	ADV
easat-5356	160	18	difficult	difficult	ADJ
easat-5356	160	19	and	and	CCONJ
easat-5356	160	20	complex	complex	ADJ
easat-5356	160	21	examples	example	NOUN
easat-5356	160	22	are	be	AUX
easat-5356	160	23	underrepresented	underrepresente	VERB
easat-5356	160	24	throughout	throughout	ADP
easat-5356	160	25	the	the	DET
easat-5356	160	26	training	training	NOUN
easat-5356	160	27	phase	phase	NOUN
easat-5356	160	28	.	.	PUNCT
easat-5356	161	1	poor	poor	ADJ
easat-5356	161	2	generalization	generalization	NOUN
easat-5356	161	3	may	may	AUX
easat-5356	161	4	result	result	VERB
easat-5356	161	5	from	from	ADP
easat-5356	161	6	this	this	PRON
easat-5356	161	7	,	,	PUNCT
easat-5356	161	8	especially	especially	ADV
easat-5356	161	9	when	when	SCONJ
easat-5356	161	10	there	there	PRON
easat-5356	161	11	is	be	VERB
easat-5356	161	12	occlusion	occlusion	NOUN
easat-5356	161	13	,	,	PUNCT
easat-5356	161	14	small	small	ADJ
easat-5356	161	15	faces	face	NOUN
easat-5356	161	16	,	,	PUNCT
easat-5356	161	17	or	or	CCONJ
easat-5356	161	18	odd	odd	ADJ
easat-5356	161	19	postures	posture	NOUN
easat-5356	161	20	.	.	PUNCT
easat-5356	162	1	to	to	PART
easat-5356	162	2	tackle	tackle	VERB
easat-5356	162	3	this	this	DET
easat-5356	162	4	problem	problem	NOUN
easat-5356	162	5	,	,	PUNCT
easat-5356	162	6	the	the	DET
easat-5356	162	7	online	online	ADJ
easat-5356	162	8	hard	hard	ADJ
easat-5356	162	9	example	example	NOUN
easat-5356	162	10	mining	mining	NOUN
easat-5356	162	11	(	(	PUNCT
easat-5356	162	12	ohem	ohem	NOUN
easat-5356	162	13	)	)	PUNCT
easat-5356	162	14	approach	approach	NOUN
easat-5356	162	15	integrates	integrate	VERB
easat-5356	162	16	hard	hard	ADJ
easat-5356	162	17	examples	example	NOUN
easat-5356	162	18	into	into	ADP
easat-5356	162	19	the	the	DET
easat-5356	162	20	stochastic	stochastic	ADJ
easat-5356	162	21	gradient	gradient	ADJ
easat-5356	162	22	descent	descent	NOUN
easat-5356	162	23	(	(	PUNCT
easat-5356	162	24	sgd	sgd	NOUN
easat-5356	162	25	)	)	PUNCT
easat-5356	162	26	training	training	NOUN
easat-5356	162	27	process	process	NOUN
easat-5356	162	28	by	by	ADP
easat-5356	162	29	choosing	choose	VERB
easat-5356	162	30	them	they	PRON
easat-5356	162	31	according	accord	VERB
easat-5356	162	32	to	to	ADP
easat-5356	162	33	their	their	PRON
easat-5356	162	34	loss	loss	NOUN
easat-5356	162	35	values	value	NOUN
easat-5356	162	36	.	.	PUNCT
easat-5356	163	1	by	by	ADP
easat-5356	163	2	focusing	focus	VERB
easat-5356	163	3	the	the	DET
easat-5356	163	4	model	model	NOUN
easat-5356	163	5	's	's	PART
easat-5356	163	6	attention	attention	NOUN
easat-5356	163	7	on	on	ADP
easat-5356	163	8	these	these	DET
easat-5356	163	9	difficult	difficult	ADJ
easat-5356	163	10	examples	example	NOUN
easat-5356	163	11	,	,	PUNCT
easat-5356	163	12	ohem	ohem	NOUN
easat-5356	163	13	helps	help	VERB
easat-5356	163	14	ensure	ensure	VERB
easat-5356	163	15	that	that	SCONJ
easat-5356	163	16	the	the	DET
easat-5356	163	17	model	model	NOUN
easat-5356	163	18	learns	learn	VERB
easat-5356	163	19	to	to	PART
easat-5356	163	20	detect	detect	VERB
easat-5356	163	21	faces	face	NOUN
easat-5356	163	22	under	under	ADP
easat-5356	163	23	more	more	ADV
easat-5356	163	24	challenging	challenging	ADJ
easat-5356	163	25	conditions	condition	NOUN
easat-5356	163	26	,	,	PUNCT
easat-5356	163	27	therefore	therefore	ADV
easat-5356	163	28	enhancing	enhance	VERB
easat-5356	163	29	overall	overall	ADJ
easat-5356	163	30	detection	detection	NOUN
easat-5356	163	31	accuracy	accuracy	NOUN
easat-5356	163	32	and	and	CCONJ
easat-5356	163	33	robustness	robustness	NOUN
easat-5356	163	34	[	[	X
easat-5356	163	35	37	37	NUM
easat-5356	163	36	]	]	SYM
easat-5356	163	37	.	.	PUNCT
easat-5356	164	1	3	3	X
easat-5356	164	2	.	.	X
easat-5356	164	3	method	method	PROPN
easat-5356	164	4	3.1	3.1	NUM
easat-5356	164	5	.	.	PUNCT
easat-5356	164	6	gcsem	gcsem	ADJ
easat-5356	164	7	-	-	PUNCT
easat-5356	164	8	yolo	yolo	PROPN
easat-5356	164	9	model	model	NOUN
easat-5356	164	10	overview	overview	VERB
easat-5356	164	11	the	the	DET
easat-5356	164	12	structure	structure	NOUN
easat-5356	164	13	of	of	ADP
easat-5356	164	14	gcsem	gcsem	NOUN
easat-5356	164	15	-	-	PUNCT
easat-5356	164	16	yolo	yolo	PROPN
easat-5356	164	17	is	be	AUX
easat-5356	164	18	depicted	depict	VERB
easat-5356	164	19	in	in	ADP
easat-5356	164	20	figure	figure	NOUN
easat-5356	164	21	1.we	1.we	NUM
easat-5356	164	22	have	have	AUX
easat-5356	164	23	made	make	VERB
easat-5356	164	24	many	many	ADJ
easat-5356	164	25	adjustments	adjustment	NOUN
easat-5356	164	26	and	and	CCONJ
easat-5356	164	27	enhancements	enhancement	NOUN
easat-5356	164	28	to	to	ADP
easat-5356	164	29	the	the	DET
easat-5356	164	30	yolov8n	yolov8n	PROPN
easat-5356	164	31	fundamental	fundamental	ADJ
easat-5356	164	32	design	design	NOUN
easat-5356	164	33	,	,	PUNCT
easat-5356	164	34	with	with	ADP
easat-5356	164	35	a	a	DET
easat-5356	164	36	particular	particular	ADJ
easat-5356	164	37	emphasis	emphasis	NOUN
easat-5356	164	38	on	on	ADP
easat-5356	164	39	the	the	DET
easat-5356	164	40	neck	neck	NOUN
easat-5356	164	41	and	and	CCONJ
easat-5356	164	42	backbone	backbone	NOUN
easat-5356	164	43	parts	part	NOUN
easat-5356	164	44	.	.	PUNCT
easat-5356	165	1	845	845	NUM
easat-5356	165	2	edelweiss	edelweiss	PROPN
easat-5356	165	3	applied	apply	VERB
easat-5356	165	4	science	science	NOUN
easat-5356	165	5	and	and	CCONJ
easat-5356	165	6	technology	technology	NOUN
easat-5356	165	7	issn	issn	PROPN
easat-5356	165	8	:	:	PUNCT
easat-5356	165	9	2576	2576	NUM
easat-5356	165	10	-	-	SYM
easat-5356	165	11	8484	8484	NUM
easat-5356	165	12	vol	vol	NOUN
easat-5356	165	13	.	.	PROPN
easat-5356	166	1	9	9	NUM
easat-5356	166	2	,	,	PUNCT
easat-5356	166	3	no	no	INTJ
easat-5356	166	4	.	.	NOUN
easat-5356	167	1	3	3	NUM
easat-5356	167	2	:	:	PUNCT
easat-5356	167	3	840	840	NUM
easat-5356	167	4	-	-	SYM
easat-5356	167	5	855	855	NUM
easat-5356	167	6	,	,	PUNCT
easat-5356	167	7	2025	2025	NUM
easat-5356	167	8	doi	doi	NOUN
easat-5356	167	9	:	:	PUNCT
easat-5356	167	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	167	11	©	©	PROPN
easat-5356	167	12	2025	2025	NUM
easat-5356	167	13	by	by	ADP
easat-5356	167	14	the	the	DET
easat-5356	167	15	authors	author	NOUN
easat-5356	167	16	;	;	PUNCT
easat-5356	167	17	licensee	licensee	PROPN
easat-5356	167	18	learning	learn	VERB
easat-5356	167	19	gate	gate	NOUN
easat-5356	167	20	figure	figure	NOUN
easat-5356	167	21	1	1	NUM
easat-5356	167	22	.	.	PUNCT
easat-5356	168	1	yolov8	yolov8	PROPN
easat-5356	168	2	's	's	PART
easat-5356	168	3	network	network	NOUN
easat-5356	168	4	structure	structure	NOUN
easat-5356	168	5	.	.	PUNCT
easat-5356	169	1	the	the	DET
easat-5356	169	2	size	size	NOUN
easat-5356	169	3	of	of	ADP
easat-5356	169	4	the	the	DET
easat-5356	169	5	feature	feature	NOUN
easat-5356	169	6	map	map	NOUN
easat-5356	169	7	is	be	AUX
easat-5356	169	8	represented	represent	VERB
easat-5356	169	9	by	by	ADP
easat-5356	169	10	the	the	DET
easat-5356	169	11	parameters	parameter	NOUN
easat-5356	169	12	w	w	ADJ
easat-5356	169	13	(	(	PUNCT
easat-5356	169	14	width	width	PROPN
easat-5356	169	15	)	)	PUNCT
easat-5356	169	16	and	and	CCONJ
easat-5356	169	17	r	r	NOUN
easat-5356	169	18	(	(	PUNCT
easat-5356	169	19	ratio	ratio	NOUN
easat-5356	169	20	)	)	PUNCT
easat-5356	169	21	in	in	ADP
easat-5356	169	22	figure	figure	NOUN
easat-5356	169	23	1	1	NUM
easat-5356	169	24	.	.	PUNCT
easat-5356	169	25	by	by	ADP
easat-5356	169	26	adjusting	adjust	VERB
easat-5356	169	27	the	the	DET
easat-5356	169	28	values	value	NOUN
easat-5356	169	29	of	of	ADP
easat-5356	169	30	w	w	PROPN
easat-5356	169	31	and	and	CCONJ
easat-5356	169	32	r	r	NOUN
easat-5356	169	33	to	to	PART
easat-5356	169	34	suit	suit	VERB
easat-5356	169	35	the	the	DET
easat-5356	169	36	requirements	requirement	NOUN
easat-5356	169	37	of	of	ADP
easat-5356	169	38	various	various	ADJ
easat-5356	169	39	application	application	NOUN
easat-5356	169	40	scenarios	scenario	NOUN
easat-5356	169	41	,	,	PUNCT
easat-5356	169	42	the	the	DET
easat-5356	169	43	model	model	NOUN
easat-5356	169	44	's	's	PART
easat-5356	169	45	size	size	NOUN
easat-5356	169	46	may	may	AUX
easat-5356	169	47	be	be	AUX
easat-5356	169	48	managed	manage	VERB
easat-5356	169	49	.	.	PUNCT
easat-5356	170	1	four	four	NUM
easat-5356	170	2	feature	feature	NOUN
easat-5356	170	3	maps	map	NOUN
easat-5356	170	4	with	with	ADP
easat-5356	170	5	dimensions	dimension	NOUN
easat-5356	170	6	of	of	ADP
easat-5356	170	7	160	160	NUM
easat-5356	170	8	×	×	NOUN
easat-5356	170	9	160	160	NUM
easat-5356	170	10	,	,	PUNCT
easat-5356	170	11	80	80	NUM
easat-5356	170	12	×	×	NOUN
easat-5356	170	13	80	80	NUM
easat-5356	170	14	,	,	PUNCT
easat-5356	170	15	40	40	NUM
easat-5356	170	16	×	×	NOUN
easat-5356	170	17	40	40	NUM
easat-5356	170	18	,	,	PUNCT
easat-5356	170	19	20	20	NUM
easat-5356	170	20	×	×	NOUN
easat-5356	170	21	20	20	NUM
easat-5356	170	22	are	be	AUX
easat-5356	170	23	produced	produce	VERB
easat-5356	170	24	by	by	ADP
easat-5356	170	25	the	the	DET
easat-5356	170	26	network	network	NOUN
easat-5356	170	27	.	.	PUNCT
easat-5356	171	1	3.2	3.2	NUM
easat-5356	171	2	.	.	PUNCT
easat-5356	172	1	feature	feature	NOUN
easat-5356	172	2	extraction	extraction	NOUN
easat-5356	172	3	module	module	NOUN
easat-5356	172	4	with	with	ADP
easat-5356	172	5	squeeze	squeeze	NOUN
easat-5356	172	6	-	-	PUNCT
easat-5356	172	7	and	and	CCONJ
easat-5356	172	8	-	-	PUNCT
easat-5356	172	9	excitation	excitation	NOUN
easat-5356	172	10	(	(	PUNCT
easat-5356	172	11	se	se	ADJ
easat-5356	172	12	)	)	PUNCT
easat-5356	172	13	and	and	CCONJ
easat-5356	172	14	group	group	NOUN
easat-5356	172	15	convolutions	convolution	NOUN
easat-5356	172	16	in	in	ADP
easat-5356	172	17	our	our	PRON
easat-5356	172	18	endeavour	endeavour	NOUN
easat-5356	172	19	to	to	PART
easat-5356	172	20	bolster	bolster	VERB
easat-5356	172	21	the	the	DET
easat-5356	172	22	feature	feature	NOUN
easat-5356	172	23	extraction	extraction	NOUN
easat-5356	172	24	process	process	NOUN
easat-5356	172	25	of	of	ADP
easat-5356	172	26	yolov8	yolov8	NOUN
easat-5356	173	1	,	,	PUNCT
easat-5356	173	2	we	we	PRON
easat-5356	173	3	have	have	AUX
easat-5356	173	4	engineered	engineer	VERB
easat-5356	173	5	an	an	DET
easat-5356	173	6	innovative	innovative	ADJ
easat-5356	173	7	gcsem	gcsem	NOUN
easat-5356	173	8	module	module	NOUN
easat-5356	173	9	that	that	PRON
easat-5356	173	10	seamlessly	seamlessly	ADV
easat-5356	173	11	integrates	integrate	VERB
easat-5356	173	12	the	the	DET
easat-5356	173	13	squeeze	squeeze	NOUN
easat-5356	173	14	-	-	PUNCT
easat-5356	173	15	and	and	CCONJ
easat-5356	173	16	-	-	PUNCT
easat-5356	173	17	excitation	excitation	NOUN
easat-5356	173	18	(	(	PUNCT
easat-5356	173	19	se	se	ADJ
easat-5356	173	20	)	)	PUNCT
easat-5356	173	21	mechanism	mechanism	NOUN
easat-5356	173	22	and	and	CCONJ
easat-5356	173	23	spatial	spatial	ADJ
easat-5356	173	24	attention	attention	NOUN
easat-5356	173	25	,	,	PUNCT
easat-5356	173	26	enhancing	enhance	VERB
easat-5356	173	27	the	the	DET
easat-5356	173	28	network	network	NOUN
easat-5356	173	29	's	's	PART
easat-5356	173	30	ability	ability	NOUN
easat-5356	173	31	to	to	PART
easat-5356	173	32	discern	discern	VERB
easat-5356	173	33	and	and	CCONJ
easat-5356	173	34	emphasize	emphasize	VERB
easat-5356	173	35	pivotal	pivotal	ADJ
easat-5356	173	36	features	feature	NOUN
easat-5356	173	37	within	within	ADP
easat-5356	173	38	the	the	DET
easat-5356	173	39	input	input	NOUN
easat-5356	173	40	data	datum	NOUN
easat-5356	173	41	.	.	PUNCT
easat-5356	174	1	group	group	NOUN
easat-5356	174	2	convolution	convolution	NOUN
easat-5356	174	3	:	:	PUNCT
easat-5356	174	4	in	in	ADP
easat-5356	174	5	our	our	PRON
easat-5356	174	6	module	module	NOUN
easat-5356	174	7	,	,	PUNCT
easat-5356	174	8	we	we	PRON
easat-5356	174	9	use	use	VERB
easat-5356	174	10	group	group	NOUN
easat-5356	174	11	convolutions	convolution	NOUN
easat-5356	174	12	to	to	PART
easat-5356	174	13	balance	balance	VERB
easat-5356	174	14	feature	feature	NOUN
easat-5356	174	15	representation	representation	NOUN
easat-5356	174	16	with	with	ADP
easat-5356	174	17	parameter	parameter	NOUN
easat-5356	174	18	economy	economy	NOUN
easat-5356	174	19	.	.	PUNCT
easat-5356	175	1	by	by	ADP
easat-5356	175	2	dividing	divide	VERB
easat-5356	175	3	the	the	DET
easat-5356	175	4	input	input	NOUN
easat-5356	175	5	channels	channel	NOUN
easat-5356	175	6	into	into	ADP
easat-5356	175	7	groups	group	NOUN
easat-5356	175	8	,	,	PUNCT
easat-5356	175	9	this	this	DET
easat-5356	175	10	method	method	NOUN
easat-5356	175	11	enables	enable	VERB
easat-5356	175	12	the	the	DET
easat-5356	175	13	network	network	NOUN
easat-5356	175	14	to	to	PART
easat-5356	175	15	concentrate	concentrate	VERB
easat-5356	175	16	on	on	ADP
easat-5356	175	17	local	local	ADJ
easat-5356	175	18	groups	group	NOUN
easat-5356	175	19	of	of	ADP
easat-5356	175	20	channels	channel	NOUN
easat-5356	175	21	and	and	CCONJ
easat-5356	175	22	acquire	acquire	VERB
easat-5356	175	23	more	more	ADJ
easat-5356	175	24	generalized	generalized	ADJ
easat-5356	175	25	features	feature	NOUN
easat-5356	175	26	.	.	PUNCT
easat-5356	176	1	this	this	DET
easat-5356	176	2	method	method	NOUN
easat-5356	176	3	lessens	lessen	VERB
easat-5356	176	4	the	the	DET
easat-5356	176	5	computational	computational	ADJ
easat-5356	176	6	load	load	NOUN
easat-5356	176	7	related	relate	VERB
easat-5356	176	8	to	to	ADP
easat-5356	176	9	managing	manage	VERB
easat-5356	176	10	big	big	ADJ
easat-5356	176	11	parameter	parameter	NOUN
easat-5356	176	12	sets	set	NOUN
easat-5356	176	13	while	while	SCONJ
easat-5356	176	14	simultaneously	simultaneously	ADV
easat-5356	176	15	enhancing	enhance	VERB
easat-5356	176	16	the	the	DET
easat-5356	176	17	model	model	NOUN
easat-5356	176	18	's	's	PART
easat-5356	176	19	capacity	capacity	NOUN
easat-5356	176	20	to	to	PART
easat-5356	176	21	identify	identify	VERB
easat-5356	176	22	various	various	ADJ
easat-5356	176	23	and	and	CCONJ
easat-5356	176	24	pertinent	pertinent	ADJ
easat-5356	176	25	patterns	pattern	NOUN
easat-5356	176	26	across	across	ADP
easat-5356	176	27	various	various	ADJ
easat-5356	176	28	spatial	spatial	ADJ
easat-5356	176	29	locations	location	NOUN
easat-5356	176	30	.	.	PUNCT
easat-5356	177	1	the	the	DET
easat-5356	177	2	technique	technique	NOUN
easat-5356	177	3	improves	improve	VERB
easat-5356	177	4	computing	compute	VERB
easat-5356	177	5	efficiency	efficiency	NOUN
easat-5356	177	6	without	without	ADP
easat-5356	177	7	compromising	compromise	VERB
easat-5356	177	8	the	the	DET
easat-5356	177	9	model	model	NOUN
easat-5356	177	10	's	's	PART
easat-5356	177	11	capacity	capacity	NOUN
easat-5356	177	12	to	to	PART
easat-5356	177	13	identify	identify	VERB
easat-5356	177	14	intricate	intricate	ADJ
easat-5356	177	15	relationships	relationship	NOUN
easat-5356	177	16	in	in	ADP
easat-5356	177	17	the	the	DET
easat-5356	177	18	data	datum	NOUN
easat-5356	177	19	by	by	ADP
easat-5356	177	20	reducing	reduce	VERB
easat-5356	177	21	the	the	DET
easat-5356	177	22	number	number	NOUN
easat-5356	177	23	of	of	ADP
easat-5356	177	24	learning	learn	VERB
easat-5356	177	25	parameters	parameter	NOUN
easat-5356	177	26	.	.	PUNCT
easat-5356	178	1	as	as	ADP
easat-5356	178	2	a	a	DET
easat-5356	178	3	result	result	NOUN
easat-5356	178	4	,	,	PUNCT
easat-5356	178	5	the	the	DET
easat-5356	178	6	network	network	NOUN
easat-5356	178	7	architecture	architecture	NOUN
easat-5356	178	8	becomes	become	VERB
easat-5356	178	9	more	more	ADV
easat-5356	178	10	effective	effective	ADJ
easat-5356	178	11	and	and	CCONJ
easat-5356	178	12	scalable	scalable	ADJ
easat-5356	178	13	.	.	PUNCT
easat-5356	179	1	mathematically	mathematically	ADV
easat-5356	179	2	,	,	PUNCT
easat-5356	179	3	the	the	DET
easat-5356	179	4	feature	feature	NOUN
easat-5356	179	5	extraction	extraction	NOUN
easat-5356	179	6	process	process	NOUN
easat-5356	179	7	within	within	ADP
easat-5356	179	8	our	our	PRON
easat-5356	179	9	module	module	NOUN
easat-5356	179	10	can	can	AUX
easat-5356	179	11	be	be	AUX
easat-5356	179	12	articulated	articulate	VERB
easat-5356	179	13	as	as	ADP
easat-5356	179	14	:	:	PUNCT
easat-5356	179	15	𝑋	𝑋	PROPN
easat-5356	179	16	=	=	PUNCT
easat-5356	179	17	conv1(conv2(𝑥	conv1(conv2(𝑥	PROPN
easat-5356	179	18	)	)	PUNCT
easat-5356	179	19	)	)	PUNCT
easat-5356	180	1	(	(	PUNCT
easat-5356	180	2	1	1	X
easat-5356	180	3	)	)	PUNCT
easat-5356	180	4	where	where	SCONJ
easat-5356	180	5	conv1	conv1	PROPN
easat-5356	180	6	denotes	denote	VERB
easat-5356	180	7	a	a	DET
easat-5356	180	8	group	group	NOUN
easat-5356	180	9	convolution	convolution	NOUN
easat-5356	180	10	operation	operation	NOUN
easat-5356	180	11	,	,	PUNCT
easat-5356	180	12	and	and	CCONJ
easat-5356	180	13	conv2	conv2	NOUN
easat-5356	180	14	represents	represent	VERB
easat-5356	180	15	a	a	DET
easat-5356	180	16	standard	standard	ADJ
easat-5356	180	17	convolution	convolution	NOUN
easat-5356	180	18	layer	layer	NOUN
easat-5356	180	19	tasked	task	VERB
easat-5356	180	20	with	with	ADP
easat-5356	180	21	downsampling	downsample	VERB
easat-5356	180	22	the	the	DET
easat-5356	180	23	feature	feature	NOUN
easat-5356	180	24	map	map	NOUN
easat-5356	180	25	,	,	PUNCT
easat-5356	180	26	the	the	DET
easat-5356	180	27	input	input	NOUN
easat-5356	180	28	data	data	NOUN
easat-5356	180	29	is	be	AUX
easat-5356	180	30	x	x	PUNCT
easat-5356	180	31	squeeze	squeeze	NOUN
easat-5356	180	32	-	-	PUNCT
easat-5356	180	33	and	and	CCONJ
easat-5356	180	34	-	-	PUNCT
easat-5356	180	35	excitation	excitation	NOUN
easat-5356	180	36	(	(	PUNCT
easat-5356	180	37	se	se	ADJ
easat-5356	180	38	)	)	PUNCT
easat-5356	180	39	mechanism	mechanism	NOUN
easat-5356	180	40	:	:	PUNCT
easat-5356	180	41	the	the	DET
easat-5356	180	42	se	se	PROPN
easat-5356	180	43	module	module	PROPN
easat-5356	180	44	cai	cai	PROPN
easat-5356	180	45	,	,	PUNCT
easat-5356	180	46	et	et	PROPN
easat-5356	180	47	al	al	PROPN
easat-5356	180	48	.	.	PUNCT
easat-5356	181	1	[	[	X
easat-5356	181	2	38	38	NUM
easat-5356	181	3	]	]	PUNCT
easat-5356	181	4	plays	play	VERB
easat-5356	181	5	a	a	DET
easat-5356	181	6	pivotal	pivotal	ADJ
easat-5356	181	7	role	role	NOUN
easat-5356	181	8	in	in	ADP
easat-5356	181	9	recalibrating	recalibrate	VERB
easat-5356	181	10	channel	channel	NOUN
easat-5356	181	11	-	-	PUNCT
easat-5356	181	12	wise	wise	ADJ
easat-5356	181	13	feature	feature	NOUN
easat-5356	181	14	responses	response	NOUN
easat-5356	181	15	by	by	ADP
easat-5356	181	16	adaptively	adaptively	ADV
easat-5356	181	17	scaling	scale	VERB
easat-5356	181	18	them	they	PRON
easat-5356	181	19	based	base	VERB
easat-5356	181	20	on	on	ADP
easat-5356	181	21	their	their	PRON
easat-5356	181	22	global	global	ADJ
easat-5356	181	23	846	846	NUM
easat-5356	181	24	edelweiss	edelweiss	PROPN
easat-5356	181	25	applied	apply	VERB
easat-5356	181	26	science	science	NOUN
easat-5356	181	27	and	and	CCONJ
easat-5356	181	28	technology	technology	NOUN
easat-5356	181	29	issn	issn	PROPN
easat-5356	181	30	:	:	PUNCT
easat-5356	181	31	2576	2576	NUM
easat-5356	181	32	-	-	SYM
easat-5356	181	33	8484	8484	NUM
easat-5356	181	34	vol	vol	NOUN
easat-5356	181	35	.	.	PROPN
easat-5356	182	1	9	9	NUM
easat-5356	182	2	,	,	PUNCT
easat-5356	182	3	no	no	INTJ
easat-5356	182	4	.	.	NOUN
easat-5356	183	1	3	3	NUM
easat-5356	183	2	:	:	PUNCT
easat-5356	183	3	840	840	NUM
easat-5356	183	4	-	-	SYM
easat-5356	183	5	855	855	NUM
easat-5356	183	6	,	,	PUNCT
easat-5356	183	7	2025	2025	NUM
easat-5356	183	8	doi	doi	NOUN
easat-5356	183	9	:	:	PUNCT
easat-5356	183	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	183	11	©	©	PROPN
easat-5356	183	12	2025	2025	NUM
easat-5356	183	13	by	by	ADP
easat-5356	183	14	the	the	DET
easat-5356	183	15	authors	author	NOUN
easat-5356	183	16	;	;	PUNCT
easat-5356	183	17	licensee	licensee	PROPN
easat-5356	183	18	learning	learn	VERB
easat-5356	183	19	gate	gate	NOUN
easat-5356	183	20	importance	importance	NOUN
easat-5356	183	21	.	.	PUNCT
easat-5356	184	1	this	this	DET
easat-5356	184	2	recalibration	recalibration	NOUN
easat-5356	184	3	improves	improve	VERB
easat-5356	184	4	feature	feature	NOUN
easat-5356	184	5	representation	representation	NOUN
easat-5356	184	6	and	and	CCONJ
easat-5356	184	7	discrimination	discrimination	NOUN
easat-5356	184	8	by	by	ADP
easat-5356	184	9	allowing	allow	VERB
easat-5356	184	10	the	the	DET
easat-5356	184	11	model	model	NOUN
easat-5356	184	12	to	to	PART
easat-5356	184	13	preferentially	preferentially	ADV
easat-5356	184	14	focus	focus	VERB
easat-5356	184	15	on	on	ADP
easat-5356	184	16	the	the	DET
easat-5356	184	17	most	most	ADV
easat-5356	184	18	important	important	ADJ
easat-5356	184	19	spatial	spatial	ADJ
easat-5356	184	20	information	information	NOUN
easat-5356	184	21	.	.	PUNCT
easat-5356	185	1	by	by	ADP
easat-5356	185	2	emphasizing	emphasize	VERB
easat-5356	185	3	relevant	relevant	ADJ
easat-5356	185	4	features	feature	NOUN
easat-5356	185	5	and	and	CCONJ
easat-5356	185	6	suppressing	suppress	VERB
easat-5356	185	7	less	less	ADV
easat-5356	185	8	informative	informative	ADJ
easat-5356	185	9	ones	one	NOUN
easat-5356	185	10	,	,	PUNCT
easat-5356	185	11	the	the	DET
easat-5356	185	12	se	se	PROPN
easat-5356	185	13	module	module	NOUN
easat-5356	185	14	not	not	PART
easat-5356	185	15	only	only	ADV
easat-5356	185	16	refines	refine	VERB
easat-5356	185	17	the	the	DET
easat-5356	185	18	model	model	NOUN
easat-5356	185	19	’s	’s	PART
easat-5356	185	20	performance	performance	NOUN
easat-5356	185	21	in	in	ADP
easat-5356	185	22	detecting	detect	VERB
easat-5356	185	23	key	key	ADJ
easat-5356	185	24	objects	object	NOUN
easat-5356	185	25	but	but	CCONJ
easat-5356	185	26	also	also	ADV
easat-5356	185	27	boosts	boost	VERB
easat-5356	185	28	its	its	PRON
easat-5356	185	29	robustness	robustness	NOUN
easat-5356	185	30	against	against	ADP
easat-5356	185	31	varying	vary	VERB
easat-5356	185	32	contextual	contextual	ADJ
easat-5356	185	33	changes	change	NOUN
easat-5356	185	34	,	,	PUNCT
easat-5356	185	35	such	such	ADJ
easat-5356	185	36	as	as	ADP
easat-5356	185	37	illumination	illumination	NOUN
easat-5356	185	38	or	or	CCONJ
easat-5356	185	39	pose	pose	NOUN
easat-5356	185	40	variations	variation	NOUN
easat-5356	185	41	,	,	PUNCT
easat-5356	185	42	making	make	VERB
easat-5356	185	43	it	it	PRON
easat-5356	185	44	more	more	ADV
easat-5356	185	45	effective	effective	ADJ
easat-5356	185	46	in	in	ADP
easat-5356	185	47	dynamic	dynamic	ADJ
easat-5356	185	48	and	and	CCONJ
easat-5356	185	49	complex	complex	ADJ
easat-5356	185	50	environments	environment	NOUN
easat-5356	185	51	.	.	PUNCT
easat-5356	186	1	after	after	ADP
easat-5356	186	2	the	the	DET
easat-5356	186	3	first	first	ADJ
easat-5356	186	4	step	step	NOUN
easat-5356	186	5	of	of	ADP
easat-5356	186	6	convolution	convolution	NOUN
easat-5356	186	7	,	,	PUNCT
easat-5356	186	8	the	the	DET
easat-5356	186	9	feature	feature	NOUN
easat-5356	186	10	map	map	NOUN
easat-5356	186	11	x	x	PUNCT
easat-5356	186	12	becomes	become	VERB
easat-5356	186	13	the	the	DET
easat-5356	186	14	feature	feature	NOUN
easat-5356	186	15	map	map	NOUN
easat-5356	186	16	u	u	NOUN
easat-5356	186	17	:	:	PUNCT
easat-5356	186	18	uc	uc	PROPN
easat-5356	186	19	=	=	PUNCT
easat-5356	186	20	∑	∑	PUNCT
easat-5356	186	21	  	  	SPACE
easat-5356	186	22	c′	c′	NOUN
easat-5356	187	1	s=1	s=1	ADP
easat-5356	187	2	vc	vc	PROPN
easat-5356	187	3	s	s	PROPN
easat-5356	187	4	∗	∗	X
easat-5356	187	5	xs	xs	PROPN
easat-5356	187	6	(	(	PUNCT
easat-5356	187	7	2	2	NUM
easat-5356	187	8	)	)	PUNCT
easat-5356	187	9	to	to	PART
easat-5356	187	10	overcome	overcome	VERB
easat-5356	187	11	the	the	DET
easat-5356	187	12	challenge	challenge	NOUN
easat-5356	187	13	of	of	ADP
easat-5356	187	14	exploiting	exploit	VERB
easat-5356	187	15	channel	channel	NOUN
easat-5356	187	16	dependency	dependency	NOUN
easat-5356	187	17	,	,	PUNCT
easat-5356	187	18	the	the	DET
easat-5356	187	19	feature	feature	NOUN
easat-5356	187	20	map	map	NOUN
easat-5356	187	21	containing	contain	VERB
easat-5356	187	22	global	global	ADJ
easat-5356	187	23	information	information	NOUN
easat-5356	187	24	is	be	AUX
easat-5356	187	25	compressed	compress	VERB
easat-5356	187	26	directly	directly	ADV
easat-5356	187	27	into	into	ADP
easat-5356	187	28	a	a	DET
easat-5356	187	29	feature	feature	NOUN
easat-5356	187	30	vector	vector	NOUN
easat-5356	187	31	,	,	PUNCT
easat-5356	187	32	and	and	CCONJ
easat-5356	187	33	the	the	DET
easat-5356	187	34	global	global	ADJ
easat-5356	187	35	spatial	spatial	ADJ
easat-5356	187	36	information	information	NOUN
easat-5356	187	37	is	be	AUX
easat-5356	187	38	selected	select	VERB
easat-5356	187	39	to	to	PART
easat-5356	187	40	be	be	AUX
easat-5356	187	41	compressed	compress	VERB
easat-5356	187	42	into	into	ADP
easat-5356	187	43	a	a	DET
easat-5356	187	44	single	single	ADJ
easat-5356	187	45	channel	channel	NOUN
easat-5356	187	46	descriptor	descriptor	NOUN
easat-5356	187	47	,	,	PUNCT
easat-5356	187	48	i.e.	i.e.	X
easat-5356	187	49	,	,	PUNCT
easat-5356	187	50	the	the	DET
easat-5356	187	51	global	global	ADJ
easat-5356	187	52	average	average	ADJ
easat-5356	187	53	pooling	pooling	NOUN
easat-5356	187	54	of	of	ADP
easat-5356	187	55	channels	channel	NOUN
easat-5356	187	56	is	be	AUX
easat-5356	187	57	employed	employ	VERB
easat-5356	187	58	.	.	PUNCT
easat-5356	188	1	because	because	SCONJ
easat-5356	188	2	all	all	PRON
easat-5356	188	3	of	of	ADP
easat-5356	188	4	the	the	DET
easat-5356	188	5	channel	channel	NOUN
easat-5356	188	6	features	feature	NOUN
easat-5356	188	7	of	of	ADP
easat-5356	188	8	the	the	DET
easat-5356	188	9	feature	feature	NOUN
easat-5356	188	10	maps	map	NOUN
easat-5356	188	11	are	be	AUX
easat-5356	188	12	condensed	condense	VERB
easat-5356	188	13	into	into	ADP
easat-5356	188	14	a	a	DET
easat-5356	188	15	single	single	ADJ
easat-5356	188	16	numerical	numerical	ADJ
easat-5356	188	17	value	value	NOUN
easat-5356	188	18	,	,	PUNCT
easat-5356	188	19	the	the	DET
easat-5356	188	20	issue	issue	NOUN
easat-5356	188	21	of	of	ADP
easat-5356	188	22	channel	channel	NOUN
easat-5356	188	23	reliance	reliance	NOUN
easat-5356	188	24	is	be	AUX
easat-5356	188	25	lessened	lessen	VERB
easat-5356	188	26	and	and	CCONJ
easat-5356	188	27	contextual	contextual	ADJ
easat-5356	188	28	information	information	NOUN
easat-5356	188	29	is	be	AUX
easat-5356	188	30	included	include	VERB
easat-5356	188	31	in	in	ADP
easat-5356	188	32	the	the	DET
easat-5356	188	33	resulting	result	VERB
easat-5356	188	34	channel	channel	NOUN
easat-5356	188	35	-	-	PUNCT
easat-5356	188	36	level	level	NOUN
easat-5356	188	37	statistics	statistic	NOUN
easat-5356	188	38	.	.	PUNCT
easat-5356	189	1	the	the	DET
easat-5356	189	2	following	follow	VERB
easat-5356	189	3	is	be	AUX
easat-5356	189	4	the	the	DET
easat-5356	189	5	definition	definition	NOUN
easat-5356	189	6	:	:	PUNCT
easat-5356	189	7	𝑧𝑐	𝑧𝑐	NOUN
easat-5356	189	8	=	=	SYM
easat-5356	190	1	1	1	NUM
easat-5356	190	2	𝐻	𝐻	NOUN
easat-5356	190	3	×	×	NOUN
easat-5356	190	4	𝑊	𝑊	PROPN
easat-5356	190	5	∑	∑	DET
easat-5356	190	6	  	  	SPACE
easat-5356	190	7	𝐻	𝐻	PROPN
easat-5356	190	8	𝑖=1	𝑖=1	PUNCT
easat-5356	190	9	∑	∑	PUNCT
easat-5356	190	10	  	  	SPACE
easat-5356	190	11	𝑊	𝑊	PROPN
easat-5356	190	12	𝑗=1	𝑗=1	PROPN
easat-5356	190	13	𝑢𝑐(𝑖	𝑢𝑐(𝑖	PROPN
easat-5356	190	14	,	,	PUNCT
easat-5356	190	15	𝑗	𝑗	NOUN
easat-5356	190	16	)	)	PUNCT
easat-5356	190	17	(	(	PUNCT
easat-5356	190	18	3	3	X
easat-5356	190	19	)	)	PUNCT
easat-5356	190	20	spatial	spatial	ADJ
easat-5356	190	21	attention	attention	NOUN
easat-5356	190	22	:	:	PUNCT
easat-5356	190	23	we	we	PRON
easat-5356	190	24	take	take	VERB
easat-5356	190	25	a	a	DET
easat-5356	190	26	two	two	NUM
easat-5356	190	27	-	-	PUNCT
easat-5356	190	28	pronged	pronged	ADJ
easat-5356	190	29	approach	approach	NOUN
easat-5356	190	30	to	to	ADP
easat-5356	190	31	spatial	spatial	ADJ
easat-5356	190	32	attention	attention	NOUN
easat-5356	190	33	,	,	PUNCT
easat-5356	190	34	extracting	extract	VERB
easat-5356	190	35	the	the	DET
easat-5356	190	36	most	most	ADV
easat-5356	190	37	important	important	ADJ
easat-5356	190	38	spatial	spatial	ADJ
easat-5356	190	39	cues	cue	NOUN
easat-5356	190	40	for	for	ADP
easat-5356	190	41	each	each	DET
easat-5356	190	42	channel	channel	NOUN
easat-5356	190	43	by	by	ADP
easat-5356	190	44	combining	combine	VERB
easat-5356	190	45	adaptive	adaptive	ADJ
easat-5356	190	46	max	max	PROPN
easat-5356	190	47	pooling	pooling	NOUN
easat-5356	190	48	with	with	ADP
easat-5356	190	49	adaptive	adaptive	ADJ
easat-5356	190	50	average	average	ADJ
easat-5356	190	51	pooling	pooling	NOUN
easat-5356	190	52	.	.	PUNCT
easat-5356	191	1	in	in	ADP
easat-5356	191	2	particular	particular	ADJ
easat-5356	191	3	,	,	PUNCT
easat-5356	191	4	both	both	PRON
easat-5356	191	5	average	average	ADJ
easat-5356	191	6	pooling	pooling	NOUN
easat-5356	191	7	and	and	CCONJ
easat-5356	191	8	maximum	maximum	ADJ
easat-5356	191	9	pooling	pooling	NOUN
easat-5356	191	10	are	be	AUX
easat-5356	191	11	used	use	VERB
easat-5356	191	12	to	to	PART
easat-5356	191	13	process	process	VERB
easat-5356	191	14	the	the	DET
easat-5356	191	15	input	input	NOUN
easat-5356	191	16	feature	feature	NOUN
easat-5356	191	17	maps	map	NOUN
easat-5356	191	18	initially	initially	ADV
easat-5356	191	19	.	.	PUNCT
easat-5356	192	1	to	to	PART
easat-5356	192	2	create	create	VERB
easat-5356	192	3	a	a	DET
easat-5356	192	4	weight	weight	NOUN
easat-5356	192	5	vector	vector	NOUN
easat-5356	192	6	,	,	PUNCT
easat-5356	192	7	these	these	DET
easat-5356	192	8	two	two	NUM
easat-5356	192	9	pooling	pooling	NOUN
easat-5356	192	10	results	result	NOUN
easat-5356	192	11	are	be	AUX
easat-5356	192	12	concatenated	concatenate	VERB
easat-5356	192	13	and	and	CCONJ
easat-5356	192	14	then	then	ADV
easat-5356	192	15	run	run	VERB
easat-5356	192	16	through	through	ADP
easat-5356	192	17	a	a	DET
easat-5356	192	18	convolutional	convolutional	ADJ
easat-5356	192	19	layer	layer	NOUN
easat-5356	192	20	and	and	CCONJ
easat-5356	192	21	a	a	DET
easat-5356	192	22	sigmoid	sigmoid	NOUN
easat-5356	192	23	function	function	NOUN
easat-5356	192	24	.	.	PUNCT
easat-5356	193	1	the	the	DET
easat-5356	193	2	weighted	weight	VERB
easat-5356	193	3	feature	feature	NOUN
easat-5356	193	4	maps	map	NOUN
easat-5356	193	5	are	be	AUX
easat-5356	193	6	then	then	ADV
easat-5356	193	7	created	create	VERB
easat-5356	193	8	by	by	ADP
easat-5356	193	9	multiplying	multiply	VERB
easat-5356	193	10	this	this	DET
easat-5356	193	11	weight	weight	NOUN
easat-5356	193	12	vector	vector	NOUN
easat-5356	193	13	by	by	ADP
easat-5356	193	14	the	the	DET
easat-5356	193	15	input	input	NOUN
easat-5356	193	16	feature	feature	NOUN
easat-5356	193	17	maps	map	NOUN
easat-5356	193	18	.	.	PUNCT
easat-5356	194	1	this	this	DET
easat-5356	194	2	dual	dual	ADJ
easat-5356	194	3	strategy	strategy	NOUN
easat-5356	194	4	ensures	ensure	VERB
easat-5356	194	5	a	a	DET
easat-5356	194	6	comprehensive	comprehensive	ADJ
easat-5356	194	7	capture	capture	NOUN
easat-5356	194	8	of	of	ADP
easat-5356	194	9	spatial	spatial	ADJ
easat-5356	194	10	features	feature	NOUN
easat-5356	194	11	by	by	ADP
easat-5356	194	12	leveraging	leverage	VERB
easat-5356	194	13	both	both	CCONJ
easat-5356	194	14	average	average	ADJ
easat-5356	194	15	and	and	CCONJ
easat-5356	194	16	extreme	extreme	ADJ
easat-5356	194	17	responses	response	NOUN
easat-5356	194	18	,	,	PUNCT
easat-5356	194	19	thereby	thereby	ADV
easat-5356	194	20	equipping	equip	VERB
easat-5356	194	21	the	the	DET
easat-5356	194	22	model	model	NOUN
easat-5356	194	23	with	with	ADP
easat-5356	194	24	enhanced	enhanced	ADJ
easat-5356	194	25	sensitivity	sensitivity	NOUN
easat-5356	194	26	to	to	ADP
easat-5356	194	27	the	the	DET
easat-5356	194	28	geometric	geometric	ADJ
easat-5356	194	29	configuration	configuration	NOUN
easat-5356	194	30	of	of	ADP
easat-5356	194	31	the	the	DET
easat-5356	194	32	input	input	NOUN
easat-5356	194	33	data	datum	NOUN
easat-5356	194	34	and	and	CCONJ
easat-5356	194	35	improving	improve	VERB
easat-5356	194	36	its	its	PRON
easat-5356	194	37	ability	ability	NOUN
easat-5356	194	38	to	to	PART
easat-5356	194	39	focus	focus	VERB
easat-5356	194	40	on	on	ADP
easat-5356	194	41	relevant	relevant	ADJ
easat-5356	194	42	spatial	spatial	ADJ
easat-5356	194	43	information	information	NOUN
easat-5356	194	44	.	.	PUNCT
easat-5356	195	1	max	max	PROPN
easat-5356	195	2	pooling	pool	VERB
easat-5356	195	3	definition	definition	NOUN
easat-5356	195	4	is	be	AUX
easat-5356	195	5	as	as	SCONJ
easat-5356	195	6	follows	follow	VERB
easat-5356	195	7	:	:	PUNCT
easat-5356	195	8	𝑀(𝑋	𝑀(𝑋	X
easat-5356	195	9	)	)	PUNCT
easat-5356	195	10	=	=	PUNCT
easat-5356	196	1	𝑚𝑎𝑥	𝑚𝑎𝑥	PRON
easat-5356	196	2	𝑖,𝑗	𝑖,𝑗	VERB
easat-5356	196	3	 	 	SPACE
easat-5356	196	4	𝑋(𝑖	𝑋(𝑖	PROPN
easat-5356	196	5	,	,	PUNCT
easat-5356	196	6	𝑗	𝑗	NOUN
easat-5356	196	7	)	)	PUNCT
easat-5356	196	8	(	(	PUNCT
easat-5356	196	9	4	4	NUM
easat-5356	196	10	)	)	PUNCT
easat-5356	196	11	within	within	ADP
easat-5356	196	12	the	the	DET
easat-5356	196	13	channel	channel	NOUN
easat-5356	196	14	dimension	dimension	NOUN
easat-5356	196	15	,	,	PUNCT
easat-5356	196	16	the	the	DET
easat-5356	196	17	outcomes	outcome	NOUN
easat-5356	196	18	of	of	ADP
easat-5356	196	19	maximum	maximum	ADJ
easat-5356	196	20	pooling	pooling	NOUN
easat-5356	196	21	and	and	CCONJ
easat-5356	196	22	average	average	ADJ
easat-5356	196	23	pooling	pooling	NOUN
easat-5356	196	24	are	be	AUX
easat-5356	196	25	then	then	ADV
easat-5356	196	26	combined	combine	VERB
easat-5356	196	27	:	:	PUNCT
easat-5356	196	28	𝐶(𝑧𝑐	𝐶(𝑧𝑐	NOUN
easat-5356	196	29	,	,	PUNCT
easat-5356	196	30	𝑀(𝑥	𝑀(𝑥	NUM
easat-5356	196	31	)	)	PUNCT
easat-5356	196	32	)	)	PUNCT
easat-5356	197	1	=	=	PUNCT
easat-5356	198	1	[	[	X
easat-5356	198	2	𝑧𝑐	𝑧𝑐	NOUN
easat-5356	198	3	;	;	PUNCT
easat-5356	198	4	𝑀(𝑋	𝑀(𝑋	NUM
easat-5356	198	5	)	)	PUNCT
easat-5356	198	6	]	]	PUNCT
easat-5356	199	1	(	(	PUNCT
easat-5356	199	2	5	5	NUM
easat-5356	199	3	)	)	PUNCT
easat-5356	199	4	after	after	ADP
easat-5356	199	5	a	a	DET
easat-5356	199	6	layer	layer	NOUN
easat-5356	199	7	of	of	ADP
easat-5356	199	8	convolution	convolution	NOUN
easat-5356	199	9	and	and	CCONJ
easat-5356	199	10	sigmoid	sigmoid	NOUN
easat-5356	199	11	activation	activation	NOUN
easat-5356	199	12	:	:	PUNCT
easat-5356	199	13	𝐾(𝐶	𝐾(𝐶	NOUN
easat-5356	199	14	)	)	PUNCT
easat-5356	199	15	=	=	PUNCT
easat-5356	200	1	𝑊	𝑊	PROPN
easat-5356	200	2	⋅	⋅	PROPN
easat-5356	200	3	𝐶	𝐶	PROPN
easat-5356	200	4	+	+	CCONJ
easat-5356	200	5	𝑏	𝑏	PROPN
easat-5356	200	6	(	(	PUNCT
easat-5356	200	7	6	6	NUM
easat-5356	200	8	)	)	PUNCT
easat-5356	200	9	𝑆(𝐾	𝑆(𝐾	NOUN
easat-5356	200	10	)	)	PUNCT
easat-5356	200	11	=	=	SYM
easat-5356	200	12	1	1	NUM
easat-5356	200	13	1	1	NUM
easat-5356	200	14	+	+	NUM
easat-5356	200	15	𝑒−𝐾	𝑒−𝐾	ADJ
easat-5356	200	16	(	(	PUNCT
easat-5356	200	17	7	7	X
easat-5356	200	18	)	)	PUNCT
easat-5356	200	19	the	the	DET
easat-5356	200	20	weighted	weight	VERB
easat-5356	200	21	feature	feature	NOUN
easat-5356	200	22	map	map	NOUN
easat-5356	200	23	is	be	AUX
easat-5356	200	24	created	create	VERB
easat-5356	200	25	by	by	ADP
easat-5356	200	26	comparing	compare	VERB
easat-5356	200	27	the	the	DET
easat-5356	200	28	weight	weight	NOUN
easat-5356	200	29	vector	vector	NOUN
easat-5356	200	30	produced	produce	VERB
easat-5356	200	31	by	by	ADP
easat-5356	200	32	the	the	DET
easat-5356	200	33	sigmoid	sigmoid	NOUN
easat-5356	200	34	function	function	NOUN
easat-5356	200	35	with	with	ADP
easat-5356	200	36	the	the	DET
easat-5356	200	37	initial	initial	ADJ
easat-5356	200	38	input	input	NOUN
easat-5356	200	39	feature	feature	NOUN
easat-5356	200	40	map	map	NOUN
easat-5356	200	41	:	:	PUNCT
easat-5356	200	42	847	847	NUM
easat-5356	200	43	edelweiss	edelweiss	PROPN
easat-5356	200	44	applied	apply	VERB
easat-5356	200	45	science	science	NOUN
easat-5356	200	46	and	and	CCONJ
easat-5356	200	47	technology	technology	NOUN
easat-5356	200	48	issn	issn	PROPN
easat-5356	200	49	:	:	PUNCT
easat-5356	200	50	2576	2576	NUM
easat-5356	200	51	-	-	SYM
easat-5356	200	52	8484	8484	NUM
easat-5356	200	53	vol	vol	NOUN
easat-5356	200	54	.	.	PROPN
easat-5356	201	1	9	9	NUM
easat-5356	201	2	,	,	PUNCT
easat-5356	201	3	no	no	INTJ
easat-5356	201	4	.	.	NOUN
easat-5356	202	1	3	3	NUM
easat-5356	202	2	:	:	PUNCT
easat-5356	202	3	840	840	NUM
easat-5356	202	4	-	-	SYM
easat-5356	202	5	855	855	NUM
easat-5356	202	6	,	,	PUNCT
easat-5356	202	7	2025	2025	NUM
easat-5356	202	8	doi	doi	NOUN
easat-5356	202	9	:	:	PUNCT
easat-5356	202	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	202	11	©	©	PROPN
easat-5356	202	12	2025	2025	NUM
easat-5356	202	13	by	by	ADP
easat-5356	202	14	the	the	DET
easat-5356	202	15	authors	author	NOUN
easat-5356	202	16	;	;	PUNCT
easat-5356	202	17	licensee	licensee	PROPN
easat-5356	202	18	learning	learn	VERB
easat-5356	202	19	gate	gate	NOUN
easat-5356	202	20	𝑊(𝑥	𝑊(𝑥	NOUN
easat-5356	202	21	)	)	PUNCT
easat-5356	202	22	=	=	SYM
easat-5356	203	1	𝑆	𝑆	PROPN
easat-5356	203	2	⊙	⊙	NOUN
easat-5356	203	3	𝑥	𝑥	X
easat-5356	204	1	(	(	PUNCT
easat-5356	204	2	8)	8)	NUM
easat-5356	204	3	the	the	DET
easat-5356	204	4	gcsem	gcsem	NOUN
easat-5356	204	5	module	module	NOUN
easat-5356	204	6	's	's	PART
easat-5356	204	7	capacity	capacity	NOUN
easat-5356	204	8	to	to	PART
easat-5356	204	9	lessen	lessen	VERB
easat-5356	204	10	the	the	DET
easat-5356	204	11	loss	loss	NOUN
easat-5356	204	12	of	of	ADP
easat-5356	204	13	spatial	spatial	ADJ
easat-5356	204	14	information	information	NOUN
easat-5356	204	15	while	while	SCONJ
easat-5356	204	16	down	down	ADJ
easat-5356	204	17	sampling	sample	VERB
easat-5356	204	18	is	be	AUX
easat-5356	204	19	a	a	DET
easat-5356	204	20	key	key	ADJ
easat-5356	204	21	benefit	benefit	NOUN
easat-5356	204	22	.	.	PUNCT
easat-5356	205	1	preserving	preserve	VERB
easat-5356	205	2	sufficient	sufficient	ADJ
easat-5356	205	3	spatial	spatial	ADJ
easat-5356	205	4	information	information	NOUN
easat-5356	205	5	in	in	ADP
easat-5356	205	6	large	large	ADJ
easat-5356	205	7	-	-	PUNCT
easat-5356	205	8	scale	scale	NOUN
easat-5356	205	9	feature	feature	NOUN
easat-5356	205	10	maps	map	NOUN
easat-5356	205	11	is	be	AUX
easat-5356	205	12	crucial	crucial	ADJ
easat-5356	205	13	for	for	ADP
easat-5356	205	14	effective	effective	ADJ
easat-5356	205	15	target	target	NOUN
easat-5356	205	16	detection	detection	NOUN
easat-5356	205	17	,	,	PUNCT
easat-5356	205	18	especially	especially	ADV
easat-5356	205	19	in	in	ADP
easat-5356	205	20	face	face	NOUN
easat-5356	205	21	recognition	recognition	NOUN
easat-5356	205	22	scenarios	scenario	NOUN
easat-5356	205	23	.	.	PUNCT
easat-5356	206	1	although	although	SCONJ
easat-5356	206	2	down	down	ADJ
easat-5356	206	3	sampling	sample	VERB
easat-5356	206	4	improves	improve	VERB
easat-5356	206	5	overall	overall	ADJ
easat-5356	206	6	semantic	semantic	ADJ
easat-5356	206	7	understanding	understanding	NOUN
easat-5356	206	8	and	and	CCONJ
easat-5356	206	9	extracts	extract	VERB
easat-5356	206	10	high	high	ADJ
easat-5356	206	11	-	-	PUNCT
easat-5356	206	12	level	level	NOUN
easat-5356	206	13	semantic	semantic	ADJ
easat-5356	206	14	information	information	NOUN
easat-5356	206	15	,	,	PUNCT
easat-5356	206	16	it	it	PRON
easat-5356	206	17	frequently	frequently	ADV
easat-5356	206	18	results	result	VERB
easat-5356	206	19	in	in	ADP
easat-5356	206	20	the	the	DET
easat-5356	206	21	loss	loss	NOUN
easat-5356	206	22	of	of	ADP
easat-5356	206	23	geographical	geographical	ADJ
easat-5356	206	24	information	information	NOUN
easat-5356	206	25	.	.	PUNCT
easat-5356	207	1	in	in	ADP
easat-5356	207	2	order	order	NOUN
easat-5356	207	3	to	to	PART
easat-5356	207	4	solve	solve	VERB
easat-5356	207	5	this	this	DET
easat-5356	207	6	issue	issue	NOUN
easat-5356	207	7	,	,	PUNCT
easat-5356	207	8	the	the	DET
easat-5356	207	9	gcsem	gcsem	NOUN
easat-5356	207	10	module	module	NOUN
easat-5356	207	11	maintains	maintain	VERB
easat-5356	207	12	spatial	spatial	ADJ
easat-5356	207	13	information	information	NOUN
easat-5356	207	14	during	during	ADP
easat-5356	207	15	down	down	ADP
easat-5356	207	16	sampling	sample	VERB
easat-5356	207	17	.	.	PUNCT
easat-5356	208	1	by	by	ADP
easat-5356	208	2	explicitly	explicitly	ADV
easat-5356	208	3	modelling	model	VERB
easat-5356	208	4	the	the	DET
easat-5356	208	5	interdependencies	interdependency	NOUN
easat-5356	208	6	between	between	ADP
easat-5356	208	7	the	the	DET
easat-5356	208	8	convolved	convolve	VERB
easat-5356	208	9	feature	feature	NOUN
easat-5356	208	10	channels	channel	NOUN
easat-5356	208	11	,	,	PUNCT
easat-5356	208	12	broad	broad	ADJ
easat-5356	208	13	local	local	ADJ
easat-5356	208	14	receptive	receptive	ADJ
easat-5356	208	15	field	field	NOUN
easat-5356	208	16	features	feature	NOUN
easat-5356	208	17	are	be	AUX
easat-5356	208	18	extracted	extract	VERB
easat-5356	208	19	from	from	ADP
easat-5356	208	20	the	the	DET
easat-5356	208	21	feature	feature	NOUN
easat-5356	208	22	map	map	NOUN
easat-5356	208	23	.	.	PUNCT
easat-5356	209	1	this	this	DET
easat-5356	209	2	approach	approach	NOUN
easat-5356	209	3	captures	capture	VERB
easat-5356	209	4	detailed	detailed	ADJ
easat-5356	209	5	and	and	CCONJ
easat-5356	209	6	localized	localized	ADJ
easat-5356	209	7	patterns	pattern	NOUN
easat-5356	209	8	in	in	ADP
easat-5356	209	9	the	the	DET
easat-5356	209	10	input	input	NOUN
easat-5356	209	11	data	datum	NOUN
easat-5356	209	12	,	,	PUNCT
easat-5356	209	13	which	which	PRON
easat-5356	209	14	are	be	AUX
easat-5356	209	15	crucial	crucial	ADJ
easat-5356	209	16	for	for	ADP
easat-5356	209	17	accurate	accurate	ADJ
easat-5356	209	18	object	object	NOUN
easat-5356	209	19	detection	detection	NOUN
easat-5356	209	20	.	.	PUNCT
easat-5356	210	1	subsequently	subsequently	ADV
easat-5356	210	2	,	,	PUNCT
easat-5356	210	3	spatial	spatial	ADJ
easat-5356	210	4	information	information	NOUN
easat-5356	210	5	extraction	extraction	NOUN
easat-5356	210	6	leverages	leverage	NOUN
easat-5356	210	7	these	these	DET
easat-5356	210	8	enhanced	enhanced	ADJ
easat-5356	210	9	receptive	receptive	ADJ
easat-5356	210	10	field	field	NOUN
easat-5356	210	11	features	feature	NOUN
easat-5356	210	12	,	,	PUNCT
easat-5356	210	13	allowing	allow	VERB
easat-5356	210	14	the	the	DET
easat-5356	210	15	model	model	NOUN
easat-5356	210	16	to	to	PART
easat-5356	210	17	effectively	effectively	ADV
easat-5356	210	18	integrate	integrate	VERB
easat-5356	210	19	both	both	DET
easat-5356	210	20	local	local	ADJ
easat-5356	210	21	and	and	CCONJ
easat-5356	210	22	contextual	contextual	ADJ
easat-5356	210	23	information	information	NOUN
easat-5356	210	24	.	.	PUNCT
easat-5356	211	1	this	this	DET
easat-5356	211	2	enriched	enrich	VERB
easat-5356	211	3	representation	representation	NOUN
easat-5356	211	4	helps	help	VERB
easat-5356	211	5	the	the	DET
easat-5356	211	6	model	model	NOUN
easat-5356	211	7	focus	focus	VERB
easat-5356	211	8	on	on	ADP
easat-5356	211	9	important	important	ADJ
easat-5356	211	10	spatial	spatial	ADJ
easat-5356	211	11	cues	cue	NOUN
easat-5356	211	12	and	and	CCONJ
easat-5356	211	13	improves	improve	VERB
easat-5356	211	14	its	its	PRON
easat-5356	211	15	ability	ability	NOUN
easat-5356	211	16	to	to	PART
easat-5356	211	17	handle	handle	VERB
easat-5356	211	18	complex	complex	ADJ
easat-5356	211	19	visual	visual	ADJ
easat-5356	211	20	tasks	task	NOUN
easat-5356	211	21	,	,	PUNCT
easat-5356	211	22	such	such	ADJ
easat-5356	211	23	as	as	ADP
easat-5356	211	24	detecting	detect	VERB
easat-5356	211	25	faces	face	NOUN
easat-5356	211	26	at	at	ADP
easat-5356	211	27	various	various	ADJ
easat-5356	211	28	scales	scale	NOUN
easat-5356	211	29	and	and	CCONJ
easat-5356	211	30	under	under	ADP
easat-5356	211	31	different	different	ADJ
easat-5356	211	32	conditions	condition	NOUN
easat-5356	211	33	.	.	PUNCT
easat-5356	212	1	even	even	ADV
easat-5356	212	2	after	after	SCONJ
easat-5356	212	3	the	the	DET
easat-5356	212	4	gcsem	gcsem	NOUN
easat-5356	212	5	module	module	NOUN
easat-5356	212	6	is	be	AUX
easat-5356	212	7	executed	execute	VERB
easat-5356	212	8	and	and	CCONJ
easat-5356	212	9	the	the	DET
easat-5356	212	10	feature	feature	NOUN
easat-5356	212	11	map	map	NOUN
easat-5356	212	12	is	be	AUX
easat-5356	212	13	restored	restore	VERB
easat-5356	212	14	to	to	ADP
easat-5356	212	15	its	its	PRON
easat-5356	212	16	original	original	ADJ
easat-5356	212	17	size	size	NOUN
easat-5356	212	18	,	,	PUNCT
easat-5356	212	19	the	the	DET
easat-5356	212	20	expanded	expand	VERB
easat-5356	212	21	intermediate	intermediate	ADJ
easat-5356	212	22	receptive	receptive	ADJ
easat-5356	212	23	field	field	NOUN
easat-5356	212	24	feature	feature	NOUN
easat-5356	212	25	maps	map	NOUN
easat-5356	212	26	allow	allow	VERB
easat-5356	212	27	for	for	ADP
easat-5356	212	28	a	a	DET
easat-5356	212	29	full	full	ADJ
easat-5356	212	30	extraction	extraction	NOUN
easat-5356	212	31	of	of	ADP
easat-5356	212	32	the	the	DET
easat-5356	212	33	spatial	spatial	ADJ
easat-5356	212	34	information	information	NOUN
easat-5356	212	35	of	of	ADP
easat-5356	212	36	small	small	ADJ
easat-5356	212	37	objects	object	NOUN
easat-5356	212	38	.	.	PUNCT
easat-5356	213	1	the	the	DET
easat-5356	213	2	gcsem	gcsem	NOUN
easat-5356	213	3	module	module	NOUN
easat-5356	213	4	outperforms	outperform	NOUN
easat-5356	213	5	traditional	traditional	ADJ
easat-5356	213	6	down	down	ADP
easat-5356	213	7	sampling	sample	VERB
easat-5356	213	8	convolutions	convolution	NOUN
easat-5356	213	9	by	by	ADP
easat-5356	213	10	significantly	significantly	ADV
easat-5356	213	11	reducing	reduce	VERB
easat-5356	213	12	the	the	DET
easat-5356	213	13	loss	loss	NOUN
easat-5356	213	14	of	of	ADP
easat-5356	213	15	spatial	spatial	ADJ
easat-5356	213	16	information	information	NOUN
easat-5356	213	17	.	.	PUNCT
easat-5356	214	1	it	it	PRON
easat-5356	214	2	enhances	enhance	VERB
easat-5356	214	3	the	the	DET
easat-5356	214	4	fidelity	fidelity	NOUN
easat-5356	214	5	of	of	ADP
easat-5356	214	6	spatial	spatial	ADJ
easat-5356	214	7	information	information	NOUN
easat-5356	214	8	within	within	ADP
easat-5356	214	9	the	the	DET
easat-5356	214	10	feature	feature	NOUN
easat-5356	214	11	maps	map	NOUN
easat-5356	214	12	,	,	PUNCT
easat-5356	214	13	ensuring	ensure	VERB
easat-5356	214	14	that	that	SCONJ
easat-5356	214	15	fine	fine	ADV
easat-5356	214	16	-	-	PUNCT
easat-5356	214	17	grained	grain	VERB
easat-5356	214	18	details	detail	NOUN
easat-5356	214	19	,	,	PUNCT
easat-5356	214	20	particularly	particularly	ADV
easat-5356	214	21	for	for	ADP
easat-5356	214	22	small	small	ADJ
easat-5356	214	23	objects	object	NOUN
easat-5356	214	24	,	,	PUNCT
easat-5356	214	25	are	be	AUX
easat-5356	214	26	preserved	preserve	VERB
easat-5356	214	27	.	.	PUNCT
easat-5356	215	1	this	this	DET
easat-5356	215	2	improvement	improvement	NOUN
easat-5356	215	3	helps	help	VERB
easat-5356	215	4	the	the	DET
easat-5356	215	5	model	model	NOUN
easat-5356	215	6	maintain	maintain	VERB
easat-5356	215	7	accurate	accurate	ADJ
easat-5356	215	8	spatial	spatial	ADJ
easat-5356	215	9	representation	representation	NOUN
easat-5356	215	10	,	,	PUNCT
easat-5356	215	11	which	which	PRON
easat-5356	215	12	is	be	AUX
easat-5356	215	13	crucial	crucial	ADJ
easat-5356	215	14	for	for	ADP
easat-5356	215	15	detecting	detect	VERB
easat-5356	215	16	small	small	ADJ
easat-5356	215	17	and	and	CCONJ
easat-5356	215	18	challenging	challenging	ADJ
easat-5356	215	19	objects	object	NOUN
easat-5356	215	20	in	in	ADP
easat-5356	215	21	complex	complex	ADJ
easat-5356	215	22	visual	visual	ADJ
easat-5356	215	23	scenes	scene	NOUN
easat-5356	215	24	.	.	PUNCT
easat-5356	216	1	3.3	3.3	NUM
easat-5356	216	2	.	.	PUNCT
easat-5356	217	1	wiouv3	wiouv3	NOUN
easat-5356	217	2	loss	loss	NOUN
easat-5356	217	3	function	function	NOUN
easat-5356	217	4	and	and	CCONJ
easat-5356	217	5	adaptive	adaptive	ADJ
easat-5356	217	6	filtering	filtering	NOUN
easat-5356	217	7	we	we	PRON
easat-5356	217	8	present	present	VERB
easat-5356	217	9	the	the	DET
easat-5356	217	10	wiouv3	wiouv3	NOUN
easat-5356	217	11	loss	loss	NOUN
easat-5356	217	12	function	function	NOUN
easat-5356	217	13	to	to	PART
easat-5356	217	14	improve	improve	VERB
easat-5356	217	15	the	the	DET
easat-5356	217	16	bounding	bounding	NOUN
easat-5356	217	17	box	box	NOUN
easat-5356	217	18	regression	regression	NOUN
easat-5356	217	19	component	component	NOUN
easat-5356	217	20	of	of	ADP
easat-5356	217	21	yolov8	yolov8	PROPN
easat-5356	217	22	,	,	PUNCT
easat-5356	217	23	a	a	DET
easat-5356	217	24	novel	novel	ADJ
easat-5356	217	25	approach	approach	NOUN
easat-5356	217	26	designed	design	VERB
easat-5356	217	27	to	to	PART
easat-5356	217	28	overcome	overcome	VERB
easat-5356	217	29	the	the	DET
easat-5356	217	30	limitations	limitation	NOUN
easat-5356	217	31	of	of	ADP
easat-5356	217	32	traditional	traditional	ADJ
easat-5356	217	33	iou	iou	NOUN
easat-5356	217	34	-	-	PUNCT
easat-5356	217	35	based	base	VERB
easat-5356	217	36	loss	loss	NOUN
easat-5356	217	37	functions	function	NOUN
easat-5356	217	38	,	,	PUNCT
easat-5356	217	39	especially	especially	ADV
easat-5356	217	40	when	when	SCONJ
easat-5356	217	41	addressing	address	VERB
easat-5356	217	42	minor	minor	ADJ
easat-5356	217	43	inaccuracies	inaccuracy	NOUN
easat-5356	217	44	in	in	ADP
easat-5356	217	45	bounding	bound	VERB
easat-5356	217	46	box	box	NOUN
easat-5356	217	47	predictions	prediction	NOUN
easat-5356	217	48	.	.	PUNCT
easat-5356	218	1	wiouv3	wiouv3	PROPN
easat-5356	218	2	integrates	integrate	VERB
easat-5356	218	3	a	a	DET
easat-5356	218	4	dynamic	dynamic	ADJ
easat-5356	218	5	,	,	PUNCT
easat-5356	218	6	non	non	ADJ
easat-5356	218	7	-	-	ADJ
easat-5356	218	8	monotonic	monotonic	ADJ
easat-5356	218	9	focusing	focus	VERB
easat-5356	218	10	mechanism	mechanism	NOUN
easat-5356	218	11	in	in	ADP
easat-5356	218	12	contrast	contrast	NOUN
easat-5356	218	13	to	to	ADP
easat-5356	218	14	traditional	traditional	ADJ
easat-5356	218	15	loss	loss	NOUN
easat-5356	218	16	functions	function	NOUN
easat-5356	218	17	that	that	PRON
easat-5356	218	18	depend	depend	VERB
easat-5356	218	19	on	on	ADP
easat-5356	218	20	a	a	DET
easat-5356	218	21	static	static	ADJ
easat-5356	218	22	focusing	focus	VERB
easat-5356	218	23	mechanism	mechanism	NOUN
easat-5356	218	24	.	.	PUNCT
easat-5356	219	1	in	in	ADP
easat-5356	219	2	addition	addition	NOUN
easat-5356	219	3	to	to	ADP
easat-5356	219	4	considering	consider	VERB
easat-5356	219	5	the	the	DET
easat-5356	219	6	aspect	aspect	NOUN
easat-5356	219	7	ratio	ratio	NOUN
easat-5356	219	8	,	,	PUNCT
easat-5356	219	9	centroid	centroid	NOUN
easat-5356	219	10	distance	distance	NOUN
easat-5356	219	11	,	,	PUNCT
easat-5356	219	12	and	and	CCONJ
easat-5356	219	13	overlap	overlap	NOUN
easat-5356	219	14	area	area	NOUN
easat-5356	219	15	,	,	PUNCT
easat-5356	219	16	it	it	PRON
easat-5356	219	17	dynamically	dynamically	ADV
easat-5356	219	18	modifies	modify	VERB
easat-5356	219	19	the	the	DET
easat-5356	219	20	focus	focus	NOUN
easat-5356	219	21	according	accord	VERB
easat-5356	219	22	to	to	ADP
easat-5356	219	23	the	the	DET
easat-5356	219	24	prediction	prediction	NOUN
easat-5356	219	25	quality	quality	NOUN
easat-5356	219	26	.	.	PUNCT
easat-5356	220	1	the	the	DET
easat-5356	220	2	model	model	NOUN
easat-5356	220	3	's	's	PART
easat-5356	220	4	capacity	capacity	NOUN
easat-5356	220	5	to	to	PART
easat-5356	220	6	manage	manage	VERB
easat-5356	220	7	small	small	ADJ
easat-5356	220	8	-	-	PUNCT
easat-5356	220	9	scale	scale	NOUN
easat-5356	220	10	faces	face	NOUN
easat-5356	220	11	or	or	CCONJ
easat-5356	220	12	objects	object	NOUN
easat-5356	220	13	with	with	ADP
easat-5356	220	14	minute	minute	ADJ
easat-5356	220	15	localization	localization	NOUN
easat-5356	220	16	mistakes	mistake	NOUN
easat-5356	220	17	is	be	AUX
easat-5356	220	18	enhanced	enhance	VERB
easat-5356	220	19	by	by	ADP
easat-5356	220	20	this	this	DET
easat-5356	220	21	adaptive	adaptive	ADJ
easat-5356	220	22	method	method	NOUN
easat-5356	220	23	,	,	PUNCT
easat-5356	220	24	leading	lead	VERB
easat-5356	220	25	to	to	ADP
easat-5356	220	26	more	more	ADV
easat-5356	220	27	accurate	accurate	ADJ
easat-5356	220	28	bounding	bounding	NOUN
easat-5356	220	29	box	box	NOUN
easat-5356	220	30	predictions	prediction	NOUN
easat-5356	220	31	and	and	CCONJ
easat-5356	220	32	better	well	ADJ
easat-5356	220	33	overall	overall	ADJ
easat-5356	220	34	performance	performance	NOUN
easat-5356	220	35	in	in	ADP
easat-5356	220	36	challenging	challenge	VERB
easat-5356	220	37	detection	detection	NOUN
easat-5356	220	38	scenarios	scenario	NOUN
easat-5356	220	39	.	.	PUNCT
easat-5356	221	1	wiou	wiou	NOUN
easat-5356	221	2	assesses	assess	VERB
easat-5356	221	3	the	the	DET
easat-5356	221	4	anchor	anchor	NOUN
easat-5356	221	5	box	box	NOUN
easat-5356	221	6	's	's	PART
easat-5356	221	7	quality	quality	NOUN
easat-5356	221	8	using	use	VERB
easat-5356	221	9	a	a	DET
easat-5356	221	10	sensible	sensible	ADJ
easat-5356	221	11	gradient	gradient	NOUN
easat-5356	221	12	gain	gain	NOUN
easat-5356	221	13	allocation	allocation	NOUN
easat-5356	221	14	technique	technique	NOUN
easat-5356	221	15	.	.	PUNCT
easat-5356	222	1	three	three	NUM
easat-5356	222	2	iterations	iteration	NOUN
easat-5356	222	3	of	of	ADP
easat-5356	222	4	wiou	wiou	NOUN
easat-5356	222	5	were	be	AUX
easat-5356	222	6	proposed	propose	VERB
easat-5356	222	7	by	by	ADP
easat-5356	222	8	tong	tong	PROPN
easat-5356	222	9	et	et	PROPN
easat-5356	222	10	al	al	PROPN
easat-5356	222	11	.	.	PUNCT
easat-5356	223	1	[	[	X
easat-5356	223	2	19	19	NUM
easat-5356	223	3	]	]	PUNCT
easat-5356	223	4	.	.	PUNCT
easat-5356	224	1	attention	attention	NOUN
easat-5356	224	2	-	-	PUNCT
easat-5356	224	3	based	base	VERB
easat-5356	224	4	projected	project	VERB
easat-5356	224	5	box	box	NOUN
easat-5356	224	6	loss	loss	NOUN
easat-5356	224	7	was	be	AUX
easat-5356	224	8	used	use	VERB
easat-5356	224	9	in	in	ADP
easat-5356	224	10	the	the	DET
easat-5356	224	11	design	design	NOUN
easat-5356	224	12	of	of	ADP
easat-5356	224	13	wiouv1	wiouv1	PROPN
easat-5356	224	14	,	,	PUNCT
easat-5356	224	15	and	and	CCONJ
easat-5356	224	16	focusing	focus	VERB
easat-5356	224	17	coefficients	coefficient	NOUN
easat-5356	224	18	were	be	AUX
easat-5356	224	19	introduced	introduce	VERB
easat-5356	224	20	in	in	ADP
easat-5356	224	21	wiouv2	wiouv2	PROPN
easat-5356	224	22	and	and	CCONJ
easat-5356	224	23	v3	v3	PROPN
easat-5356	224	24	.	.	PUNCT
easat-5356	225	1	wiouv3	wiouv3	PROPN
easat-5356	225	2	innovatively	innovatively	ADV
easat-5356	225	3	incorporates	incorporate	VERB
easat-5356	225	4	adaptive	adaptive	ADJ
easat-5356	225	5	weights	weight	NOUN
easat-5356	225	6	,	,	PUNCT
easat-5356	225	7	denoted	denote	VERB
easat-5356	225	8	by	by	ADP
easat-5356	225	9	r	r	NOUN
easat-5356	225	10	,	,	PUNCT
easat-5356	225	11	which	which	PRON
easat-5356	225	12	are	be	AUX
easat-5356	225	13	dynamically	dynamically	ADV
easat-5356	225	14	adjusted	adjust	VERB
easat-5356	225	15	based	base	VERB
easat-5356	225	16	on	on	ADP
easat-5356	225	17	the	the	DET
easat-5356	225	18	delta	delta	NOUN
easat-5356	225	19	and	and	CCONJ
easat-5356	225	20	alpha	alpha	NOUN
easat-5356	225	21	parameters	parameter	NOUN
easat-5356	225	22	.	.	PUNCT
easat-5356	226	1	these	these	DET
easat-5356	226	2	parameters	parameter	NOUN
easat-5356	226	3	are	be	AUX
easat-5356	226	4	sensitive	sensitive	ADJ
easat-5356	226	5	to	to	ADP
easat-5356	226	6	the	the	DET
easat-5356	226	7	scale	scale	NOUN
easat-5356	226	8	of	of	ADP
easat-5356	226	9	the	the	DET
easat-5356	226	10	bounding	bounding	NOUN
easat-5356	226	11	boxes	box	NOUN
easat-5356	226	12	,	,	PUNCT
easat-5356	226	13	ensuring	ensure	VERB
easat-5356	226	14	a	a	DET
easat-5356	226	15	balanced	balanced	ADJ
easat-5356	226	16	error	error	NOUN
easat-5356	226	17	distribution	distribution	NOUN
easat-5356	226	18	across	across	ADP
easat-5356	226	19	predictions	prediction	NOUN
easat-5356	226	20	of	of	ADP
easat-5356	226	21	varying	vary	VERB
easat-5356	226	22	sizes	size	NOUN
easat-5356	226	23	.	.	PUNCT
easat-5356	227	1	the	the	DET
easat-5356	227	2	wiouv3	wiouv3	PROPN
easat-5356	227	3	loss	loss	NOUN
easat-5356	227	4	function	function	NOUN
easat-5356	227	5	is	be	AUX
easat-5356	227	6	articulated	articulate	VERB
easat-5356	227	7	as	as	SCONJ
easat-5356	227	8	follows	follow	VERB
easat-5356	227	9	:	:	PUNCT
easat-5356	227	10	𝐿𝑊𝐼𝑜𝑈𝑣3	𝐿𝑊𝐼𝑜𝑈𝑣3	SYM
easat-5356	228	1	=	=	SYM
easat-5356	228	2	𝑟	𝑟	NOUN
easat-5356	228	3	×	×	NOUN
easat-5356	228	4	𝐿𝑊𝐼𝑜𝑈𝑣1	𝐿𝑊𝐼𝑜𝑈𝑣1	PUNCT
easat-5356	228	5	(	(	PUNCT
easat-5356	228	6	9	9	NUM
easat-5356	228	7	)	)	PUNCT
easat-5356	228	8	𝑟	𝑟	NOUN
easat-5356	228	9	=	=	SYM
easat-5356	228	10	𝛽	𝛽	NOUN
easat-5356	228	11	𝛿𝛼𝛽−𝛿	𝛿𝛼𝛽−𝛿	X
easat-5356	228	12	(	(	PUNCT
easat-5356	228	13	10	10	NUM
easat-5356	228	14	)	)	PUNCT
easat-5356	228	15	𝛽	𝛽	NOUN
easat-5356	228	16	=	=	SYM
easat-5356	228	17	𝐿∗	𝐿∗	PROPN
easat-5356	228	18	𝐼𝑜𝑈	𝐼𝑜𝑈	PROPN
easat-5356	228	19	𝐿𝐼𝑜𝑈	𝐿𝐼𝑜𝑈	PROPN
easat-5356	228	20	−	−	PROPN
easat-5356	228	21	∈	∈	PROPN
easat-5356	229	1	[	[	X
easat-5356	229	2	0	0	NUM
easat-5356	229	3	,	,	PUNCT
easat-5356	229	4	+	+	NOUN
easat-5356	229	5	∞	∞	NUM
easat-5356	229	6	)	)	PUNCT
easat-5356	229	7	(	(	PUNCT
easat-5356	229	8	11	11	NUM
easat-5356	229	9	)	)	PUNCT
easat-5356	229	10	where	where	SCONJ
easat-5356	229	11	is	be	AUX
easat-5356	229	12	calculated	calculate	VERB
easat-5356	229	13	based	base	VERB
easat-5356	229	14	on	on	ADP
easat-5356	229	15	the	the	DET
easat-5356	229	16	iou	iou	PROPN
easat-5356	229	17	target	target	NOUN
easat-5356	229	18	values	value	NOUN
easat-5356	229	19	and	and	CCONJ
easat-5356	229	20	mean	mean	ADJ
easat-5356	229	21	iou	iou	NOUN
easat-5356	229	22	,	,	PUNCT
easat-5356	229	23	ensuring	ensure	VERB
easat-5356	229	24	smooth	smooth	ADJ
easat-5356	229	25	adjustments	adjustment	NOUN
easat-5356	229	26	to	to	ADP
easat-5356	229	27	loss	loss	NOUN
easat-5356	229	28	gradients	gradient	NOUN
easat-5356	229	29	.	.	PUNCT
easat-5356	230	1	and	and	CCONJ
easat-5356	230	2	are	be	AUX
easat-5356	230	3	hyperparameters	hyperparameter	NOUN
easat-5356	230	4	in	in	ADP
easat-5356	230	5	the	the	DET
easat-5356	230	6	equation	equation	NOUN
easat-5356	230	7	that	that	PRON
easat-5356	230	8	can	can	AUX
easat-5356	230	9	be	be	AUX
easat-5356	230	10	changed	change	VERB
easat-5356	230	11	to	to	PART
easat-5356	230	12	suit	suit	VERB
easat-5356	230	13	various	various	ADJ
easat-5356	230	14	models	model	NOUN
easat-5356	230	15	.	.	PUNCT
easat-5356	231	1	848	848	NUM
easat-5356	231	2	edelweiss	edelweiss	PROPN
easat-5356	231	3	applied	apply	VERB
easat-5356	231	4	science	science	NOUN
easat-5356	231	5	and	and	CCONJ
easat-5356	231	6	technology	technology	NOUN
easat-5356	231	7	issn	issn	PROPN
easat-5356	231	8	:	:	PUNCT
easat-5356	231	9	2576	2576	NUM
easat-5356	231	10	-	-	SYM
easat-5356	231	11	8484	8484	NUM
easat-5356	231	12	vol	vol	NOUN
easat-5356	231	13	.	.	PROPN
easat-5356	232	1	9	9	NUM
easat-5356	232	2	,	,	PUNCT
easat-5356	232	3	no	no	INTJ
easat-5356	232	4	.	.	NOUN
easat-5356	233	1	3	3	NUM
easat-5356	233	2	:	:	PUNCT
easat-5356	233	3	840	840	NUM
easat-5356	233	4	-	-	SYM
easat-5356	233	5	855	855	NUM
easat-5356	233	6	,	,	PUNCT
easat-5356	233	7	2025	2025	NUM
easat-5356	233	8	doi	doi	NOUN
easat-5356	233	9	:	:	PUNCT
easat-5356	233	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	233	11	©	©	PROPN
easat-5356	233	12	2025	2025	NUM
easat-5356	233	13	by	by	ADP
easat-5356	233	14	the	the	DET
easat-5356	233	15	authors	author	NOUN
easat-5356	233	16	;	;	PUNCT
easat-5356	233	17	licensee	licensee	PROPN
easat-5356	233	18	learning	learn	VERB
easat-5356	233	19	gate	gate	PROPN
easat-5356	233	20	building	building	NOUN
easat-5356	233	21	upon	upon	SCONJ
easat-5356	233	22	wiouv3	wiouv3	NOUN
easat-5356	233	23	,	,	PUNCT
easat-5356	233	24	we	we	PRON
easat-5356	233	25	introduce	introduce	VERB
easat-5356	233	26	an	an	DET
easat-5356	233	27	adaptive	adaptive	ADJ
easat-5356	233	28	filter	filter	NOUN
easat-5356	233	29	that	that	PRON
easat-5356	233	30	effectively	effectively	ADV
easat-5356	233	31	filters	filter	VERB
easat-5356	233	32	out	out	ADP
easat-5356	233	33	most	most	ADV
easat-5356	233	34	easily	easily	ADV
easat-5356	233	35	distinguishable	distinguishable	ADJ
easat-5356	233	36	negative	negative	ADJ
easat-5356	233	37	samples	sample	NOUN
easat-5356	233	38	,	,	PUNCT
easat-5356	233	39	thereby	thereby	ADV
easat-5356	233	40	reducing	reduce	VERB
easat-5356	233	41	the	the	DET
easat-5356	233	42	number	number	NOUN
easat-5356	233	43	of	of	ADP
easat-5356	233	44	samples	sample	NOUN
easat-5356	233	45	the	the	DET
easat-5356	233	46	classifier	classifier	NOUN
easat-5356	233	47	must	must	AUX
easat-5356	233	48	process	process	VERB
easat-5356	233	49	.	.	PUNCT
easat-5356	234	1	this	this	DET
easat-5356	234	2	operation	operation	NOUN
easat-5356	234	3	addresses	address	VERB
easat-5356	234	4	a	a	DET
easat-5356	234	5	key	key	ADJ
easat-5356	234	6	challenge	challenge	NOUN
easat-5356	234	7	:	:	PUNCT
easat-5356	234	8	when	when	SCONJ
easat-5356	234	9	detecting	detect	VERB
easat-5356	234	10	small	small	ADJ
easat-5356	234	11	-	-	PUNCT
easat-5356	234	12	scale	scale	NOUN
easat-5356	234	13	faces	face	NOUN
easat-5356	234	14	,	,	PUNCT
easat-5356	234	15	anchors	anchor	NOUN
easat-5356	234	16	are	be	AUX
easat-5356	234	17	densely	densely	ADV
easat-5356	234	18	sampled	sample	VERB
easat-5356	234	19	from	from	ADP
easat-5356	234	20	the	the	DET
easat-5356	234	21	shallow	shallow	ADJ
easat-5356	234	22	feature	feature	NOUN
easat-5356	234	23	maps	map	NOUN
easat-5356	234	24	of	of	ADP
easat-5356	234	25	the	the	DET
easat-5356	234	26	feature	feature	NOUN
easat-5356	234	27	pyramid	pyramid	NOUN
easat-5356	234	28	.	.	PUNCT
easat-5356	235	1	while	while	SCONJ
easat-5356	235	2	this	this	PRON
easat-5356	235	3	leads	lead	VERB
easat-5356	235	4	to	to	ADP
easat-5356	235	5	good	good	ADJ
easat-5356	235	6	recall	recall	NOUN
easat-5356	235	7	for	for	ADP
easat-5356	235	8	small	small	ADJ
easat-5356	235	9	-	-	PUNCT
easat-5356	235	10	scale	scale	NOUN
easat-5356	235	11	faces	face	NOUN
easat-5356	235	12	,	,	PUNCT
easat-5356	235	13	it	it	PRON
easat-5356	235	14	also	also	ADV
easat-5356	235	15	adds	add	VERB
easat-5356	235	16	a	a	DET
easat-5356	235	17	large	large	ADJ
easat-5356	235	18	number	number	NOUN
easat-5356	235	19	of	of	ADP
easat-5356	235	20	false	false	ADJ
easat-5356	235	21	positives	positive	NOUN
easat-5356	235	22	,	,	PUNCT
easat-5356	235	23	which	which	PRON
easat-5356	235	24	exacerbates	exacerbate	VERB
easat-5356	235	25	the	the	DET
easat-5356	235	26	imbalance	imbalance	NOUN
easat-5356	235	27	between	between	ADP
easat-5356	235	28	positive	positive	ADJ
easat-5356	235	29	and	and	CCONJ
easat-5356	235	30	negative	negative	ADJ
easat-5356	235	31	samples	sample	NOUN
easat-5356	235	32	.	.	PUNCT
easat-5356	236	1	our	our	PRON
easat-5356	236	2	adaptive	adaptive	ADJ
easat-5356	236	3	filter	filter	NOUN
easat-5356	236	4	greatly	greatly	ADV
easat-5356	236	5	lessens	lessen	VERB
easat-5356	236	6	the	the	DET
easat-5356	236	7	model	model	NOUN
easat-5356	236	8	's	's	PART
easat-5356	236	9	computational	computational	ADJ
easat-5356	236	10	load	load	NOUN
easat-5356	236	11	by	by	ADP
easat-5356	236	12	automatically	automatically	ADV
easat-5356	236	13	eliminating	eliminate	VERB
easat-5356	236	14	the	the	DET
easat-5356	236	15	majority	majority	NOUN
easat-5356	236	16	of	of	ADP
easat-5356	236	17	easily	easily	ADV
easat-5356	236	18	separable	separable	ADJ
easat-5356	236	19	negative	negative	ADJ
easat-5356	236	20	samples	sample	NOUN
easat-5356	236	21	.	.	PUNCT
easat-5356	237	1	the	the	DET
easat-5356	237	2	atw	atw	PROPN
easat-5356	237	3	module	module	NOUN
easat-5356	237	4	takes	take	VERB
easat-5356	237	5	the	the	DET
easat-5356	237	6	coordinates	coordinate	NOUN
easat-5356	237	7	and	and	CCONJ
easat-5356	237	8	scores	score	NOUN
easat-5356	237	9	of	of	ADP
easat-5356	237	10	anchor	anchor	NOUN
easat-5356	237	11	boxes	box	NOUN
easat-5356	237	12	as	as	ADP
easat-5356	237	13	inputs	input	NOUN
easat-5356	237	14	,	,	PUNCT
easat-5356	237	15	using	use	VERB
easat-5356	237	16	a	a	DET
easat-5356	237	17	percentilebased	percentilebased	ADJ
easat-5356	237	18	approach	approach	NOUN
easat-5356	237	19	to	to	PART
easat-5356	237	20	determine	determine	VERB
easat-5356	237	21	the	the	DET
easat-5356	237	22	filtering	filter	VERB
easat-5356	237	23	threshold	threshold	NOUN
easat-5356	237	24	.	.	PUNCT
easat-5356	238	1	the	the	DET
easat-5356	238	2	threshold	threshold	NOUN
easat-5356	238	3	is	be	AUX
easat-5356	238	4	dynamically	dynamically	ADV
easat-5356	238	5	calculated	calculate	VERB
easat-5356	238	6	based	base	VERB
easat-5356	238	7	on	on	ADP
easat-5356	238	8	the	the	DET
easat-5356	238	9	score	score	NOUN
easat-5356	238	10	distribution	distribution	NOUN
easat-5356	238	11	of	of	ADP
easat-5356	238	12	the	the	DET
easat-5356	238	13	anchors	anchor	NOUN
easat-5356	238	14	.	.	PUNCT
easat-5356	239	1	for	for	ADP
easat-5356	239	2	instance	instance	NOUN
easat-5356	239	3	,	,	PUNCT
easat-5356	239	4	if	if	SCONJ
easat-5356	239	5	the	the	DET
easat-5356	239	6	percentile	percentile	NOUN
easat-5356	239	7	is	be	AUX
easat-5356	239	8	set	set	VERB
easat-5356	239	9	to	to	ADP
easat-5356	239	10	95	95	NUM
easat-5356	239	11	,	,	PUNCT
easat-5356	239	12	the	the	DET
easat-5356	239	13	threshold	threshold	NOUN
easat-5356	239	14	corresponds	correspond	VERB
easat-5356	239	15	to	to	ADP
easat-5356	239	16	the	the	DET
easat-5356	239	17	95th	95th	ADJ
easat-5356	239	18	percentile	percentile	NOUN
easat-5356	239	19	of	of	ADP
easat-5356	239	20	the	the	DET
easat-5356	239	21	score	score	NOUN
easat-5356	239	22	distribution	distribution	NOUN
easat-5356	239	23	.	.	PUNCT
easat-5356	240	1	this	this	DET
easat-5356	240	2	dynamic	dynamic	ADJ
easat-5356	240	3	adjustment	adjustment	NOUN
easat-5356	240	4	allows	allow	VERB
easat-5356	240	5	the	the	DET
easat-5356	240	6	threshold	threshold	NOUN
easat-5356	240	7	to	to	PART
easat-5356	240	8	automatically	automatically	ADV
easat-5356	240	9	adapt	adapt	VERB
easat-5356	240	10	to	to	ADP
easat-5356	240	11	the	the	DET
easat-5356	240	12	score	score	NOUN
easat-5356	240	13	distribution	distribution	NOUN
easat-5356	240	14	,	,	PUNCT
easat-5356	240	15	ensuring	ensure	VERB
easat-5356	240	16	that	that	SCONJ
easat-5356	240	17	it	it	PRON
easat-5356	240	18	is	be	AUX
easat-5356	240	19	optimized	optimize	VERB
easat-5356	240	20	for	for	ADP
easat-5356	240	21	different	different	ADJ
easat-5356	240	22	datasets	dataset	NOUN
easat-5356	240	23	and	and	CCONJ
easat-5356	240	24	tasks	task	NOUN
easat-5356	240	25	,	,	PUNCT
easat-5356	240	26	ultimately	ultimately	ADV
easat-5356	240	27	improving	improve	VERB
easat-5356	240	28	the	the	DET
easat-5356	240	29	efficiency	efficiency	NOUN
easat-5356	240	30	and	and	CCONJ
easat-5356	240	31	effectiveness	effectiveness	NOUN
easat-5356	240	32	of	of	ADP
easat-5356	240	33	the	the	DET
easat-5356	240	34	model	model	NOUN
easat-5356	240	35	.	.	PUNCT
easat-5356	241	1	3.4	3.4	NUM
easat-5356	241	2	small	small	ADJ
easat-5356	241	3	-	-	PUNCT
easat-5356	241	4	scale	scale	NOUN
easat-5356	241	5	detection	detection	NOUN
easat-5356	241	6	head	head	NOUN
easat-5356	241	7	to	to	PART
easat-5356	241	8	tackle	tackle	VERB
easat-5356	241	9	the	the	DET
easat-5356	241	10	detection	detection	NOUN
easat-5356	241	11	of	of	ADP
easat-5356	241	12	small	small	ADJ
easat-5356	241	13	-	-	PUNCT
easat-5356	241	14	scale	scale	NOUN
easat-5356	241	15	targets	target	NOUN
easat-5356	241	16	while	while	SCONJ
easat-5356	241	17	considering	consider	VERB
easat-5356	241	18	the	the	DET
easat-5356	241	19	platform	platform	NOUN
easat-5356	241	20	's	's	PART
easat-5356	241	21	resource	resource	NOUN
easat-5356	241	22	limitations	limitation	NOUN
easat-5356	241	23	,	,	PUNCT
easat-5356	241	24	this	this	DET
easat-5356	241	25	paper	paper	NOUN
easat-5356	241	26	strategically	strategically	ADV
easat-5356	241	27	integrates	integrate	VERB
easat-5356	241	28	the	the	DET
easat-5356	241	29	efficient	efficient	ADJ
easat-5356	241	30	gcsem	gcsem	NOUN
easat-5356	241	31	module	module	NOUN
easat-5356	241	32	to	to	PART
easat-5356	241	33	design	design	VERB
easat-5356	241	34	the	the	DET
easat-5356	241	35	feature	feature	NOUN
easat-5356	241	36	fusion	fusion	NOUN
easat-5356	241	37	network	network	NOUN
easat-5356	241	38	.	.	PUNCT
easat-5356	242	1	an	an	DET
easat-5356	242	2	additional	additional	ADJ
easat-5356	242	3	detection	detection	NOUN
easat-5356	242	4	scale	scale	NOUN
easat-5356	242	5	is	be	AUX
easat-5356	242	6	introduced	introduce	VERB
easat-5356	242	7	to	to	PART
easat-5356	242	8	complement	complement	VERB
easat-5356	242	9	yolov8	yolov8	PROPN
easat-5356	242	10	's	's	PART
easat-5356	242	11	original	original	ADJ
easat-5356	242	12	three	three	NUM
easat-5356	242	13	,	,	PUNCT
easat-5356	242	14	enhancing	enhance	VERB
easat-5356	242	15	the	the	DET
easat-5356	242	16	fusion	fusion	NOUN
easat-5356	242	17	of	of	ADP
easat-5356	242	18	shallow	shallow	ADJ
easat-5356	242	19	features	feature	NOUN
easat-5356	242	20	with	with	ADP
easat-5356	242	21	more	more	ADV
easat-5356	242	22	comprehensive	comprehensive	ADJ
easat-5356	242	23	positional	positional	ADJ
easat-5356	242	24	information	information	NOUN
easat-5356	242	25	.	.	PUNCT
easat-5356	243	1	the	the	DET
easat-5356	243	2	gcsem	gcsem	NOUN
easat-5356	243	3	module	module	NOUN
easat-5356	243	4	is	be	AUX
easat-5356	243	5	strategically	strategically	ADV
easat-5356	243	6	inserted	insert	VERB
easat-5356	243	7	as	as	ADP
easat-5356	243	8	a	a	DET
easat-5356	243	9	feature	feature	NOUN
easat-5356	243	10	processing	processing	NOUN
easat-5356	243	11	block	block	NOUN
easat-5356	243	12	the	the	DET
easat-5356	243	13	backbone	backbone	NOUN
easat-5356	243	14	network	network	NOUN
easat-5356	243	15	.	.	PUNCT
easat-5356	244	1	as	as	SCONJ
easat-5356	244	2	can	can	AUX
easat-5356	244	3	be	be	AUX
easat-5356	244	4	seen	see	VERB
easat-5356	244	5	in	in	ADP
easat-5356	244	6	figure	figure	NOUN
easat-5356	244	7	2	2	NUM
easat-5356	244	8	,	,	PUNCT
easat-5356	244	9	the	the	DET
easat-5356	244	10	incorporation	incorporation	NOUN
easat-5356	244	11	of	of	ADP
easat-5356	244	12	the	the	DET
easat-5356	244	13	gcsem	gcsem	NOUN
easat-5356	244	14	module	module	NOUN
easat-5356	244	15	serves	serve	VERB
easat-5356	244	16	to	to	PART
easat-5356	244	17	mitigate	mitigate	VERB
easat-5356	244	18	the	the	DET
easat-5356	244	19	resource	resource	NOUN
easat-5356	244	20	demands	demand	NOUN
easat-5356	244	21	associated	associate	VERB
easat-5356	244	22	with	with	ADP
easat-5356	244	23	multi	multi	ADJ
easat-5356	244	24	-	-	ADJ
easat-5356	244	25	scale	scale	ADJ
easat-5356	244	26	feature	feature	NOUN
easat-5356	244	27	fusion	fusion	NOUN
easat-5356	244	28	.	.	PUNCT
easat-5356	245	1	the	the	DET
easat-5356	245	2	resultant	resultant	NOUN
easat-5356	245	3	model	model	NOUN
easat-5356	245	4	,	,	PUNCT
easat-5356	245	5	now	now	ADV
easat-5356	245	6	capable	capable	ADJ
easat-5356	245	7	of	of	ADP
easat-5356	245	8	four	four	NUM
easat-5356	245	9	-	-	PUNCT
easat-5356	245	10	scale	scale	NOUN
easat-5356	245	11	detection	detection	NOUN
easat-5356	245	12	,	,	PUNCT
easat-5356	245	13	witnesses	witness	VERB
easat-5356	245	14	a	a	DET
easat-5356	245	15	marked	mark	VERB
easat-5356	245	16	enhancement	enhancement	NOUN
easat-5356	245	17	in	in	ADP
easat-5356	245	18	detection	detection	NOUN
easat-5356	245	19	performance	performance	NOUN
easat-5356	245	20	,	,	PUNCT
easat-5356	245	21	particularly	particularly	ADV
easat-5356	245	22	for	for	ADP
easat-5356	245	23	small	small	ADJ
easat-5356	245	24	-	-	PUNCT
easat-5356	245	25	scale	scale	NOUN
easat-5356	245	26	targets	target	NOUN
easat-5356	245	27	,	,	PUNCT
easat-5356	245	28	thus	thus	ADV
easat-5356	245	29	pushing	push	VERB
easat-5356	245	30	the	the	DET
easat-5356	245	31	boundaries	boundary	NOUN
easat-5356	245	32	of	of	ADP
easat-5356	245	33	what	what	PRON
easat-5356	245	34	is	be	AUX
easat-5356	245	35	achievable	achievable	ADJ
easat-5356	245	36	within	within	ADP
easat-5356	245	37	the	the	DET
easat-5356	245	38	constraints	constraint	NOUN
easat-5356	245	39	of	of	ADP
easat-5356	245	40	limited	limited	ADJ
easat-5356	245	41	computational	computational	ADJ
easat-5356	245	42	resources	resource	NOUN
easat-5356	245	43	.	.	PUNCT
easat-5356	246	1	figure	figure	NOUN
easat-5356	246	2	2	2	NUM
easat-5356	246	3	.	.	PUNCT
easat-5356	247	1	the	the	DET
easat-5356	247	2	terms	term	NOUN
easat-5356	247	3	large	large	ADJ
easat-5356	247	4	,	,	PUNCT
easat-5356	247	5	medium	medium	ADJ
easat-5356	247	6	,	,	PUNCT
easat-5356	247	7	tiny	tiny	ADJ
easat-5356	247	8	,	,	PUNCT
easat-5356	247	9	and	and	CCONJ
easat-5356	247	10	xsmall	xsmall	NOUN
easat-5356	247	11	are	be	AUX
easat-5356	247	12	used	use	VERB
easat-5356	247	13	to	to	PART
easat-5356	247	14	indicate	indicate	VERB
easat-5356	247	15	an	an	DET
easat-5356	247	16	object	object	NOUN
easat-5356	247	17	's	's	PART
easat-5356	247	18	size	size	NOUN
easat-5356	247	19	.	.	PUNCT
easat-5356	248	1	849	849	NUM
easat-5356	248	2	edelweiss	edelweiss	PROPN
easat-5356	248	3	applied	apply	VERB
easat-5356	248	4	science	science	NOUN
easat-5356	248	5	and	and	CCONJ
easat-5356	248	6	technology	technology	NOUN
easat-5356	248	7	issn	issn	PROPN
easat-5356	248	8	:	:	PUNCT
easat-5356	248	9	2576	2576	NUM
easat-5356	248	10	-	-	SYM
easat-5356	248	11	8484	8484	NUM
easat-5356	248	12	vol	vol	NOUN
easat-5356	248	13	.	.	PROPN
easat-5356	249	1	9	9	NUM
easat-5356	249	2	,	,	PUNCT
easat-5356	249	3	no	no	INTJ
easat-5356	249	4	.	.	NOUN
easat-5356	250	1	3	3	NUM
easat-5356	250	2	:	:	PUNCT
easat-5356	250	3	840	840	NUM
easat-5356	250	4	-	-	SYM
easat-5356	250	5	855	855	NUM
easat-5356	250	6	,	,	PUNCT
easat-5356	250	7	2025	2025	NUM
easat-5356	250	8	doi	doi	NOUN
easat-5356	250	9	:	:	PUNCT
easat-5356	250	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	250	11	©	©	PROPN
easat-5356	250	12	2025	2025	NUM
easat-5356	250	13	by	by	ADP
easat-5356	250	14	the	the	DET
easat-5356	250	15	authors	author	NOUN
easat-5356	250	16	;	;	PUNCT
easat-5356	250	17	licensee	licensee	PROPN
easat-5356	250	18	learning	learning	NOUN
easat-5356	250	19	gate	gate	NOUN
easat-5356	250	20	4	4	NUM
easat-5356	250	21	.	.	PUNCT
easat-5356	250	22	experiment	experiment	NOUN
easat-5356	250	23	4.1	4.1	NUM
easat-5356	250	24	.	.	PUNCT
easat-5356	251	1	wider	wide	ADJ
easat-5356	251	2	face	face	VERB
easat-5356	251	3	datasets	dataset	NOUN
easat-5356	251	4	with	with	ADP
easat-5356	251	5	32,203	32,203	NUM
easat-5356	251	6	photos	photo	NOUN
easat-5356	251	7	and	and	CCONJ
easat-5356	251	8	393,703	393,703	NUM
easat-5356	251	9	tagged	tag	VERB
easat-5356	251	10	faces—158,989	faces—158,989	PROPN
easat-5356	251	11	of	of	ADP
easat-5356	251	12	which	which	PRON
easat-5356	251	13	are	be	AUX
easat-5356	251	14	in	in	ADP
easat-5356	251	15	the	the	DET
easat-5356	251	16	training	training	NOUN
easat-5356	251	17	set	set	NOUN
easat-5356	251	18	and	and	CCONJ
easat-5356	251	19	39,496	39,496	NUM
easat-5356	251	20	in	in	ADP
easat-5356	251	21	the	the	DET
easat-5356	251	22	validation	validation	NOUN
easat-5356	251	23	set	set	NOUN
easat-5356	251	24	—	—	PUNCT
easat-5356	251	25	the	the	DET
easat-5356	251	26	wider	wide	ADJ
easat-5356	251	27	face	face	NOUN
easat-5356	251	28	datasets	dataset	NOUN
easat-5356	251	29	serve	serve	VERB
easat-5356	251	30	as	as	ADP
easat-5356	251	31	a	a	DET
easat-5356	251	32	standard	standard	NOUN
easat-5356	251	33	for	for	ADP
easat-5356	251	34	face	face	NOUN
easat-5356	251	35	detection	detection	NOUN
easat-5356	251	36	.	.	PUNCT
easat-5356	252	1	there	there	PRON
easat-5356	252	2	are	be	VERB
easat-5356	252	3	three	three	NUM
easat-5356	252	4	detection	detection	NOUN
easat-5356	252	5	complexity	complexity	NOUN
easat-5356	252	6	levels	level	NOUN
easat-5356	252	7	in	in	ADP
easat-5356	252	8	each	each	DET
easat-5356	252	9	subset	subset	NOUN
easat-5356	252	10	:	:	PUNCT
easat-5356	252	11	easy	easy	ADJ
easat-5356	252	12	,	,	PUNCT
easat-5356	252	13	medium	medium	ADJ
easat-5356	252	14	,	,	PUNCT
easat-5356	252	15	and	and	CCONJ
easat-5356	252	16	hard	hard	ADV
easat-5356	252	17	.	.	PUNCT
easat-5356	253	1	as	as	SCONJ
easat-5356	253	2	can	can	AUX
easat-5356	253	3	be	be	AUX
easat-5356	253	4	seen	see	VERB
easat-5356	253	5	in	in	ADP
easat-5356	253	6	figure	figure	NOUN
easat-5356	253	7	3	3	NUM
easat-5356	253	8	,	,	PUNCT
easat-5356	253	9	the	the	DET
easat-5356	253	10	scale	scale	NOUN
easat-5356	253	11	,	,	PUNCT
easat-5356	253	12	placement	placement	NOUN
easat-5356	253	13	,	,	PUNCT
easat-5356	253	14	lighting	lighting	NOUN
easat-5356	253	15	,	,	PUNCT
easat-5356	253	16	expression	expression	NOUN
easat-5356	253	17	,	,	PUNCT
easat-5356	253	18	and	and	CCONJ
easat-5356	253	19	obstruction	obstruction	NOUN
easat-5356	253	20	of	of	ADP
easat-5356	253	21	these	these	DET
easat-5356	253	22	faces	face	NOUN
easat-5356	253	23	differ	differ	VERB
easat-5356	253	24	significantly	significantly	ADV
easat-5356	253	25	.	.	PUNCT
easat-5356	254	1	figure	figure	VERB
easat-5356	254	2	3	3	NUM
easat-5356	254	3	.	.	PUNCT
easat-5356	255	1	examples	example	NOUN
easat-5356	255	2	of	of	ADP
easat-5356	255	3	the	the	DET
easat-5356	255	4	wider	wide	ADJ
easat-5356	255	5	face	face	NOUN
easat-5356	255	6	dataset	dataset	NOUN
easat-5356	255	7	.	.	PUNCT
easat-5356	256	1	these	these	DET
easat-5356	256	2	public	public	ADJ
easat-5356	256	3	datasets	dataset	NOUN
easat-5356	256	4	wider	wide	ADJ
easat-5356	256	5	is	be	AUX
easat-5356	256	6	the	the	DET
easat-5356	256	7	primary	primary	ADJ
easat-5356	256	8	source	source	NOUN
easat-5356	256	9	of	of	ADP
easat-5356	256	10	the	the	DET
easat-5356	256	11	photographs	photograph	NOUN
easat-5356	256	12	chosen	choose	VERB
easat-5356	256	13	for	for	ADP
easat-5356	256	14	wider	wide	ADJ
easat-5356	256	15	face	face	NOUN
easat-5356	256	16	.	.	PUNCT
easat-5356	257	1	the	the	DET
easat-5356	257	2	61	61	NUM
easat-5356	257	3	wider	wide	ADJ
easat-5356	257	4	event	event	NOUN
easat-5356	257	5	categories	category	NOUN
easat-5356	257	6	were	be	AUX
easat-5356	257	7	chosen	choose	VERB
easat-5356	257	8	by	by	ADP
easat-5356	257	9	the	the	DET
easat-5356	257	10	creators	creator	NOUN
easat-5356	257	11	,	,	PUNCT
easat-5356	257	12	who	who	PRON
easat-5356	257	13	are	be	AUX
easat-5356	257	14	from	from	ADP
easat-5356	257	15	the	the	DET
easat-5356	257	16	chinese	chinese	ADJ
easat-5356	257	17	university	university	PROPN
easat-5356	257	18	of	of	ADP
easat-5356	257	19	hong	hong	PROPN
easat-5356	257	20	kong	kong	PROPN
easat-5356	257	21	.	.	PUNCT
easat-5356	258	1	for	for	ADP
easat-5356	258	2	each	each	DET
easat-5356	258	3	category	category	NOUN
easat-5356	258	4	,	,	PUNCT
easat-5356	258	5	40	40	NUM
easat-5356	258	6	%	%	NOUN
easat-5356	258	7	,	,	PUNCT
easat-5356	258	8	10	10	NUM
easat-5356	258	9	%	%	NOUN
easat-5356	258	10	,	,	PUNCT
easat-5356	258	11	and	and	CCONJ
easat-5356	258	12	50	50	NUM
easat-5356	258	13	%	%	NOUN
easat-5356	258	14	were	be	AUX
easat-5356	258	15	chosen	choose	VERB
easat-5356	258	16	at	at	ADP
easat-5356	258	17	random	random	ADJ
easat-5356	258	18	to	to	PART
easat-5356	258	19	serve	serve	VERB
easat-5356	258	20	as	as	ADP
easat-5356	258	21	the	the	DET
easat-5356	258	22	training	training	NOUN
easat-5356	258	23	,	,	PUNCT
easat-5356	258	24	validation	validation	NOUN
easat-5356	258	25	,	,	PUNCT
easat-5356	258	26	and	and	CCONJ
easat-5356	258	27	testing	testing	NOUN
easat-5356	258	28	sets	set	NOUN
easat-5356	258	29	.	.	PUNCT
easat-5356	259	1	850	850	NUM
easat-5356	259	2	edelweiss	edelweiss	PROPN
easat-5356	259	3	applied	apply	VERB
easat-5356	259	4	science	science	NOUN
easat-5356	259	5	and	and	CCONJ
easat-5356	259	6	technology	technology	NOUN
easat-5356	259	7	issn	issn	PROPN
easat-5356	259	8	:	:	PUNCT
easat-5356	259	9	2576	2576	NUM
easat-5356	259	10	-	-	SYM
easat-5356	259	11	8484	8484	NUM
easat-5356	259	12	vol	vol	NOUN
easat-5356	259	13	.	.	PROPN
easat-5356	260	1	9	9	NUM
easat-5356	260	2	,	,	PUNCT
easat-5356	260	3	no	no	INTJ
easat-5356	260	4	.	.	NOUN
easat-5356	261	1	3	3	NUM
easat-5356	261	2	:	:	PUNCT
easat-5356	261	3	840	840	NUM
easat-5356	261	4	-	-	SYM
easat-5356	261	5	855	855	NUM
easat-5356	261	6	,	,	PUNCT
easat-5356	261	7	2025	2025	NUM
easat-5356	261	8	doi	doi	NOUN
easat-5356	261	9	:	:	PUNCT
easat-5356	261	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	261	11	©	©	PROPN
easat-5356	261	12	2025	2025	NUM
easat-5356	261	13	by	by	ADP
easat-5356	261	14	the	the	DET
easat-5356	261	15	authors	author	NOUN
easat-5356	261	16	;	;	PUNCT
easat-5356	261	17	licensee	licensee	PROPN
easat-5356	261	18	learning	learn	VERB
easat-5356	261	19	gate	gate	NOUN
easat-5356	261	20	figure	figure	NOUN
easat-5356	261	21	4	4	NUM
easat-5356	261	22	.	.	PUNCT
easat-5356	261	23	manually	manually	ADV
easat-5356	261	24	labelled	label	VERB
easat-5356	261	25	object	object	NOUN
easat-5356	261	26	details	detail	NOUN
easat-5356	261	27	in	in	ADP
easat-5356	261	28	the	the	DET
easat-5356	261	29	wider	wide	ADJ
easat-5356	261	30	face	face	NOUN
easat-5356	261	31	dataset	dataset	VERB
easat-5356	261	32	python	python	NOUN
easat-5356	261	33	3.8	3.8	NUM
easat-5356	261	34	,	,	PUNCT
easat-5356	261	35	pytorch	pytorch	NOUN
easat-5356	261	36	2.1	2.1	NUM
easat-5356	261	37	,	,	PUNCT
easat-5356	261	38	and	and	CCONJ
easat-5356	261	39	cuda	cuda	PROPN
easat-5356	261	40	11.8	11.8	NUM
easat-5356	261	41	were	be	AUX
easat-5356	261	42	used	use	VERB
easat-5356	261	43	in	in	ADP
easat-5356	261	44	the	the	DET
easat-5356	261	45	desktop	desktop	NOUN
easat-5356	261	46	computing	compute	VERB
easat-5356	261	47	environment	environment	NOUN
easat-5356	261	48	for	for	ADP
easat-5356	261	49	the	the	DET
easat-5356	261	50	trials	trial	NOUN
easat-5356	261	51	,	,	PUNCT
easat-5356	261	52	which	which	PRON
easat-5356	261	53	were	be	AUX
easat-5356	261	54	conducted	conduct	VERB
easat-5356	261	55	on	on	ADP
easat-5356	261	56	a	a	DET
easat-5356	261	57	windows	window	NOUN
easat-5356	261	58	11	11	NUM
easat-5356	261	59	operating	operating	NOUN
easat-5356	261	60	system	system	NOUN
easat-5356	261	61	.	.	PUNCT
easat-5356	262	1	an	an	DET
easat-5356	262	2	nvidia	nvidia	PROPN
easat-5356	262	3	3090	3090	NUM
easat-5356	262	4	graphics	graphic	NOUN
easat-5356	262	5	card	card	NOUN
easat-5356	262	6	was	be	AUX
easat-5356	262	7	used	use	VERB
easat-5356	262	8	as	as	ADP
easat-5356	262	9	the	the	DET
easat-5356	262	10	hardware	hardware	NOUN
easat-5356	262	11	for	for	ADP
easat-5356	262	12	these	these	DET
easat-5356	262	13	tests	test	NOUN
easat-5356	262	14	.	.	PUNCT
easat-5356	263	1	the	the	DET
easat-5356	263	2	code	code	NOUN
easat-5356	263	3	for	for	ADP
easat-5356	263	4	implementing	implement	VERB
easat-5356	263	5	the	the	DET
easat-5356	263	6	neural	neural	ADJ
easat-5356	263	7	network	network	NOUN
easat-5356	263	8	was	be	AUX
easat-5356	263	9	modified	modify	VERB
easat-5356	263	10	from	from	ADP
easat-5356	263	11	ultralytics	ultralytic	NOUN
easat-5356	263	12	version	version	PROPN
easat-5356	263	13	8.0.202	8.0.202	PROPN
easat-5356	263	14	.	.	PUNCT
easat-5356	264	1	the	the	DET
easat-5356	264	2	input	input	NOUN
easat-5356	264	3	images	image	NOUN
easat-5356	264	4	were	be	AUX
easat-5356	264	5	resized	resize	VERB
easat-5356	264	6	to	to	ADP
easat-5356	264	7	640	640	NUM
easat-5356	264	8	×	×	NOUN
easat-5356	264	9	640	640	NUM
easat-5356	264	10	pixels	pixel	NOUN
easat-5356	264	11	,	,	PUNCT
easat-5356	264	12	and	and	CCONJ
easat-5356	264	13	training	training	NOUN
easat-5356	264	14	was	be	AUX
easat-5356	264	15	carried	carry	VERB
easat-5356	264	16	out	out	ADP
easat-5356	264	17	for	for	ADP
easat-5356	264	18	300	300	NUM
easat-5356	264	19	epochs	epoch	NOUN
easat-5356	264	20	.	.	PUNCT
easat-5356	265	1	optimization	optimization	NOUN
easat-5356	265	2	was	be	AUX
easat-5356	265	3	performed	perform	VERB
easat-5356	265	4	using	use	VERB
easat-5356	265	5	the	the	DET
easat-5356	265	6	stochastic	stochastic	ADJ
easat-5356	265	7	gradient	gradient	ADJ
easat-5356	265	8	descent	descent	NOUN
easat-5356	265	9	(	(	PUNCT
easat-5356	265	10	sgd	sgd	NOUN
easat-5356	265	11	)	)	PUNCT
easat-5356	265	12	algorithm	algorithm	NOUN
easat-5356	265	13	,	,	PUNCT
easat-5356	265	14	starting	start	VERB
easat-5356	265	15	with	with	ADP
easat-5356	265	16	an	an	DET
easat-5356	265	17	initial	initial	ADJ
easat-5356	265	18	learning	learning	NOUN
easat-5356	265	19	rate	rate	NOUN
easat-5356	265	20	of	of	ADP
easat-5356	265	21	0.01	0.01	NUM
easat-5356	265	22	,	,	PUNCT
easat-5356	265	23	which	which	PRON
easat-5356	265	24	was	be	AUX
easat-5356	265	25	progressively	progressively	ADV
easat-5356	265	26	reduced	reduce	VERB
easat-5356	265	27	to	to	ADP
easat-5356	265	28	0.0001	0.0001	NUM
easat-5356	265	29	.	.	PUNCT
easat-5356	266	1	to	to	PART
easat-5356	266	2	ensure	ensure	VERB
easat-5356	266	3	a	a	DET
easat-5356	266	4	fair	fair	ADJ
easat-5356	266	5	and	and	CCONJ
easat-5356	266	6	unbiased	unbiased	ADJ
easat-5356	266	7	comparison	comparison	NOUN
easat-5356	266	8	among	among	ADP
easat-5356	266	9	models	model	NOUN
easat-5356	266	10	,	,	PUNCT
easat-5356	266	11	each	each	PRON
easat-5356	266	12	was	be	AUX
easat-5356	266	13	trained	train	VERB
easat-5356	266	14	from	from	ADP
easat-5356	266	15	scratch	scratch	NOUN
easat-5356	266	16	without	without	ADP
easat-5356	266	17	employing	employ	VERB
easat-5356	266	18	pretrained	pretraine	VERB
easat-5356	266	19	weights	weight	NOUN
easat-5356	266	20	.	.	PUNCT
easat-5356	267	1	this	this	DET
easat-5356	267	2	configuration	configuration	NOUN
easat-5356	267	3	made	make	VERB
easat-5356	267	4	it	it	PRON
easat-5356	267	5	possible	possible	ADJ
easat-5356	267	6	to	to	PART
easat-5356	267	7	assess	assess	VERB
easat-5356	267	8	the	the	DET
easat-5356	267	9	model	model	NOUN
easat-5356	267	10	's	's	PART
easat-5356	267	11	performance	performance	NOUN
easat-5356	267	12	only	only	ADV
easat-5356	267	13	using	use	VERB
easat-5356	267	14	its	its	PRON
easat-5356	267	15	architecture	architecture	NOUN
easat-5356	267	16	and	and	CCONJ
easat-5356	267	17	training	training	NOUN
easat-5356	267	18	.	.	PUNCT
easat-5356	268	1	4.2	4.2	NUM
easat-5356	268	2	.	.	PUNCT
easat-5356	268	3	indicators	indicator	NOUN
easat-5356	268	4	of	of	ADP
easat-5356	268	5	experimental	experimental	ADJ
easat-5356	268	6	evaluation	evaluation	NOUN
easat-5356	268	7	the	the	DET
easat-5356	268	8	suggested	suggested	ADJ
easat-5356	268	9	model	model	NOUN
easat-5356	268	10	's	's	PART
easat-5356	268	11	detection	detection	NOUN
easat-5356	268	12	capabilities	capability	NOUN
easat-5356	268	13	and	and	CCONJ
easat-5356	268	14	model	model	NOUN
easat-5356	268	15	size	size	NOUN
easat-5356	268	16	were	be	AUX
easat-5356	268	17	assessed	assess	VERB
easat-5356	268	18	.	.	PUNCT
easat-5356	269	1	the	the	DET
easat-5356	269	2	accuracy	accuracy	NOUN
easat-5356	269	3	of	of	ADP
easat-5356	269	4	each	each	DET
easat-5356	269	5	object	object	NOUN
easat-5356	269	6	category	category	NOUN
easat-5356	269	7	was	be	AUX
easat-5356	269	8	assessed	assess	VERB
easat-5356	269	9	using	use	VERB
easat-5356	269	10	the	the	DET
easat-5356	269	11	precision	precision	NOUN
easat-5356	269	12	(	(	PUNCT
easat-5356	269	13	p	p	NOUN
easat-5356	269	14	)	)	PUNCT
easat-5356	269	15	,	,	PUNCT
easat-5356	269	16	recall	recall	INTJ
easat-5356	269	17	(	(	PUNCT
easat-5356	269	18	r	r	NOUN
easat-5356	269	19	)	)	PUNCT
easat-5356	269	20	,	,	PUNCT
easat-5356	269	21	average	average	ADJ
easat-5356	269	22	precision	precision	NOUN
easat-5356	269	23	(	(	PUNCT
easat-5356	269	24	ap	ap	PROPN
easat-5356	269	25	)	)	PUNCT
easat-5356	269	26	,	,	PUNCT
easat-5356	269	27	and	and	CCONJ
easat-5356	269	28	mean	mean	VERB
easat-5356	269	29	average	average	ADJ
easat-5356	269	30	precision	precision	NOUN
easat-5356	269	31	(	(	PUNCT
easat-5356	269	32	map	map	NOUN
easat-5356	269	33	)	)	PUNCT
easat-5356	269	34	.	.	PUNCT
easat-5356	270	1	additionally	additionally	ADV
easat-5356	270	2	,	,	PUNCT
easat-5356	270	3	the	the	DET
easat-5356	270	4	giga	giga	ADJ
easat-5356	270	5	floating	float	VERB
easat-5356	270	6	-	-	PUNCT
easat-5356	270	7	point	point	NOUN
easat-5356	270	8	operations	operation	NOUN
easat-5356	270	9	per	per	ADP
easat-5356	270	10	second	second	ADJ
easat-5356	270	11	(	(	PUNCT
easat-5356	270	12	gflops	gflop	NOUN
easat-5356	270	13	)	)	PUNCT
easat-5356	270	14	metric	metric	NOUN
easat-5356	270	15	was	be	AUX
easat-5356	270	16	used	use	VERB
easat-5356	270	17	to	to	PART
easat-5356	270	18	evaluate	evaluate	VERB
easat-5356	270	19	the	the	DET
easat-5356	270	20	networks	network	NOUN
easat-5356	270	21	'	'	PART
easat-5356	270	22	computational	computational	ADJ
easat-5356	270	23	complexity	complexity	NOUN
easat-5356	270	24	.	.	PUNCT
easat-5356	271	1	frames	frame	NOUN
easat-5356	271	2	per	per	ADP
easat-5356	271	3	second	second	ADJ
easat-5356	271	4	(	(	PUNCT
easat-5356	271	5	fps	fps	PROPN
easat-5356	271	6	)	)	PUNCT
easat-5356	271	7	was	be	AUX
easat-5356	271	8	used	use	VERB
easat-5356	271	9	to	to	PART
easat-5356	271	10	gauge	gauge	VERB
easat-5356	271	11	the	the	DET
easat-5356	271	12	model	model	NOUN
easat-5356	271	13	's	's	PART
easat-5356	271	14	real	real	ADJ
easat-5356	271	15	-	-	PUNCT
easat-5356	271	16	time	time	NOUN
easat-5356	271	17	processing	processing	NOUN
easat-5356	271	18	performance	performance	NOUN
easat-5356	271	19	,	,	PUNCT
easat-5356	271	20	and	and	CCONJ
easat-5356	271	21	the	the	DET
easat-5356	271	22	number	number	NOUN
easat-5356	271	23	of	of	ADP
easat-5356	271	24	parameters	parameter	NOUN
easat-5356	271	25	dictated	dictate	VERB
easat-5356	271	26	the	the	DET
easat-5356	271	27	model	model	NOUN
easat-5356	271	28	's	's	PART
easat-5356	271	29	size	size	NOUN
easat-5356	271	30	.	.	PUNCT
easat-5356	272	1	precision	precision	NOUN
easat-5356	272	2	is	be	AUX
easat-5356	272	3	defined	define	VERB
easat-5356	272	4	as	as	SCONJ
easat-5356	272	5	follows	follow	VERB
easat-5356	272	6	:	:	PUNCT
easat-5356	272	7	precision	precision	NOUN
easat-5356	272	8	is	be	AUX
easat-5356	272	9	the	the	DET
easat-5356	272	10	ratio	ratio	NOUN
easat-5356	272	11	of	of	ADP
easat-5356	272	12	all	all	DET
easat-5356	272	13	detected	detect	VERB
easat-5356	272	14	samples	sample	NOUN
easat-5356	272	15	to	to	ADP
easat-5356	272	16	the	the	DET
easat-5356	272	17	number	number	NOUN
easat-5356	272	18	of	of	ADP
easat-5356	272	19	positive	positive	ADJ
easat-5356	272	20	samples	sample	NOUN
easat-5356	272	21	predicted	predict	VERB
easat-5356	272	22	by	by	ADP
easat-5356	272	23	the	the	DET
easat-5356	272	24	model	model	NOUN
easat-5356	272	25	:	:	PUNCT
easat-5356	272	26	precision	precision	NOUN
easat-5356	272	27	=	=	SYM
easat-5356	272	28	𝑇𝑃	𝑇𝑃	PROPN
easat-5356	272	29	𝑇𝑃	𝑇𝑃	PROPN
easat-5356	272	30	+	+	CCONJ
easat-5356	272	31	𝐹𝑃	𝐹𝑃	PROPN
easat-5356	272	32	(	(	PUNCT
easat-5356	272	33	12	12	NUM
easat-5356	272	34	)	)	PUNCT
easat-5356	272	35	851	851	NUM
easat-5356	272	36	edelweiss	edelweiss	PROPN
easat-5356	272	37	applied	apply	VERB
easat-5356	272	38	science	science	NOUN
easat-5356	272	39	and	and	CCONJ
easat-5356	272	40	technology	technology	NOUN
easat-5356	272	41	issn	issn	PROPN
easat-5356	272	42	:	:	PUNCT
easat-5356	272	43	2576	2576	NUM
easat-5356	272	44	-	-	SYM
easat-5356	272	45	8484	8484	NUM
easat-5356	272	46	vol	vol	NOUN
easat-5356	272	47	.	.	PROPN
easat-5356	273	1	9	9	NUM
easat-5356	273	2	,	,	PUNCT
easat-5356	273	3	no	no	INTJ
easat-5356	273	4	.	.	NOUN
easat-5356	274	1	3	3	NUM
easat-5356	274	2	:	:	PUNCT
easat-5356	274	3	840	840	NUM
easat-5356	274	4	-	-	SYM
easat-5356	274	5	855	855	NUM
easat-5356	274	6	,	,	PUNCT
easat-5356	274	7	2025	2025	NUM
easat-5356	274	8	doi	doi	NOUN
easat-5356	274	9	:	:	PUNCT
easat-5356	274	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	274	11	©	©	PROPN
easat-5356	274	12	2025	2025	NUM
easat-5356	274	13	by	by	ADP
easat-5356	274	14	the	the	DET
easat-5356	274	15	authors	author	NOUN
easat-5356	274	16	;	;	PUNCT
easat-5356	274	17	licensee	licensee	PROPN
easat-5356	274	18	learning	learn	VERB
easat-5356	274	19	gate	gate	PROPN
easat-5356	274	20	precision	precision	NOUN
easat-5356	274	21	is	be	AUX
easat-5356	274	22	calculated	calculate	VERB
easat-5356	274	23	as	as	SCONJ
easat-5356	274	24	follows	follow	VERB
easat-5356	274	25	:	:	PUNCT
easat-5356	274	26	precision	precision	NOUN
easat-5356	274	27	is	be	AUX
easat-5356	274	28	the	the	DET
easat-5356	274	29	ratio	ratio	NOUN
easat-5356	274	30	of	of	ADP
easat-5356	274	31	all	all	DET
easat-5356	274	32	detected	detect	VERB
easat-5356	274	33	samples	sample	NOUN
easat-5356	274	34	to	to	ADP
easat-5356	274	35	the	the	DET
easat-5356	274	36	number	number	NOUN
easat-5356	274	37	of	of	ADP
easat-5356	274	38	positive	positive	ADJ
easat-5356	274	39	samples	sample	NOUN
easat-5356	274	40	identified	identify	VERB
easat-5356	274	41	by	by	ADP
easat-5356	274	42	the	the	DET
easat-5356	274	43	model	model	NOUN
easat-5356	274	44	:	:	PUNCT
easat-5356	274	45	recall	recall	NOUN
easat-5356	274	46	=	=	SYM
easat-5356	274	47	𝑇𝑃	𝑇𝑃	PROPN
easat-5356	274	48	𝑇𝑃	𝑇𝑃	PROPN
easat-5356	274	49	+	+	CCONJ
easat-5356	275	1	𝐹𝑁	𝐹𝑁	PROPN
easat-5356	275	2	(	(	PUNCT
easat-5356	275	3	13	13	NUM
easat-5356	275	4	)	)	PUNCT
easat-5356	275	5	the	the	DET
easat-5356	275	6	area	area	NOUN
easat-5356	275	7	under	under	ADP
easat-5356	275	8	the	the	DET
easat-5356	275	9	precision	precision	NOUN
easat-5356	275	10	-	-	PUNCT
easat-5356	275	11	recall	recall	NOUN
easat-5356	275	12	curve	curve	NOUN
easat-5356	275	13	is	be	AUX
easat-5356	275	14	the	the	DET
easat-5356	275	15	mean	mean	ADJ
easat-5356	275	16	precision	precision	NOUN
easat-5356	275	17	(	(	PUNCT
easat-5356	275	18	ap	ap	PROPN
easat-5356	275	19	)	)	PUNCT
easat-5356	275	20	,	,	PUNCT
easat-5356	275	21	which	which	PRON
easat-5356	275	22	is	be	AUX
easat-5356	275	23	calculated	calculate	VERB
easat-5356	275	24	as	as	SCONJ
easat-5356	275	25	follows	follow	VERB
easat-5356	275	26	:	:	PUNCT
easat-5356	275	27	ap	ap	PROPN
easat-5356	276	1	=	=	SYM
easat-5356	276	2	∫	∫	PROPN
easat-5356	276	3	  	  	SPACE
easat-5356	276	4	1	1	NUM
easat-5356	276	5	0	0	NUM
easat-5356	276	6	precision(recall)𝑑(recall	precision(recall)𝑑(recall	NOUN
easat-5356	276	7	)	)	PUNCT
easat-5356	276	8	(	(	PUNCT
easat-5356	276	9	14	14	NUM
easat-5356	276	10	)	)	PUNCT
easat-5356	276	11	the	the	DET
easat-5356	276	12	model	model	NOUN
easat-5356	276	13	's	's	PART
easat-5356	276	14	detection	detection	NOUN
easat-5356	276	15	performance	performance	NOUN
easat-5356	276	16	across	across	ADP
easat-5356	276	17	all	all	DET
easat-5356	276	18	categories	category	NOUN
easat-5356	276	19	is	be	AUX
easat-5356	276	20	measured	measure	VERB
easat-5356	276	21	by	by	ADP
easat-5356	276	22	mean	mean	ADJ
easat-5356	276	23	average	average	ADJ
easat-5356	276	24	precision(map	precision(map	NOUN
easat-5356	276	25	)	)	PUNCT
easat-5356	276	26	,	,	PUNCT
easat-5356	276	27	which	which	PRON
easat-5356	276	28	is	be	AUX
easat-5356	276	29	the	the	DET
easat-5356	276	30	outcome	outcome	NOUN
easat-5356	276	31	of	of	ADP
easat-5356	276	32	the	the	DET
easat-5356	276	33	weighted	weight	VERB
easat-5356	276	34	average	average	NOUN
easat-5356	276	35	of	of	ADP
easat-5356	276	36	ap	ap	PROPN
easat-5356	276	37	values	value	NOUN
easat-5356	276	38	for	for	ADP
easat-5356	276	39	each	each	DET
easat-5356	276	40	sample	sample	NOUN
easat-5356	276	41	category	category	NOUN
easat-5356	276	42	.	.	PUNCT
easat-5356	277	1	the	the	DET
easat-5356	277	2	formula	formula	NOUN
easat-5356	277	3	is	be	AUX
easat-5356	277	4	displayed	display	VERB
easat-5356	277	5	below	below	ADV
easat-5356	277	6	:	:	PUNCT
easat-5356	277	7	map	map	VERB
easat-5356	277	8	=	=	SYM
easat-5356	277	9	1	1	NUM
easat-5356	277	10	𝑁	𝑁	PROPN
easat-5356	277	11	∑	∑	DET
easat-5356	277	12	  	  	SPACE
easat-5356	277	13	𝑁	𝑁	PROPN
easat-5356	277	14	𝑖=1	𝑖=1	PROPN
easat-5356	277	15	𝐴𝑃𝑖	𝐴𝑃𝑖	NOUN
easat-5356	277	16	(	(	PUNCT
easat-5356	277	17	15	15	NUM
easat-5356	277	18	)	)	PUNCT
easat-5356	277	19	the	the	DET
easat-5356	277	20	value	value	NOUN
easat-5356	277	21	having	have	VERB
easat-5356	277	22	a	a	DET
easat-5356	277	23	category	category	NOUN
easat-5356	277	24	index	index	NOUN
easat-5356	277	25	value	value	NOUN
easat-5356	277	26	of	of	ADP
easat-5356	277	27	is	be	AUX
easat-5356	277	28	indicated	indicate	VERB
easat-5356	277	29	in	in	ADP
easat-5356	277	30	the	the	DET
easat-5356	277	31	equation	equation	NOUN
easat-5356	277	32	above	above	ADV
easat-5356	277	33	.	.	PUNCT
easat-5356	278	1	indicates	indicate	VERB
easat-5356	278	2	how	how	SCONJ
easat-5356	278	3	many	many	ADJ
easat-5356	278	4	sample	sample	NOUN
easat-5356	278	5	categories	category	NOUN
easat-5356	278	6	there	there	PRON
easat-5356	278	7	are	be	VERB
easat-5356	278	8	in	in	ADP
easat-5356	278	9	the	the	DET
easat-5356	278	10	training	training	NOUN
easat-5356	278	11	dataset	dataset	NOUN
easat-5356	278	12	.	.	PUNCT
easat-5356	279	1	when	when	SCONJ
easat-5356	279	2	the	the	DET
easat-5356	279	3	detection	detection	NOUN
easat-5356	279	4	model	model	NOUN
easat-5356	279	5	iou	iou	PROPN
easat-5356	279	6	is	be	AUX
easat-5356	279	7	set	set	VERB
easat-5356	279	8	to	to	ADP
easat-5356	279	9	0.5	0.5	NUM
easat-5356	279	10	,	,	PUNCT
easat-5356	279	11	the	the	DET
easat-5356	279	12	average	average	ADJ
easat-5356	279	13	accuracy	accuracy	NOUN
easat-5356	279	14	is	be	AUX
easat-5356	279	15	indicated	indicate	VERB
easat-5356	279	16	by	by	ADP
easat-5356	279	17	map0.5	map0.5	PROPN
easat-5356	279	18	:	:	PUNCT
easat-5356	279	19	when	when	SCONJ
easat-5356	279	20	the	the	DET
easat-5356	279	21	detection	detection	NOUN
easat-5356	279	22	model	model	NOUN
easat-5356	279	23	iou	iou	PROPN
easat-5356	279	24	is	be	AUX
easat-5356	279	25	set	set	VERB
easat-5356	279	26	to	to	ADP
easat-5356	279	27	0.5	0.5	NUM
easat-5356	279	28	to	to	PART
easat-5356	279	29	0.95	0.95	NUM
easat-5356	279	30	(	(	PUNCT
easat-5356	279	31	taken	take	VERB
easat-5356	279	32	at	at	ADP
easat-5356	279	33	0.5	0.5	NUM
easat-5356	279	34	intervals	interval	NOUN
easat-5356	279	35	)	)	PUNCT
easat-5356	279	36	,	,	PUNCT
easat-5356	279	37	the	the	DET
easat-5356	279	38	average	average	ADJ
easat-5356	279	39	accuracy	accuracy	NOUN
easat-5356	279	40	is	be	AUX
easat-5356	279	41	indicated	indicate	VERB
easat-5356	279	42	by	by	ADP
easat-5356	279	43	map	map	VERB
easat-5356	279	44	0.5:0.95	0.5:0.95	PROPN
easat-5356	279	45	.	.	PUNCT
easat-5356	280	1	5	5	NUM
easat-5356	280	2	.	.	PUNCT
easat-5356	280	3	yolov8	yolov8	NOUN
easat-5356	280	4	comparative	comparative	ADJ
easat-5356	280	5	analysis	analysis	NOUN
easat-5356	280	6	table	table	NOUN
easat-5356	280	7	1and	1and	NUM
easat-5356	280	8	table	table	NOUN
easat-5356	280	9	2	2	NUM
easat-5356	280	10	as	as	ADV
easat-5356	280	11	well	well	ADV
easat-5356	280	12	as	as	ADP
easat-5356	280	13	figure	figure	NOUN
easat-5356	280	14	5	5	NUM
easat-5356	280	15	display	display	VERB
easat-5356	280	16	the	the	DET
easat-5356	280	17	gcsem	gcsem	ADJ
easat-5356	280	18	-yolo	-yolo	NOUN
easat-5356	280	19	and	and	CCONJ
easat-5356	280	20	yolov8	yolov8	NOUN
easat-5356	280	21	models	model	NOUN
easat-5356	280	22	'	'	PART
easat-5356	280	23	sizes	size	NOUN
easat-5356	280	24	and	and	CCONJ
easat-5356	280	25	performance	performance	NOUN
easat-5356	280	26	on	on	ADP
easat-5356	280	27	the	the	DET
easat-5356	280	28	wider	wide	ADJ
easat-5356	280	29	face	face	NOUN
easat-5356	280	30	dataset	dataset	NOUN
easat-5356	280	31	.	.	PUNCT
easat-5356	281	1	table	table	NOUN
easat-5356	281	2	1	1	NUM
easat-5356	281	3	.	.	PUNCT
easat-5356	282	1	a	a	DET
easat-5356	282	2	comparison	comparison	NOUN
easat-5356	282	3	between	between	ADP
easat-5356	282	4	yolov8	yolov8	NOUN
easat-5356	282	5	and	and	CCONJ
easat-5356	282	6	gcsem	gcsem	PROPN
easat-5356	282	7	-	-	PUNCT
easat-5356	282	8	yolo	yolo	PROPN
easat-5356	282	9	.	.	PUNCT
easat-5356	283	1	network	network	NOUN
easat-5356	283	2	precision	precision	NOUN
easat-5356	283	3	(	(	PUNCT
easat-5356	283	4	%	%	INTJ
easat-5356	283	5	)	)	PUNCT
easat-5356	283	6	recall	recall	NOUN
easat-5356	283	7	(	(	PUNCT
easat-5356	283	8	%	%	INTJ
easat-5356	283	9	)	)	PUNCT
easat-5356	284	1	map50	map50	PROPN
easat-5356	284	2	(	(	PUNCT
easat-5356	284	3	%	%	INTJ
easat-5356	284	4	)	)	PUNCT
easat-5356	284	5	map50	map50	PROPN
easat-5356	284	6	-	-	PUNCT
easat-5356	284	7	95	95	NUM
easat-5356	284	8	(	(	PUNCT
easat-5356	284	9	%	%	INTJ
easat-5356	284	10	)	)	PUNCT
easat-5356	284	11	fps	fps	PROPN
easat-5356	284	12	yolov8n	yolov8n	PROPN
easat-5356	284	13	84.7	84.7	NUM
easat-5356	284	14	58.0	58.0	NUM
easat-5356	284	15	66.7	66.7	NUM
easat-5356	284	16	36.6	36.6	NUM
easat-5356	284	17	162	162	NUM
easat-5356	284	18	gcsem	gcsem	ADJ
easat-5356	284	19	-	-	PUNCT
easat-5356	284	20	yolo	yolo	NOUN
easat-5356	284	21	-	-	PUNCT
easat-5356	284	22	n	n	PROPN
easat-5356	284	23	89.3	89.3	NUM
easat-5356	284	24	69.9	69.9	NUM
easat-5356	284	25	78.8	78.8	NUM
easat-5356	284	26	45.1	45.1	NUM
easat-5356	284	27	142	142	NUM
easat-5356	284	28	yolov8s	yolov8	NOUN
easat-5356	284	29	85.2	85.2	NUM
easat-5356	284	30	58.6	58.6	NUM
easat-5356	284	31	69.5	69.5	NUM
easat-5356	284	32	40.9	40.9	NUM
easat-5356	284	33	148	148	NUM
easat-5356	284	34	gcsem	gcsem	ADJ
easat-5356	284	35	-yolo	-yolo	PROPN
easat-5356	284	36	-	-	PUNCT
easat-5356	284	37	s	s	PART
easat-5356	284	38	92.2	92.2	NUM
easat-5356	284	39	72.9	72.9	NUM
easat-5356	284	40	82.5	82.5	NUM
easat-5356	284	41	47.1	47.1	NUM
easat-5356	284	42	117	117	NUM
easat-5356	284	43	yolov8	yolov8	NOUN
easat-5356	284	44	m	m	VERB
easat-5356	284	45	87.6	87.6	NUM
easat-5356	284	46	60.9	60.9	NUM
easat-5356	284	47	72.3	72.3	NUM
easat-5356	284	48	43.6	43.6	NUM
easat-5356	284	49	117	117	NUM
easat-5356	284	50	gcsem	gcsem	ADJ
easat-5356	284	51	-yolo	-yolo	PROPN
easat-5356	284	52	-	-	PUNCT
easat-5356	284	53	m	m	PROPN
easat-5356	284	54	95.3	95.3	NUM
easat-5356	284	55	76.8	76.8	NUM
easat-5356	284	56	88.1	88.1	NUM
easat-5356	284	57	49.5	49.5	NUM
easat-5356	284	58	99	99	NUM
easat-5356	284	59	yolov8l	yolov8l	NOUN
easat-5356	284	60	88.9	88.9	NUM
easat-5356	284	61	65.1	65.1	NUM
easat-5356	284	62	76.9	76.9	NUM
easat-5356	284	63	49.2	49.2	NUM
easat-5356	284	64	94	94	NUM
easat-5356	284	65	gcsem	gcsem	ADJ
easat-5356	284	66	-yolo	-yolo	PROPN
easat-5356	284	67	-	-	PUNCT
easat-5356	284	68	l	l	NOUN
easat-5356	284	69	98.4	98.4	NUM
easat-5356	284	70	82.9	82.9	NUM
easat-5356	284	71	95.1	95.1	NUM
easat-5356	284	72	55.7	55.7	NUM
easat-5356	284	73	88	88	NUM
easat-5356	284	74	in	in	ADP
easat-5356	284	75	terms	term	NOUN
easat-5356	284	76	of	of	ADP
easat-5356	284	77	detection	detection	NOUN
easat-5356	284	78	accuracy	accuracy	NOUN
easat-5356	284	79	at	at	ADP
easat-5356	284	80	various	various	ADJ
easat-5356	284	81	scales	scale	NOUN
easat-5356	284	82	,	,	PUNCT
easat-5356	284	83	the	the	DET
easat-5356	284	84	results	result	NOUN
easat-5356	284	85	in	in	ADP
easat-5356	284	86	table	table	NOUN
easat-5356	284	87	1	1	NUM
easat-5356	284	88	show	show	VERB
easat-5356	284	89	that	that	SCONJ
easat-5356	284	90	gcsem	gcsem	ADJ
easat-5356	284	91	-	-	PUNCT
easat-5356	284	92	yolo	yolo	PROPN
easat-5356	284	93	performs	perform	VERB
easat-5356	284	94	better	well	ADJ
easat-5356	284	95	than	than	ADP
easat-5356	284	96	the	the	DET
easat-5356	284	97	conventional	conventional	ADJ
easat-5356	284	98	yolov8	yolov8	NOUN
easat-5356	284	99	model	model	NOUN
easat-5356	284	100	.	.	PUNCT
easat-5356	285	1	in	in	ADP
easat-5356	285	2	particular	particular	ADJ
easat-5356	285	3	,	,	PUNCT
easat-5356	285	4	gcsem	gcsem	ADJ
easat-5356	285	5	-	-	PUNCT
easat-5356	285	6	yolo	yolo	NOUN
easat-5356	285	7	-	-	PUNCT
easat-5356	285	8	n	n	NOUN
easat-5356	285	9	outperforms	outperform	NOUN
easat-5356	285	10	yolov8n	yolov8n	NOUN
easat-5356	285	11	by	by	ADP
easat-5356	285	12	12.1	12.1	NUM
easat-5356	285	13	%	%	NOUN
easat-5356	285	14	,	,	PUNCT
easat-5356	285	15	achieving	achieve	VERB
easat-5356	285	16	a	a	DET
easat-5356	285	17	map50	map50	ADJ
easat-5356	285	18	score	score	NOUN
easat-5356	285	19	of	of	ADP
easat-5356	285	20	78.8	78.8	NUM
easat-5356	285	21	%	%	NOUN
easat-5356	285	22	.	.	PUNCT
easat-5356	286	1	furthermore	furthermore	ADV
easat-5356	286	2	,	,	PUNCT
easat-5356	286	3	at	at	ADP
easat-5356	286	4	all	all	DET
easat-5356	286	5	scales	scale	NOUN
easat-5356	286	6	,	,	PUNCT
easat-5356	286	7	gcsem	gcsem	ADJ
easat-5356	286	8	-	-	PUNCT
easat-5356	286	9	yolo	yolo	ADJ
easat-5356	286	10	attains	attain	NOUN
easat-5356	286	11	better	well	ADJ
easat-5356	286	12	accuracies	accuracy	NOUN
easat-5356	286	13	based	base	VERB
easat-5356	286	14	on	on	ADP
easat-5356	286	15	an	an	DET
easat-5356	286	16	fps	fps	NOUN
easat-5356	286	17	that	that	PRON
easat-5356	286	18	is	be	AUX
easat-5356	286	19	just	just	ADV
easat-5356	286	20	marginally	marginally	ADV
easat-5356	286	21	less	less	ADJ
easat-5356	286	22	than	than	ADP
easat-5356	286	23	yolov8	yolov8	PROPN
easat-5356	286	24	.	.	PUNCT
easat-5356	287	1	according	accord	VERB
easat-5356	287	2	to	to	ADP
easat-5356	287	3	these	these	DET
easat-5356	287	4	findings	finding	NOUN
easat-5356	287	5	,	,	PUNCT
easat-5356	287	6	gcsem	gcsem	ADJ
easat-5356	287	7	-	-	PUNCT
easat-5356	287	8	yolo	yolo	NOUN
easat-5356	287	9	reduces	reduce	VERB
easat-5356	287	10	computational	computational	ADJ
easat-5356	287	11	costs	cost	NOUN
easat-5356	287	12	and	and	CCONJ
easat-5356	287	13	increases	increase	VERB
easat-5356	287	14	detection	detection	NOUN
easat-5356	287	15	accuracy	accuracy	NOUN
easat-5356	287	16	while	while	SCONJ
easat-5356	287	17	using	use	VERB
easat-5356	287	18	a	a	DET
easat-5356	287	19	smaller	small	ADJ
easat-5356	287	20	model	model	NOUN
easat-5356	287	21	size	size	NOUN
easat-5356	287	22	.	.	PUNCT
easat-5356	288	1	852	852	NUM
easat-5356	288	2	edelweiss	edelweiss	PROPN
easat-5356	288	3	applied	apply	VERB
easat-5356	288	4	science	science	NOUN
easat-5356	288	5	and	and	CCONJ
easat-5356	288	6	technology	technology	NOUN
easat-5356	288	7	issn	issn	PROPN
easat-5356	288	8	:	:	PUNCT
easat-5356	288	9	2576	2576	NUM
easat-5356	288	10	-	-	SYM
easat-5356	288	11	8484	8484	NUM
easat-5356	288	12	vol	vol	NOUN
easat-5356	288	13	.	.	PROPN
easat-5356	289	1	9	9	NUM
easat-5356	289	2	,	,	PUNCT
easat-5356	289	3	no	no	INTJ
easat-5356	289	4	.	.	NOUN
easat-5356	290	1	3	3	NUM
easat-5356	290	2	:	:	PUNCT
easat-5356	290	3	840	840	NUM
easat-5356	290	4	-	-	SYM
easat-5356	290	5	855	855	NUM
easat-5356	290	6	,	,	PUNCT
easat-5356	290	7	2025	2025	NUM
easat-5356	290	8	doi	doi	NOUN
easat-5356	290	9	:	:	PUNCT
easat-5356	290	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	290	11	©	©	PROPN
easat-5356	290	12	2025	2025	NUM
easat-5356	290	13	by	by	ADP
easat-5356	290	14	the	the	DET
easat-5356	290	15	authors	author	NOUN
easat-5356	290	16	;	;	PUNCT
easat-5356	290	17	licensee	licensee	PROPN
easat-5356	290	18	learning	learn	VERB
easat-5356	290	19	gate	gate	PROPN
easat-5356	290	20	figure	figure	NOUN
easat-5356	290	21	5	5	NUM
easat-5356	290	22	.	.	PUNCT
easat-5356	290	23	detection	detection	NOUN
easat-5356	290	24	accuracy	accuracy	NOUN
easat-5356	290	25	versus	versus	ADP
easat-5356	290	26	scale	scale	NOUN
easat-5356	290	27	.	.	PUNCT
easat-5356	291	1	regardless	regardless	ADV
easat-5356	291	2	of	of	ADP
easat-5356	291	3	the	the	DET
easat-5356	291	4	scale	scale	NOUN
easat-5356	291	5	,	,	PUNCT
easat-5356	291	6	figure	figure	NOUN
easat-5356	291	7	5	5	NUM
easat-5356	291	8	shows	show	VERB
easat-5356	291	9	that	that	SCONJ
easat-5356	291	10	the	the	DET
easat-5356	291	11	suggested	suggest	VERB
easat-5356	291	12	detection	detection	NOUN
easat-5356	291	13	model	model	NOUN
easat-5356	291	14	continuously	continuously	ADV
easat-5356	291	15	performs	perform	VERB
easat-5356	291	16	better	well	ADJ
easat-5356	291	17	than	than	ADP
easat-5356	291	18	the	the	DET
easat-5356	291	19	baseline	baseline	NOUN
easat-5356	291	20	model	model	NOUN
easat-5356	291	21	.	.	PUNCT
easat-5356	292	1	subsequent	subsequent	ADJ
easat-5356	292	2	study	study	NOUN
easat-5356	292	3	shows	show	VERB
easat-5356	292	4	that	that	SCONJ
easat-5356	292	5	the	the	DET
easat-5356	292	6	suggested	suggest	VERB
easat-5356	292	7	model	model	NOUN
easat-5356	292	8	's	's	PART
easat-5356	292	9	detection	detection	NOUN
easat-5356	292	10	performance	performance	NOUN
easat-5356	292	11	differs	differ	VERB
easat-5356	292	12	dramatically	dramatically	ADV
easat-5356	292	13	from	from	ADP
easat-5356	292	14	the	the	DET
easat-5356	292	15	baseline	baseline	NOUN
easat-5356	292	16	model	model	NOUN
easat-5356	292	17	(	(	PUNCT
easat-5356	292	18	in	in	ADP
easat-5356	292	19	this	this	DET
easat-5356	292	20	case	case	NOUN
easat-5356	292	21	,	,	PUNCT
easat-5356	292	22	yolov8	yolov8	PROPN
easat-5356	292	23	)	)	PUNCT
easat-5356	292	24	as	as	SCONJ
easat-5356	292	25	the	the	DET
easat-5356	292	26	model	model	NOUN
easat-5356	292	27	size	size	NOUN
easat-5356	292	28	grows	grow	VERB
easat-5356	292	29	.	.	PUNCT
easat-5356	293	1	according	accord	VERB
easat-5356	293	2	to	to	ADP
easat-5356	293	3	the	the	DET
easat-5356	293	4	map50	map50	PROPN
easat-5356	293	5	indicator	indicator	NOUN
easat-5356	293	6	,	,	PUNCT
easat-5356	293	7	gcsem	gcsem	ADJ
easat-5356	293	8	-	-	PUNCT
easat-5356	293	9	yolo	yolo	NOUN
easat-5356	293	10	-	-	PUNCT
easat-5356	293	11	n	n	NOUN
easat-5356	293	12	performs	perform	VERB
easat-5356	293	13	12.1	12.1	NUM
easat-5356	293	14	%	%	NOUN
easat-5356	293	15	better	well	ADJ
easat-5356	293	16	than	than	ADP
easat-5356	293	17	yolov8	yolov8	NOUN
easat-5356	293	18	in	in	ADP
easat-5356	293	19	the	the	DET
easat-5356	293	20	"	"	PUNCT
easat-5356	293	21	n"-scale	n"-scale	ADJ
easat-5356	293	22	setup	setup	NOUN
easat-5356	293	23	.	.	PUNCT
easat-5356	294	1	nevertheless	nevertheless	ADV
easat-5356	294	2	,	,	PUNCT
easat-5356	294	3	the	the	DET
easat-5356	294	4	map50	map50	PROPN
easat-5356	294	5	indication	indication	NOUN
easat-5356	294	6	is	be	AUX
easat-5356	294	7	18.2	18.2	NUM
easat-5356	294	8	%	%	NOUN
easat-5356	294	9	larger	large	ADJ
easat-5356	294	10	when	when	SCONJ
easat-5356	294	11	the	the	DET
easat-5356	294	12	scale	scale	NOUN
easat-5356	294	13	is	be	AUX
easat-5356	294	14	raised	raise	VERB
easat-5356	294	15	to	to	ADP
easat-5356	294	16	"	"	PUNCT
easat-5356	294	17	l	l	NOUN
easat-5356	294	18	,	,	PUNCT
easat-5356	294	19	"	"	PUNCT
easat-5356	294	20	greatly	greatly	ADV
easat-5356	294	21	increasing	increase	VERB
easat-5356	294	22	the	the	DET
easat-5356	294	23	difference	difference	NOUN
easat-5356	294	24	between	between	ADP
easat-5356	294	25	the	the	DET
easat-5356	294	26	two	two	NUM
easat-5356	294	27	models	model	NOUN
easat-5356	294	28	.	.	PUNCT
easat-5356	295	1	this	this	DET
easat-5356	295	2	occurrence	occurrence	NOUN
easat-5356	295	3	suggests	suggest	VERB
easat-5356	295	4	that	that	SCONJ
easat-5356	295	5	as	as	ADP
easat-5356	295	6	the	the	DET
easat-5356	295	7	number	number	NOUN
easat-5356	295	8	of	of	ADP
easat-5356	295	9	model	model	NOUN
easat-5356	295	10	parameters	parameter	NOUN
easat-5356	295	11	and	and	CCONJ
easat-5356	295	12	computational	computational	ADJ
easat-5356	295	13	complexity	complexity	NOUN
easat-5356	295	14	rise	rise	NOUN
easat-5356	295	15	,	,	PUNCT
easat-5356	295	16	gcsem	gcsem	NOUN
easat-5356	295	17	-	-	PUNCT
easat-5356	295	18	yolo	yolo	PROPN
easat-5356	295	19	's	's	PART
easat-5356	295	20	detection	detection	NOUN
easat-5356	295	21	accuracy	accuracy	NOUN
easat-5356	295	22	improves	improve	VERB
easat-5356	295	23	more	more	ADV
easat-5356	295	24	quickly	quickly	ADV
easat-5356	295	25	.	.	PUNCT
easat-5356	296	1	therefore	therefore	ADV
easat-5356	296	2	,	,	PUNCT
easat-5356	296	3	when	when	SCONJ
easat-5356	296	4	implemented	implement	VERB
easat-5356	296	5	on	on	ADP
easat-5356	296	6	a	a	DET
easat-5356	296	7	hardware	hardware	NOUN
easat-5356	296	8	platform	platform	NOUN
easat-5356	296	9	with	with	ADP
easat-5356	296	10	adequate	adequate	ADJ
easat-5356	296	11	computational	computational	ADJ
easat-5356	296	12	capabilities	capability	NOUN
easat-5356	296	13	,	,	PUNCT
easat-5356	296	14	the	the	DET
easat-5356	296	15	suggested	suggest	VERB
easat-5356	296	16	model	model	NOUN
easat-5356	296	17	should	should	AUX
easat-5356	296	18	offer	offer	VERB
easat-5356	296	19	high	high	ADJ
easat-5356	296	20	object	object	NOUN
easat-5356	296	21	-	-	PUNCT
easat-5356	296	22	detection	detection	NOUN
easat-5356	296	23	accuracy	accuracy	NOUN
easat-5356	296	24	,	,	PUNCT
easat-5356	296	25	particularly	particularly	ADV
easat-5356	296	26	in	in	ADP
easat-5356	296	27	large	large	ADJ
easat-5356	296	28	-	-	PUNCT
easat-5356	296	29	scale	scale	NOUN
easat-5356	296	30	settings	setting	NOUN
easat-5356	296	31	.	.	PUNCT
easat-5356	297	1	table	table	NOUN
easat-5356	297	2	2	2	NUM
easat-5356	297	3	.	.	PUNCT
easat-5356	297	4	comparing	compare	VERB
easat-5356	297	5	the	the	DET
easat-5356	297	6	detection	detection	NOUN
easat-5356	297	7	of	of	ADP
easat-5356	297	8	faces	face	NOUN
easat-5356	297	9	of	of	ADP
easat-5356	297	10	varying	vary	VERB
easat-5356	297	11	sizes	size	NOUN
easat-5356	297	12	using	use	VERB
easat-5356	297	13	yolov8	yolov8	PROPN
easat-5356	297	14	.	.	PUNCT
easat-5356	298	1	network	network	PROPN
easat-5356	298	2	ap	ap	PROPN
easat-5356	298	3	-	-	PUNCT
easat-5356	298	4	small	small	ADJ
easat-5356	298	5	(	(	PUNCT
easat-5356	298	6	%	%	NOUN
easat-5356	298	7	)	)	PUNCT
easat-5356	298	8	ap	ap	NOUN
easat-5356	298	9	-	-	PUNCT
easat-5356	298	10	medium	medium	NOUN
easat-5356	298	11	(	(	PUNCT
easat-5356	298	12	%	%	NOUN
easat-5356	298	13	)	)	PUNCT
easat-5356	298	14	ap	ap	NOUN
easat-5356	298	15	-	-	PUNCT
easat-5356	298	16	large	large	ADJ
easat-5356	298	17	(	(	PUNCT
easat-5356	298	18	%	%	NOUN
easat-5356	298	19	)	)	PUNCT
easat-5356	298	20	network	network	NOUN
easat-5356	298	21	ap	ap	PROPN
easat-5356	298	22	-	-	PUNCT
easat-5356	298	23	easy	easy	ADJ
easat-5356	298	24	(	(	PUNCT
easat-5356	298	25	%	%	INTJ
easat-5356	298	26	)	)	PUNCT
easat-5356	298	27	ap	ap	NOUN
easat-5356	298	28	-	-	PUNCT
easat-5356	298	29	medium	medium	NOUN
easat-5356	298	30	(	(	PUNCT
easat-5356	298	31	%	%	NOUN
easat-5356	298	32	)	)	PUNCT
easat-5356	298	33	ap	ap	NOUN
easat-5356	298	34	-	-	ADJ
easat-5356	298	35	hard	hard	ADV
easat-5356	298	36	(	(	PUNCT
easat-5356	298	37	%	%	INTJ
easat-5356	298	38	)	)	PUNCT
easat-5356	298	39	yolov8n	yolov8n	NOUN
easat-5356	298	40	75.6	75.6	NUM
easat-5356	298	41	75.1	75.1	NUM
easat-5356	298	42	61.3	61.3	NUM
easat-5356	298	43	gcsem	gcsem	ADJ
easat-5356	298	44	-	-	PUNCT
easat-5356	298	45	yolo	yolo	NOUN
easat-5356	298	46	-	-	PUNCT
easat-5356	298	47	n	n	PROPN
easat-5356	298	48	91.5	91.5	NUM
easat-5356	298	49	89.9	89.9	NUM
easat-5356	298	50	79.3	79.3	NUM
easat-5356	298	51	yolov8s	yolov8	NOUN
easat-5356	298	52	78.4	78.4	NUM
easat-5356	298	53	78.2	78.2	NUM
easat-5356	298	54	61.9	61.9	NUM
easat-5356	298	55	gcsem	gcsem	ADJ
easat-5356	298	56	-yolo	-yolo	PROPN
easat-5356	298	57	-	-	PUNCT
easat-5356	298	58	s	s	PROPN
easat-5356	298	59	92.0	92.0	NUM
easat-5356	298	60	90.1	90.1	NUM
easat-5356	298	61	81.1	81.1	NUM
easat-5356	298	62	yolov8	yolov8	NOUN
easat-5356	298	63	m	m	VERB
easat-5356	298	64	79.1	79.1	NUM
easat-5356	298	65	79.0	79.0	NUM
easat-5356	298	66	62.1	62.1	NUM
easat-5356	298	67	gcsem	gcsem	ADJ
easat-5356	298	68	-yolo	-yolo	PROPN
easat-5356	298	69	-	-	PUNCT
easat-5356	298	70	m	m	PROPN
easat-5356	298	71	92.9	92.9	NUM
easat-5356	298	72	90.8	90.8	NUM
easat-5356	298	73	83.5	83.5	NUM
easat-5356	298	74	yolov8l	yolov8l	NOUN
easat-5356	298	75	82.7	82.7	NUM
easat-5356	298	76	82.2	82.2	NUM
easat-5356	298	77	65.8	65.8	NUM
easat-5356	298	78	gcsem	gcsem	ADJ
easat-5356	298	79	-yolo	-yolo	PROPN
easat-5356	298	80	-	-	PUNCT
easat-5356	298	81	l	l	NOUN
easat-5356	298	82	94.3	94.3	NUM
easat-5356	298	83	93.0	93.0	NUM
easat-5356	298	84	84.7	84.7	NUM
easat-5356	298	85	853	853	NUM
easat-5356	298	86	edelweiss	edelweiss	PROPN
easat-5356	298	87	applied	apply	VERB
easat-5356	298	88	science	science	NOUN
easat-5356	298	89	and	and	CCONJ
easat-5356	298	90	technology	technology	NOUN
easat-5356	298	91	issn	issn	PROPN
easat-5356	298	92	:	:	PUNCT
easat-5356	298	93	2576	2576	NUM
easat-5356	298	94	-	-	SYM
easat-5356	298	95	8484	8484	NUM
easat-5356	298	96	vol	vol	NOUN
easat-5356	298	97	.	.	PROPN
easat-5356	299	1	9	9	NUM
easat-5356	299	2	,	,	PUNCT
easat-5356	299	3	no	no	INTJ
easat-5356	299	4	.	.	NOUN
easat-5356	300	1	3	3	NUM
easat-5356	300	2	:	:	PUNCT
easat-5356	300	3	840	840	NUM
easat-5356	300	4	-	-	SYM
easat-5356	300	5	855	855	NUM
easat-5356	300	6	,	,	PUNCT
easat-5356	300	7	2025	2025	NUM
easat-5356	300	8	doi	doi	NOUN
easat-5356	300	9	:	:	PUNCT
easat-5356	300	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	300	11	©	©	PROPN
easat-5356	300	12	2025	2025	NUM
easat-5356	300	13	by	by	ADP
easat-5356	300	14	the	the	DET
easat-5356	300	15	authors	author	NOUN
easat-5356	300	16	;	;	PUNCT
easat-5356	300	17	licensee	licensee	PROPN
easat-5356	300	18	learning	learn	VERB
easat-5356	300	19	gate	gate	PROPN
easat-5356	300	20	table	table	NOUN
easat-5356	300	21	2	2	NUM
easat-5356	300	22	shows	show	VERB
easat-5356	300	23	that	that	SCONJ
easat-5356	300	24	gcsem	gcsem	ADJ
easat-5356	300	25	-	-	PUNCT
easat-5356	300	26	yolo	yolo	ADJ
easat-5356	300	27	outperform	outperform	NOUN
easat-5356	300	28	yolov8	yolov8	NOUN
easat-5356	300	29	in	in	ADP
easat-5356	300	30	dealing	deal	VERB
easat-5356	300	31	with	with	ADP
easat-5356	300	32	large	large	ADJ
easat-5356	300	33	-	-	PUNCT
easat-5356	300	34	size	size	NOUN
easat-5356	300	35	objects	object	NOUN
easat-5356	300	36	.	.	PUNCT
easat-5356	301	1	in	in	ADP
easat-5356	301	2	addition	addition	NOUN
easat-5356	301	3	,	,	PUNCT
easat-5356	301	4	gcsem	gcsem	NOUN
easat-5356	301	5	-	-	PUNCT
easat-5356	301	6	yolo	yolo	NOUN
easat-5356	301	7	also	also	ADV
easat-5356	301	8	outperforms	outperform	VERB
easat-5356	301	9	yolov8	yolov8	NOUN
easat-5356	301	10	in	in	ADP
easat-5356	301	11	detecting	detect	VERB
easat-5356	301	12	hard	hard	ADJ
easat-5356	301	13	targets	target	NOUN
easat-5356	301	14	,	,	PUNCT
easat-5356	301	15	especially	especially	ADV
easat-5356	301	16	in	in	ADP
easat-5356	301	17	the	the	DET
easat-5356	301	18	l	l	NOUN
easat-5356	301	19	-	-	PUNCT
easat-5356	301	20	size	size	NOUN
easat-5356	301	21	model	model	NOUN
easat-5356	301	22	,	,	PUNCT
easat-5356	301	23	where	where	SCONJ
easat-5356	301	24	gcsem	gcsem	NOUN
easat-5356	301	25	-	-	PUNCT
easat-5356	301	26	yolo	yolo	PROPN
easat-5356	301	27	-	-	PUNCT
easat-5356	301	28	l	l	NOUN
easat-5356	301	29	outperforms	outperform	NOUN
easat-5356	301	30	yolov8l	yolov8l	NOUN
easat-5356	301	31	by	by	ADP
easat-5356	301	32	18%-21.4	18%-21.4	NOUN
easat-5356	301	33	%	%	NOUN
easat-5356	301	34	.	.	PUNCT
easat-5356	302	1	according	accord	VERB
easat-5356	302	2	to	to	ADP
easat-5356	302	3	these	these	DET
easat-5356	302	4	findings	finding	NOUN
easat-5356	302	5	,	,	PUNCT
easat-5356	302	6	gcsem	gcsem	ADJ
easat-5356	302	7	-	-	PUNCT
easat-5356	302	8	yolo	yolo	PROPN
easat-5356	302	9	offers	offer	VERB
easat-5356	302	10	notable	notable	ADJ
easat-5356	302	11	benefits	benefit	NOUN
easat-5356	302	12	when	when	SCONJ
easat-5356	302	13	it	it	PRON
easat-5356	302	14	comes	come	VERB
easat-5356	302	15	to	to	ADP
easat-5356	302	16	identifying	identify	VERB
easat-5356	302	17	tiny	tiny	ADJ
easat-5356	302	18	and	and	CCONJ
easat-5356	302	19	medium	medium	ADJ
easat-5356	302	20	-	-	PUNCT
easat-5356	302	21	sized	sized	ADJ
easat-5356	302	22	faces	face	NOUN
easat-5356	302	23	,	,	PUNCT
easat-5356	302	24	particularly	particularly	ADV
easat-5356	302	25	small	small	ADJ
easat-5356	302	26	faces	face	NOUN
easat-5356	302	27	.	.	PUNCT
easat-5356	303	1	6	6	X
easat-5356	303	2	.	.	X
easat-5356	303	3	ablation	ablation	NOUN
easat-5356	303	4	experiment	experiment	NOUN
easat-5356	303	5	this	this	DET
easat-5356	303	6	study	study	NOUN
easat-5356	303	7	carried	carry	VERB
easat-5356	303	8	out	out	ADP
easat-5356	303	9	a	a	DET
easat-5356	303	10	number	number	NOUN
easat-5356	303	11	of	of	ADP
easat-5356	303	12	compared	compare	VERB
easat-5356	303	13	experiments	experiment	NOUN
easat-5356	303	14	to	to	PART
easat-5356	303	15	evaluate	evaluate	VERB
easat-5356	303	16	the	the	DET
easat-5356	303	17	effects	effect	NOUN
easat-5356	303	18	of	of	ADP
easat-5356	303	19	including	include	VERB
easat-5356	303	20	the	the	DET
easat-5356	303	21	gcsem	gcsem	NOUN
easat-5356	303	22	module	module	NOUN
easat-5356	303	23	in	in	ADP
easat-5356	303	24	the	the	DET
easat-5356	303	25	model	model	NOUN
easat-5356	303	26	and	and	CCONJ
easat-5356	303	27	to	to	PART
easat-5356	303	28	make	make	VERB
easat-5356	303	29	ablation	ablation	NOUN
easat-5356	303	30	trials	trial	NOUN
easat-5356	303	31	easier	easy	ADJ
easat-5356	303	32	in	in	ADP
easat-5356	303	33	the	the	DET
easat-5356	303	34	future	future	NOUN
easat-5356	303	35	.	.	PUNCT
easat-5356	304	1	at	at	ADP
easat-5356	304	2	various	various	ADJ
easat-5356	304	3	backbone	backbone	NOUN
easat-5356	304	4	network	network	NOUN
easat-5356	304	5	tiers	tier	NOUN
easat-5356	304	6	,	,	PUNCT
easat-5356	304	7	the	the	DET
easat-5356	304	8	gcsem	gcsem	NOUN
easat-5356	304	9	module	module	NOUN
easat-5356	304	10	was	be	AUX
easat-5356	304	11	utilized	utilize	VERB
easat-5356	304	12	in	in	ADP
easat-5356	304	13	place	place	NOUN
easat-5356	304	14	of	of	ADP
easat-5356	304	15	the	the	DET
easat-5356	304	16	conv	conv	NOUN
easat-5356	304	17	module	module	NOUN
easat-5356	304	18	,	,	PUNCT
easat-5356	304	19	with	with	ADP
easat-5356	304	20	yolov8n	yolov8n	NOUN
easat-5356	304	21	serving	serve	VERB
easat-5356	304	22	as	as	ADP
easat-5356	304	23	the	the	DET
easat-5356	304	24	baseline	baseline	PROPN
easat-5356	304	25	model	model	NOUN
easat-5356	304	26	.	.	PUNCT
easat-5356	305	1	table	table	NOUN
easat-5356	305	2	3	3	NUM
easat-5356	305	3	displays	display	VERB
easat-5356	305	4	the	the	DET
easat-5356	305	5	experimental	experimental	ADJ
easat-5356	305	6	outcomes	outcome	NOUN
easat-5356	305	7	,	,	PUNCT
easat-5356	305	8	with	with	ADP
easat-5356	305	9	the	the	DET
easat-5356	305	10	"	"	PUNCT
easat-5356	305	11	+	+	ADJ
easat-5356	305	12	"	"	PUNCT
easat-5356	305	13	sign	sign	NOUN
easat-5356	305	14	denoting	denote	VERB
easat-5356	305	15	modifications	modification	NOUN
easat-5356	305	16	to	to	ADP
easat-5356	305	17	the	the	DET
easat-5356	305	18	baseline	baseline	PROPN
easat-5356	305	19	model	model	NOUN
easat-5356	305	20	.	.	PUNCT
easat-5356	306	1	according	accord	VERB
easat-5356	306	2	to	to	ADP
easat-5356	306	3	the	the	DET
easat-5356	306	4	findings	finding	NOUN
easat-5356	306	5	,	,	PUNCT
easat-5356	306	6	gcim	gcim	PROPN
easat-5356	306	7	-	-	PUNCT
easat-5356	306	8	yolo	yolo	PROPN
easat-5356	306	9	-	-	PUNCT
easat-5356	306	10	n	n	CCONJ
easat-5356	306	11	's	's	PART
easat-5356	306	12	map50	map50	PROPN
easat-5356	306	13	was	be	AUX
easat-5356	306	14	10	10	NUM
easat-5356	306	15	percentage	percentage	NOUN
easat-5356	306	16	points	point	NOUN
easat-5356	306	17	greater	great	ADJ
easat-5356	306	18	than	than	ADP
easat-5356	306	19	the	the	DET
easat-5356	306	20	original	original	ADJ
easat-5356	306	21	yolov8n	yolov8n	NOUN
easat-5356	306	22	and	and	CCONJ
easat-5356	306	23	3	3	NUM
easat-5356	306	24	percentage	percentage	NOUN
easat-5356	306	25	points	point	NOUN
easat-5356	306	26	higher	high	ADJ
easat-5356	306	27	than	than	ADP
easat-5356	306	28	yolov8n	yolov8n	NOUN
easat-5356	306	29	.	.	PUNCT
easat-5356	307	1	the	the	DET
easat-5356	307	2	effectiveness	effectiveness	NOUN
easat-5356	307	3	of	of	ADP
easat-5356	307	4	the	the	DET
easat-5356	307	5	gcsem	gcsem	NOUN
easat-5356	307	6	module	module	NOUN
easat-5356	307	7	was	be	AUX
easat-5356	307	8	validated	validate	VERB
easat-5356	307	9	by	by	ADP
easat-5356	307	10	ablation	ablation	NOUN
easat-5356	307	11	tests	test	NOUN
easat-5356	307	12	on	on	ADP
easat-5356	307	13	the	the	DET
easat-5356	307	14	wider	wide	ADJ
easat-5356	307	15	face	face	NOUN
easat-5356	307	16	dataset	dataset	NOUN
easat-5356	307	17	,	,	PUNCT
easat-5356	307	18	which	which	PRON
easat-5356	307	19	showed	show	VERB
easat-5356	307	20	that	that	SCONJ
easat-5356	307	21	the	the	DET
easat-5356	307	22	suggested	suggest	VERB
easat-5356	307	23	enhancement	enhancement	NOUN
easat-5356	307	24	significantly	significantly	ADV
easat-5356	307	25	increased	increase	VERB
easat-5356	307	26	the	the	DET
easat-5356	307	27	accuracy	accuracy	NOUN
easat-5356	307	28	of	of	ADP
easat-5356	307	29	face	face	NOUN
easat-5356	307	30	detection	detection	NOUN
easat-5356	307	31	.	.	PUNCT
easat-5356	308	1	a	a	PRON
easat-5356	308	2	:	:	PUNCT
easat-5356	308	3	instead	instead	ADV
easat-5356	308	4	of	of	ADP
easat-5356	308	5	using	use	VERB
easat-5356	308	6	the	the	DET
easat-5356	308	7	original	original	ADJ
easat-5356	308	8	iou	iou	NOUN
easat-5356	308	9	loss	loss	NOUN
easat-5356	308	10	function	function	NOUN
easat-5356	308	11	,	,	PUNCT
easat-5356	308	12	the	the	DET
easat-5356	308	13	wiouv3	wiouv3	NOUN
easat-5356	308	14	loss	loss	NOUN
easat-5356	308	15	function	function	NOUN
easat-5356	308	16	is	be	AUX
easat-5356	308	17	used	use	VERB
easat-5356	308	18	.	.	PUNCT
easat-5356	309	1	add	add	VERB
easat-5356	309	2	atw	atw	PROPN
easat-5356	309	3	module	module	NOUN
easat-5356	309	4	before	before	ADP
easat-5356	309	5	computing	compute	VERB
easat-5356	309	6	wiouv3	wiouv3	NOUN
easat-5356	309	7	loss	loss	NOUN
easat-5356	309	8	.	.	PUNCT
easat-5356	310	1	b	b	X
easat-5356	310	2	:	:	PUNCT
easat-5356	310	3	add	add	VERB
easat-5356	310	4	the	the	DET
easat-5356	310	5	xsmall	xsmall	NOUN
easat-5356	310	6	-	-	PUNCT
easat-5356	310	7	detection	detection	NOUN
easat-5356	310	8	head	head	NOUN
easat-5356	310	9	.	.	PUNCT
easat-5356	311	1	c	c	X
easat-5356	311	2	:	:	PUNCT
easat-5356	311	3	the	the	DET
easat-5356	311	4	gcsem	gcsem	NOUN
easat-5356	311	5	module	module	NOUN
easat-5356	311	6	is	be	AUX
easat-5356	311	7	utilized	utilize	VERB
easat-5356	311	8	as	as	ADP
easat-5356	311	9	the	the	DET
easat-5356	311	10	feature	feature	NOUN
easat-5356	311	11	extraction	extraction	NOUN
easat-5356	311	12	module	module	NOUN
easat-5356	311	13	in	in	ADP
easat-5356	311	14	place	place	NOUN
easat-5356	311	15	of	of	ADP
easat-5356	311	16	the	the	DET
easat-5356	311	17	conv	conv	NOUN
easat-5356	311	18	module	module	NOUN
easat-5356	311	19	.	.	PUNCT
easat-5356	312	1	table	table	NOUN
easat-5356	312	2	3	3	NUM
easat-5356	312	3	.	.	PUNCT
easat-5356	312	4	ablation	ablation	NOUN
easat-5356	312	5	study	study	NOUN
easat-5356	312	6	results	result	VERB
easat-5356	312	7	on	on	ADP
easat-5356	312	8	the	the	DET
easat-5356	312	9	wider	wide	ADJ
easat-5356	312	10	face	face	NOUN
easat-5356	312	11	dataset	dataset	NOUN
easat-5356	312	12	.	.	PUNCT
easat-5356	313	1	network	network	NOUN
easat-5356	313	2	precision/%	precision/%	X
easat-5356	313	3	recall/%	recall/%	ADV
easat-5356	313	4	map0.5/%	map0.5/%	VERB
easat-5356	313	5	map0.5:0.95/%	map0.5:0.95/%	ADJ
easat-5356	313	6	yolov8s	yolov8	NOUN
easat-5356	313	7	84.7	84.7	NUM
easat-5356	313	8	58.0	58.0	NUM
easat-5356	313	9	66.7	66.7	NUM
easat-5356	313	10	36.6	36.6	NUM
easat-5356	313	11	+	+	NOUN
easat-5356	313	12	a	a	DET
easat-5356	313	13	85.9	85.9	NUM
easat-5356	313	14	60.8	60.8	NUM
easat-5356	313	15	71.2	71.2	NUM
easat-5356	313	16	39.0	39.0	NUM
easat-5356	313	17	+	+	PROPN
easat-5356	313	18	a+b	a+b	PROPN
easat-5356	313	19	86.4	86.4	NUM
easat-5356	313	20	62.1	62.1	NUM
easat-5356	313	21	72.9	72.9	NUM
easat-5356	313	22	41.1	41.1	NUM
easat-5356	313	23	+	+	NOUN
easat-5356	313	24	a+b+c	a+b+c	ADV
easat-5356	313	25	89.3	89.3	NUM
easat-5356	313	26	69.5	69.5	NUM
easat-5356	313	27	78.8	78.8	NUM
easat-5356	313	28	45.1	45.1	NUM
easat-5356	313	29	as	as	SCONJ
easat-5356	313	30	observed	observe	VERB
easat-5356	313	31	from	from	ADP
easat-5356	313	32	the	the	DET
easat-5356	313	33	experimental	experimental	ADJ
easat-5356	313	34	results	result	NOUN
easat-5356	313	35	in	in	ADP
easat-5356	313	36	table	table	NOUN
easat-5356	313	37	3	3	NUM
easat-5356	313	38	,	,	PUNCT
easat-5356	313	39	each	each	DET
easat-5356	313	40	improvement	improvement	NOUN
easat-5356	313	41	strategy	strategy	NOUN
easat-5356	313	42	contributes	contribute	VERB
easat-5356	313	43	to	to	ADP
easat-5356	313	44	varying	vary	VERB
easat-5356	313	45	degrees	degree	NOUN
easat-5356	313	46	of	of	ADP
easat-5356	313	47	enhancement	enhancement	NOUN
easat-5356	313	48	in	in	ADP
easat-5356	313	49	detection	detection	NOUN
easat-5356	313	50	performance	performance	NOUN
easat-5356	313	51	when	when	SCONJ
easat-5356	313	52	applied	apply	VERB
easat-5356	313	53	to	to	ADP
easat-5356	313	54	the	the	DET
easat-5356	313	55	baseline	baseline	PROPN
easat-5356	313	56	model	model	NOUN
easat-5356	313	57	.	.	PUNCT
easat-5356	314	1	the	the	DET
easat-5356	314	2	introduction	introduction	NOUN
easat-5356	314	3	of	of	ADP
easat-5356	314	4	wiouv3	wiouv3	NOUN
easat-5356	314	5	in	in	ADP
easat-5356	314	6	the	the	DET
easat-5356	314	7	prediction	prediction	NOUN
easat-5356	314	8	box	box	NOUN
easat-5356	314	9	regression	regression	NOUN
easat-5356	314	10	loss	loss	NOUN
easat-5356	314	11	,	,	PUNCT
easat-5356	314	12	combined	combine	VERB
easat-5356	314	13	with	with	ADP
easat-5356	314	14	the	the	DET
easat-5356	314	15	atw	atw	PROPN
easat-5356	314	16	module	module	NOUN
easat-5356	314	17	,	,	PUNCT
easat-5356	314	18	improves	improve	VERB
easat-5356	314	19	the	the	DET
easat-5356	314	20	model	model	NOUN
easat-5356	314	21	's	's	PART
easat-5356	314	22	localization	localization	NOUN
easat-5356	314	23	ability	ability	NOUN
easat-5356	314	24	through	through	ADP
easat-5356	314	25	a	a	DET
easat-5356	314	26	smarter	smart	ADJ
easat-5356	314	27	sample	sample	NOUN
easat-5356	314	28	allocation	allocation	NOUN
easat-5356	314	29	strategy	strategy	NOUN
easat-5356	314	30	,	,	PUNCT
easat-5356	314	31	leading	lead	VERB
easat-5356	314	32	to	to	ADP
easat-5356	314	33	a	a	DET
easat-5356	314	34	4.5	4.5	NUM
easat-5356	314	35	%	%	NOUN
easat-5356	314	36	increase	increase	NOUN
easat-5356	314	37	in	in	ADP
easat-5356	314	38	map50	map50	PROPN
easat-5356	314	39	.	.	PUNCT
easat-5356	315	1	adding	add	VERB
easat-5356	315	2	the	the	DET
easat-5356	315	3	xsmall	xsmall	NOUN
easat-5356	315	4	detection	detection	NOUN
easat-5356	315	5	head	head	NOUN
easat-5356	315	6	to	to	ADP
easat-5356	315	7	the	the	DET
easat-5356	315	8	detection	detection	NOUN
easat-5356	315	9	head	head	NOUN
easat-5356	315	10	architecture	architecture	NOUN
easat-5356	315	11	improved	improve	VERB
easat-5356	315	12	the	the	DET
easat-5356	315	13	map50	map50	NOUN
easat-5356	315	14	by	by	ADP
easat-5356	315	15	1.7	1.7	NUM
easat-5356	315	16	%	%	NOUN
easat-5356	315	17	.	.	PUNCT
easat-5356	316	1	an	an	DET
easat-5356	316	2	effective	effective	ADJ
easat-5356	316	3	attention	attention	NOUN
easat-5356	316	4	mechanism	mechanism	NOUN
easat-5356	316	5	that	that	PRON
easat-5356	316	6	improves	improve	VERB
easat-5356	316	7	concentration	concentration	NOUN
easat-5356	316	8	on	on	ADP
easat-5356	316	9	important	important	ADJ
easat-5356	316	10	information	information	NOUN
easat-5356	316	11	in	in	ADP
easat-5356	316	12	the	the	DET
easat-5356	316	13	feature	feature	NOUN
easat-5356	316	14	map	map	NOUN
easat-5356	316	15	is	be	AUX
easat-5356	316	16	integrated	integrate	VERB
easat-5356	316	17	into	into	ADP
easat-5356	316	18	the	the	DET
easat-5356	316	19	gcsem	gcsem	NOUN
easat-5356	316	20	module	module	NOUN
easat-5356	316	21	,	,	PUNCT
easat-5356	316	22	which	which	PRON
easat-5356	316	23	takes	take	VERB
easat-5356	316	24	the	the	DET
easat-5356	316	25	place	place	NOUN
easat-5356	316	26	of	of	ADP
easat-5356	316	27	the	the	DET
easat-5356	316	28	original	original	ADJ
easat-5356	316	29	conv	conv	NOUN
easat-5356	316	30	module	module	NOUN
easat-5356	316	31	in	in	ADP
easat-5356	316	32	the	the	DET
easat-5356	316	33	backbone	backbone	NOUN
easat-5356	316	34	.	.	PUNCT
easat-5356	317	1	as	as	ADP
easat-5356	317	2	a	a	DET
easat-5356	317	3	result	result	NOUN
easat-5356	317	4	,	,	PUNCT
easat-5356	317	5	map50	map50	PROPN
easat-5356	317	6	significantly	significantly	ADV
easat-5356	317	7	improves	improve	VERB
easat-5356	317	8	by	by	ADP
easat-5356	317	9	5.9	5.9	NUM
easat-5356	317	10	%	%	NOUN
easat-5356	317	11	.	.	PUNCT
easat-5356	318	1	since	since	SCONJ
easat-5356	318	2	there	there	PRON
easat-5356	318	3	are	be	VERB
easat-5356	318	4	a	a	DET
easat-5356	318	5	lot	lot	NOUN
easat-5356	318	6	of	of	ADP
easat-5356	318	7	little	little	ADJ
easat-5356	318	8	targets	target	NOUN
easat-5356	318	9	in	in	ADP
easat-5356	318	10	the	the	DET
easat-5356	318	11	dataset	dataset	NOUN
easat-5356	318	12	,	,	PUNCT
easat-5356	318	13	the	the	DET
easat-5356	318	14	gcsem	gcsem	NOUN
easat-5356	318	15	module	module	NOUN
easat-5356	318	16	's	's	PART
easat-5356	318	17	fusion	fusion	NOUN
easat-5356	318	18	of	of	ADP
easat-5356	318	19	shallow	shallow	ADJ
easat-5356	318	20	feature	feature	NOUN
easat-5356	318	21	information	information	NOUN
easat-5356	318	22	successfully	successfully	ADV
easat-5356	318	23	lowers	lower	VERB
easat-5356	318	24	the	the	DET
easat-5356	318	25	rate	rate	NOUN
easat-5356	318	26	at	at	ADP
easat-5356	318	27	which	which	PRON
easat-5356	318	28	small	small	ADJ
easat-5356	318	29	targets	target	NOUN
easat-5356	318	30	leak	leak	VERB
easat-5356	318	31	out	out	ADP
easat-5356	318	32	.	.	PUNCT
easat-5356	319	1	all	all	DET
easat-5356	319	2	things	thing	NOUN
easat-5356	319	3	considered	consider	VERB
easat-5356	319	4	,	,	PUNCT
easat-5356	319	5	our	our	PRON
easat-5356	319	6	suggested	suggest	VERB
easat-5356	319	7	feature	feature	NOUN
easat-5356	319	8	fusion	fusion	NOUN
easat-5356	319	9	network	network	NOUN
easat-5356	319	10	shows	show	VERB
easat-5356	319	11	promise	promise	NOUN
easat-5356	319	12	in	in	ADP
easat-5356	319	13	significantly	significantly	ADV
easat-5356	319	14	enhancing	enhance	VERB
easat-5356	319	15	the	the	DET
easat-5356	319	16	model	model	NOUN
easat-5356	319	17	's	's	PART
easat-5356	319	18	detection	detection	NOUN
easat-5356	319	19	performance	performance	NOUN
easat-5356	319	20	,	,	PUNCT
easat-5356	319	21	especially	especially	ADV
easat-5356	319	22	for	for	ADP
easat-5356	319	23	small	small	ADJ
easat-5356	319	24	-	-	PUNCT
easat-5356	319	25	scale	scale	NOUN
easat-5356	319	26	faces	face	NOUN
easat-5356	319	27	.	.	PUNCT
easat-5356	320	1	7	7	X
easat-5356	320	2	.	.	X
easat-5356	320	3	conclusions	conclusion	NOUN
easat-5356	320	4	in	in	ADP
easat-5356	320	5	order	order	NOUN
easat-5356	320	6	to	to	PART
easat-5356	320	7	tackle	tackle	VERB
easat-5356	320	8	the	the	DET
easat-5356	320	9	problem	problem	NOUN
easat-5356	320	10	of	of	ADP
easat-5356	320	11	small	small	ADJ
easat-5356	320	12	target	target	NOUN
easat-5356	320	13	recognition	recognition	NOUN
easat-5356	320	14	in	in	ADP
easat-5356	320	15	face	face	NOUN
easat-5356	320	16	detection	detection	NOUN
easat-5356	320	17	tasks	task	NOUN
easat-5356	320	18	,	,	PUNCT
easat-5356	320	19	we	we	PRON
easat-5356	320	20	suggest	suggest	VERB
easat-5356	320	21	the	the	DET
easat-5356	320	22	gcsem	gcsem	ADJ
easat-5356	320	23	-	-	PUNCT
easat-5356	320	24	yolo	yolo	PROPN
easat-5356	320	25	model	model	NOUN
easat-5356	320	26	,	,	PUNCT
easat-5356	320	27	which	which	PRON
easat-5356	320	28	is	be	AUX
easat-5356	320	29	based	base	VERB
easat-5356	320	30	on	on	ADP
easat-5356	320	31	yolov8	yolov8	PROPN
easat-5356	320	32	.	.	PUNCT
easat-5356	321	1	the	the	DET
easat-5356	321	2	model	model	NOUN
easat-5356	321	3	successfully	successfully	ADV
easat-5356	321	4	addresses	address	VERB
easat-5356	321	5	typical	typical	ADJ
easat-5356	321	6	face	face	NOUN
easat-5356	321	7	detection	detection	NOUN
easat-5356	321	8	problems	problem	NOUN
easat-5356	321	9	.	.	PUNCT
easat-5356	322	1	the	the	DET
easat-5356	322	2	new	new	ADJ
easat-5356	322	3	gcsem	gcsem	ADJ
easat-5356	322	4	neck	neck	NOUN
easat-5356	322	5	structure	structure	NOUN
easat-5356	322	6	,	,	PUNCT
easat-5356	322	7	which	which	PRON
easat-5356	322	8	makes	make	VERB
easat-5356	322	9	multi	multi	ADJ
easat-5356	322	10	-	-	ADJ
easat-5356	322	11	scale	scale	ADJ
easat-5356	322	12	feature	feature	NOUN
easat-5356	322	13	fusion	fusion	NOUN
easat-5356	322	14	easier	easy	ADJ
easat-5356	322	15	,	,	PUNCT
easat-5356	322	16	is	be	AUX
easat-5356	322	17	a	a	DET
easat-5356	322	18	significant	significant	ADJ
easat-5356	322	19	algorithmic	algorithmic	ADJ
easat-5356	322	20	advance	advance	NOUN
easat-5356	322	21	.	.	PUNCT
easat-5356	323	1	by	by	ADP
easat-5356	323	2	achieving	achieve	VERB
easat-5356	323	3	a	a	DET
easat-5356	323	4	balanced	balanced	ADJ
easat-5356	323	5	integration	integration	NOUN
easat-5356	323	6	of	of	ADP
easat-5356	323	7	spatial	spatial	ADJ
easat-5356	323	8	and	and	CCONJ
easat-5356	323	9	semantic	semantic	ADJ
easat-5356	323	10	information	information	NOUN
easat-5356	323	11	in	in	ADP
easat-5356	323	12	the	the	DET
easat-5356	323	13	feature	feature	NOUN
easat-5356	323	14	map	map	NOUN
easat-5356	323	15	,	,	PUNCT
easat-5356	323	16	this	this	DET
easat-5356	323	17	structure	structure	NOUN
easat-5356	323	18	enhances	enhance	VERB
easat-5356	323	19	the	the	DET
easat-5356	323	20	model	model	NOUN
easat-5356	323	21	's	's	PART
easat-5356	323	22	capacity	capacity	NOUN
easat-5356	323	23	to	to	PART
easat-5356	323	24	identify	identify	VERB
easat-5356	323	25	faces	face	NOUN
easat-5356	323	26	at	at	ADP
easat-5356	323	27	various	various	ADJ
easat-5356	323	28	scales	scale	NOUN
easat-5356	323	29	.	.	PUNCT
easat-5356	324	1	gcsem	gcsem	NOUN
easat-5356	324	2	decreases	decrease	VERB
easat-5356	324	3	the	the	DET
easat-5356	324	4	sparsity	sparsity	NOUN
easat-5356	324	5	of	of	ADP
easat-5356	324	6	spatial	spatial	ADJ
easat-5356	324	7	information	information	NOUN
easat-5356	324	8	brought	bring	VERB
easat-5356	324	9	on	on	ADP
easat-5356	324	10	by	by	ADP
easat-5356	324	11	down	down	ADP
easat-5356	324	12	sampling	sample	VERB
easat-5356	324	13	and	and	CCONJ
easat-5356	324	14	improves	improve	VERB
easat-5356	324	15	feature	feature	NOUN
easat-5356	324	16	extraction	extraction	NOUN
easat-5356	324	17	performance	performance	NOUN
easat-5356	324	18	as	as	ADP
easat-5356	324	19	compared	compare	VERB
easat-5356	324	20	to	to	ADP
easat-5356	324	21	the	the	DET
easat-5356	324	22	conventional	conventional	ADJ
easat-5356	324	23	convolutional	convolutional	ADJ
easat-5356	324	24	layer	layer	NOUN
easat-5356	324	25	.	.	PUNCT
easat-5356	325	1	furthermore	furthermore	ADV
easat-5356	325	2	,	,	PUNCT
easat-5356	325	3	a	a	DET
easat-5356	325	4	key	key	ADJ
easat-5356	325	5	factor	factor	NOUN
easat-5356	325	6	in	in	ADP
easat-5356	325	7	enhancing	enhance	VERB
easat-5356	325	8	detection	detection	NOUN
easat-5356	325	9	performance	performance	NOUN
easat-5356	325	10	for	for	ADP
easat-5356	325	11	small	small	ADJ
easat-5356	325	12	faces	face	NOUN
easat-5356	325	13	is	be	AUX
easat-5356	325	14	the	the	DET
easat-5356	325	15	xsmall	xsmall	NOUN
easat-5356	325	16	detection	detection	NOUN
easat-5356	325	17	head	head	NOUN
easat-5356	325	18	,	,	PUNCT
easat-5356	325	19	which	which	PRON
easat-5356	325	20	is	be	AUX
easat-5356	325	21	854	854	NUM
easat-5356	325	22	edelweiss	edelweiss	PROPN
easat-5356	325	23	applied	apply	VERB
easat-5356	325	24	science	science	NOUN
easat-5356	325	25	and	and	CCONJ
easat-5356	325	26	technology	technology	NOUN
easat-5356	325	27	issn	issn	PROPN
easat-5356	325	28	:	:	PUNCT
easat-5356	325	29	2576	2576	NUM
easat-5356	325	30	-	-	SYM
easat-5356	325	31	8484	8484	NUM
easat-5356	325	32	vol	vol	NOUN
easat-5356	325	33	.	.	PROPN
easat-5356	326	1	9	9	NUM
easat-5356	326	2	,	,	PUNCT
easat-5356	326	3	no	no	INTJ
easat-5356	326	4	.	.	NOUN
easat-5356	327	1	3	3	NUM
easat-5356	327	2	:	:	PUNCT
easat-5356	327	3	840	840	NUM
easat-5356	327	4	-	-	SYM
easat-5356	327	5	855	855	NUM
easat-5356	327	6	,	,	PUNCT
easat-5356	327	7	2025	2025	NUM
easat-5356	327	8	doi	doi	NOUN
easat-5356	327	9	:	:	PUNCT
easat-5356	327	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	327	11	©	©	PROPN
easat-5356	327	12	2025	2025	NUM
easat-5356	327	13	by	by	ADP
easat-5356	327	14	the	the	DET
easat-5356	327	15	authors	author	NOUN
easat-5356	327	16	;	;	PUNCT
easat-5356	327	17	licensee	licensee	PROPN
easat-5356	327	18	learning	learning	NOUN
easat-5356	327	19	gate	gate	NOUN
easat-5356	327	20	especially	especially	ADV
easat-5356	327	21	made	make	VERB
easat-5356	327	22	for	for	ADP
easat-5356	327	23	small	small	ADJ
easat-5356	327	24	target	target	NOUN
easat-5356	327	25	face	face	NOUN
easat-5356	327	26	detection	detection	NOUN
easat-5356	327	27	.	.	PUNCT
easat-5356	328	1	the	the	DET
easat-5356	328	2	computational	computational	ADJ
easat-5356	328	3	complexity	complexity	NOUN
easat-5356	328	4	of	of	ADP
easat-5356	328	5	the	the	DET
easat-5356	328	6	model	model	NOUN
easat-5356	328	7	is	be	AUX
easat-5356	328	8	further	far	ADV
easat-5356	328	9	decreased	decrease	VERB
easat-5356	328	10	,	,	PUNCT
easat-5356	328	11	increasing	increase	VERB
easat-5356	328	12	its	its	PRON
easat-5356	328	13	efficiency	efficiency	NOUN
easat-5356	328	14	,	,	PUNCT
easat-5356	328	15	through	through	ADP
easat-5356	328	16	the	the	DET
easat-5356	328	17	application	application	NOUN
easat-5356	328	18	of	of	ADP
easat-5356	328	19	wiouv3	wiouv3	NOUN
easat-5356	328	20	loss	loss	NOUN
easat-5356	328	21	and	and	CCONJ
easat-5356	328	22	a	a	DET
easat-5356	328	23	novel	novel	ADJ
easat-5356	328	24	anchor	anchor	NOUN
easat-5356	328	25	frame	frame	NOUN
easat-5356	328	26	filtering	filter	VERB
easat-5356	328	27	approach	approach	NOUN
easat-5356	328	28	.	.	PUNCT
easat-5356	329	1	gcsem	gcsem	ADJ
easat-5356	329	2	-	-	PUNCT
easat-5356	329	3	yolo	yolo	PROPN
easat-5356	329	4	performs	perform	VERB
easat-5356	329	5	better	well	ADJ
easat-5356	329	6	than	than	ADP
easat-5356	329	7	yolov8	yolov8	NOUN
easat-5356	329	8	in	in	ADP
easat-5356	329	9	terms	term	NOUN
easat-5356	329	10	of	of	ADP
easat-5356	329	11	detection	detection	NOUN
easat-5356	329	12	accuracy	accuracy	NOUN
easat-5356	329	13	on	on	ADP
easat-5356	329	14	practically	practically	ADV
easat-5356	329	15	all	all	PRON
easat-5356	329	16	scales	scale	NOUN
easat-5356	329	17	,	,	PUNCT
easat-5356	329	18	according	accord	VERB
easat-5356	329	19	to	to	ADP
easat-5356	329	20	experimental	experimental	ADJ
easat-5356	329	21	results	result	NOUN
easat-5356	329	22	on	on	ADP
easat-5356	329	23	the	the	DET
easat-5356	329	24	wider	wide	ADJ
easat-5356	329	25	face	face	NOUN
easat-5356	329	26	dataset	dataset	NOUN
easat-5356	329	27	.	.	PUNCT
easat-5356	330	1	these	these	DET
easat-5356	330	2	findings	finding	NOUN
easat-5356	330	3	indicate	indicate	VERB
easat-5356	330	4	how	how	SCONJ
easat-5356	330	5	well	well	ADV
easat-5356	330	6	the	the	DET
easat-5356	330	7	gcsem	gcsem	ADJ
easat-5356	330	8	-	-	PUNCT
easat-5356	330	9	yolo	yolo	PROPN
easat-5356	330	10	model	model	NOUN
easat-5356	330	11	works	work	VERB
easat-5356	330	12	to	to	PART
easat-5356	330	13	improve	improve	VERB
easat-5356	330	14	tiny	tiny	ADJ
easat-5356	330	15	target	target	NOUN
easat-5356	330	16	detection	detection	NOUN
easat-5356	330	17	for	for	ADP
easat-5356	330	18	facial	facial	ADJ
easat-5356	330	19	photos	photo	NOUN
easat-5356	330	20	.	.	PUNCT
easat-5356	331	1	transparency	transparency	NOUN
easat-5356	331	2	:	:	PUNCT
easat-5356	331	3	the	the	DET
easat-5356	331	4	authors	author	NOUN
easat-5356	331	5	confirm	confirm	VERB
easat-5356	331	6	that	that	SCONJ
easat-5356	331	7	the	the	DET
easat-5356	331	8	manuscript	manuscript	NOUN
easat-5356	331	9	is	be	AUX
easat-5356	331	10	an	an	DET
easat-5356	331	11	honest	honest	ADJ
easat-5356	331	12	,	,	PUNCT
easat-5356	331	13	accurate	accurate	ADJ
easat-5356	331	14	,	,	PUNCT
easat-5356	331	15	and	and	CCONJ
easat-5356	331	16	transparent	transparent	ADJ
easat-5356	331	17	account	account	NOUN
easat-5356	331	18	of	of	ADP
easat-5356	331	19	the	the	DET
easat-5356	331	20	study	study	NOUN
easat-5356	331	21	;	;	PUNCT
easat-5356	331	22	that	that	SCONJ
easat-5356	331	23	no	no	DET
easat-5356	331	24	vital	vital	ADJ
easat-5356	331	25	features	feature	NOUN
easat-5356	331	26	of	of	ADP
easat-5356	331	27	the	the	DET
easat-5356	331	28	study	study	NOUN
easat-5356	331	29	have	have	AUX
easat-5356	331	30	been	be	AUX
easat-5356	331	31	omitted	omit	VERB
easat-5356	331	32	;	;	PUNCT
easat-5356	331	33	and	and	CCONJ
easat-5356	331	34	that	that	SCONJ
easat-5356	331	35	any	any	DET
easat-5356	331	36	discrepancies	discrepancy	NOUN
easat-5356	331	37	from	from	ADP
easat-5356	331	38	the	the	DET
easat-5356	331	39	study	study	NOUN
easat-5356	331	40	as	as	SCONJ
easat-5356	331	41	planned	plan	VERB
easat-5356	331	42	have	have	AUX
easat-5356	331	43	been	be	AUX
easat-5356	331	44	explained	explain	VERB
easat-5356	331	45	.	.	PUNCT
easat-5356	332	1	this	this	DET
easat-5356	332	2	study	study	NOUN
easat-5356	332	3	followed	follow	VERB
easat-5356	332	4	all	all	DET
easat-5356	332	5	ethical	ethical	ADJ
easat-5356	332	6	practices	practice	NOUN
easat-5356	332	7	during	during	ADP
easat-5356	332	8	writing	writing	NOUN
easat-5356	332	9	.	.	PUNCT
easat-5356	333	1	copyright	copyright	NOUN
easat-5356	333	2	:	:	PUNCT
easat-5356	333	3	©	©	PROPN
easat-5356	333	4	2025	2025	NUM
easat-5356	333	5	by	by	ADP
easat-5356	333	6	the	the	DET
easat-5356	333	7	authors	author	NOUN
easat-5356	333	8	.	.	PUNCT
easat-5356	334	1	this	this	DET
easat-5356	334	2	open	open	ADJ
easat-5356	334	3	-	-	PUNCT
easat-5356	334	4	access	access	NOUN
easat-5356	334	5	article	article	NOUN
easat-5356	334	6	is	be	AUX
easat-5356	334	7	distributed	distribute	VERB
easat-5356	334	8	under	under	ADP
easat-5356	334	9	the	the	DET
easat-5356	334	10	terms	term	NOUN
easat-5356	334	11	and	and	CCONJ
easat-5356	334	12	conditions	condition	NOUN
easat-5356	334	13	of	of	ADP
easat-5356	334	14	the	the	DET
easat-5356	334	15	creative	creative	ADJ
easat-5356	334	16	commons	common	NOUN
easat-5356	334	17	attribution	attribution	NOUN
easat-5356	334	18	(	(	PUNCT
easat-5356	334	19	cc	cc	NOUN
easat-5356	334	20	by	by	ADP
easat-5356	334	21	)	)	PUNCT
easat-5356	334	22	license	license	NOUN
easat-5356	334	23	(	(	PUNCT
easat-5356	334	24	https://creativecommons.org/licenses/by/4.0/	https://creativecommons.org/licenses/by/4.0/	PROPN
easat-5356	334	25	)	)	PUNCT
easat-5356	334	26	.	.	PUNCT
easat-5356	335	1	references	reference	NOUN
easat-5356	335	2	[	[	X
easat-5356	335	3	1	1	X
easat-5356	335	4	]	]	PUNCT
easat-5356	335	5	j.	j.	PROPN
easat-5356	335	6	meher	meher	PROPN
easat-5356	335	7	,	,	PUNCT
easat-5356	335	8	h.	h.	PROPN
easat-5356	335	9	allende	allende	PROPN
easat-5356	335	10	-	-	PUNCT
easat-5356	335	11	cid	cid	PROPN
easat-5356	335	12	,	,	PUNCT
easat-5356	335	13	and	and	CCONJ
easat-5356	335	14	t.	t.	PROPN
easat-5356	335	15	e.	e.	PROPN
easat-5356	335	16	nordling	nordling	PROPN
easat-5356	335	17	,	,	PUNCT
easat-5356	335	18	"	"	PUNCT
easat-5356	335	19	a	a	DET
easat-5356	335	20	survey	survey	NOUN
easat-5356	335	21	and	and	CCONJ
easat-5356	335	22	classification	classification	NOUN
easat-5356	335	23	of	of	ADP
easat-5356	335	24	face	face	NOUN
easat-5356	335	25	alignment	alignment	NOUN
easat-5356	335	26	methods	method	NOUN
easat-5356	335	27	based	base	VERB
easat-5356	335	28	on	on	ADP
easat-5356	335	29	face	face	NOUN
easat-5356	335	30	models	model	NOUN
easat-5356	335	31	,	,	PUNCT
easat-5356	335	32	"	"	PUNCT
easat-5356	335	33	arxiv	arxiv	PROPN
easat-5356	335	34	preprint	preprint	VERB
easat-5356	335	35	arxiv:2311.03082	arxiv:2311.03082	PROPN
easat-5356	335	36	,	,	PUNCT
easat-5356	335	37	2023	2023	NUM
easat-5356	335	38	.	.	PUNCT
easat-5356	336	1	[	[	X
easat-5356	336	2	2	2	NUM
easat-5356	336	3	]	]	PUNCT
easat-5356	336	4	a.	a.	NOUN
easat-5356	336	5	dey	dey	PROPN
easat-5356	336	6	,	,	PUNCT
easat-5356	336	7	t.	t.	PROPN
easat-5356	336	8	senapati	senapati	PROPN
easat-5356	336	9	,	,	PUNCT
easat-5356	336	10	m.	m.	NOUN
easat-5356	336	11	pal	pal	NOUN
easat-5356	336	12	,	,	PUNCT
easat-5356	336	13	and	and	CCONJ
easat-5356	336	14	g.	g.	PROPN
easat-5356	336	15	chen	chen	PROPN
easat-5356	336	16	,	,	PUNCT
easat-5356	336	17	"	"	PUNCT
easat-5356	336	18	pythagorean	pythagorean	VERB
easat-5356	336	19	fuzzy	fuzzy	ADJ
easat-5356	336	20	soft	soft	ADJ
easat-5356	336	21	rms	rm	NOUN
easat-5356	336	22	approach	approach	NOUN
easat-5356	336	23	to	to	ADP
easat-5356	336	24	decision	decision	NOUN
easat-5356	336	25	making	making	NOUN
easat-5356	336	26	and	and	CCONJ
easat-5356	336	27	medical	medical	ADJ
easat-5356	336	28	diagnosis	diagnosis	NOUN
easat-5356	336	29	,	,	PUNCT
easat-5356	336	30	"	"	PUNCT
easat-5356	336	31	afrika	afrika	ADJ
easat-5356	336	32	matematika	matematika	NOUN
easat-5356	336	33	,	,	PUNCT
easat-5356	336	34	vol	vol	NOUN
easat-5356	336	35	.	.	PROPN
easat-5356	336	36	33	33	NUM
easat-5356	336	37	,	,	PUNCT
easat-5356	336	38	no	no	INTJ
easat-5356	336	39	.	.	NOUN
easat-5356	336	40	4	4	NUM
easat-5356	336	41	,	,	PUNCT
easat-5356	336	42	p.	p.	NOUN
easat-5356	336	43	97	97	NUM
easat-5356	336	44	,	,	PUNCT
easat-5356	336	45	2022	2022	NUM
easat-5356	336	46	.	.	PUNCT
easat-5356	337	1	https://doi.org/10.1007/s13370-022-01031-7	https://doi.org/10.1007/s13370-022-01031-7	NUM
easat-5356	338	1	[	[	X
easat-5356	338	2	3	3	X
easat-5356	338	3	]	]	X
easat-5356	338	4	h.	h.	PROPN
easat-5356	338	5	du	du	PROPN
easat-5356	338	6	,	,	PUNCT
easat-5356	338	7	h.	h.	PROPN
easat-5356	338	8	shi	shi	PROPN
easat-5356	338	9	,	,	PUNCT
easat-5356	338	10	d.	d.	PROPN
easat-5356	338	11	zeng	zeng	PROPN
easat-5356	338	12	,	,	PUNCT
easat-5356	338	13	x.-p	x.-p	PROPN
easat-5356	338	14	.	.	PUNCT
easat-5356	339	1	zhang	zhang	PROPN
easat-5356	339	2	,	,	PUNCT
easat-5356	339	3	and	and	CCONJ
easat-5356	339	4	t.	t.	PROPN
easat-5356	339	5	mei	mei	PROPN
easat-5356	339	6	,	,	PUNCT
easat-5356	339	7	"	"	PUNCT
easat-5356	339	8	the	the	DET
easat-5356	339	9	elements	element	NOUN
easat-5356	339	10	of	of	ADP
easat-5356	339	11	end	end	NOUN
easat-5356	339	12	-	-	PUNCT
easat-5356	339	13	to	to	ADP
easat-5356	339	14	-	-	PUNCT
easat-5356	339	15	end	end	NOUN
easat-5356	339	16	deep	deep	ADJ
easat-5356	339	17	face	face	NOUN
easat-5356	339	18	recognition	recognition	NOUN
easat-5356	339	19	:	:	PUNCT
easat-5356	339	20	a	a	DET
easat-5356	339	21	survey	survey	NOUN
easat-5356	339	22	of	of	ADP
easat-5356	339	23	recent	recent	ADJ
easat-5356	339	24	advances	advance	NOUN
easat-5356	339	25	,	,	PUNCT
easat-5356	339	26	"	"	PUNCT
easat-5356	339	27	acm	acm	PROPN
easat-5356	339	28	computing	computing	NOUN
easat-5356	339	29	surveys	survey	NOUN
easat-5356	339	30	(	(	PUNCT
easat-5356	339	31	csur	csur	NOUN
easat-5356	339	32	)	)	PUNCT
easat-5356	339	33	,	,	PUNCT
easat-5356	339	34	vol	vol	NOUN
easat-5356	339	35	.	.	PROPN
easat-5356	339	36	54	54	NUM
easat-5356	339	37	,	,	PUNCT
easat-5356	339	38	no	no	NOUN
easat-5356	339	39	.	.	PUNCT
easat-5356	339	40	10s	10	NOUN
easat-5356	339	41	,	,	PUNCT
easat-5356	339	42	pp	pp	ADJ
easat-5356	339	43	.	.	PUNCT
easat-5356	340	1	1	1	NUM
easat-5356	340	2	-	-	SYM
easat-5356	340	3	42	42	NUM
easat-5356	340	4	,	,	PUNCT
easat-5356	340	5	2022	2022	NUM
easat-5356	340	6	.	.	PUNCT
easat-5356	341	1	https://doi.org/10.1145/3507902	https://doi.org/10.1145/3507902	VERB
easat-5356	342	1	[	[	X
easat-5356	342	2	4	4	X
easat-5356	342	3	]	]	PUNCT
easat-5356	342	4	j.	j.	PROPN
easat-5356	342	5	you	you	PROPN
easat-5356	342	6	,	,	PUNCT
easat-5356	342	7	y.	y.	PROPN
easat-5356	342	8	sun	sun	PROPN
easat-5356	342	9	,	,	PUNCT
easat-5356	342	10	m.	m.	NOUN
easat-5356	342	11	guo	guo	PROPN
easat-5356	342	12	,	,	PUNCT
easat-5356	342	13	c.	c.	PROPN
easat-5356	342	14	feng	feng	PROPN
easat-5356	342	15	,	,	PUNCT
easat-5356	342	16	and	and	CCONJ
easat-5356	342	17	j.	j.	PROPN
easat-5356	342	18	ran	run	VERB
easat-5356	342	19	,	,	PUNCT
easat-5356	342	20	"	"	PUNCT
easat-5356	342	21	lvface	lvface	NOUN
easat-5356	342	22	:	:	PUNCT
easat-5356	342	23	large	large	ADJ
easat-5356	342	24	vision	vision	NOUN
easat-5356	342	25	model	model	NOUN
easat-5356	342	26	for	for	ADP
easat-5356	342	27	face	face	NOUN
easat-5356	342	28	recogniton	recogniton	NOUN
easat-5356	342	29	,	,	PUNCT
easat-5356	342	30	"	"	PUNCT
easat-5356	342	31	arxiv	arxiv	PROPN
easat-5356	342	32	preprint	preprint	VERB
easat-5356	342	33	arxiv:2501.13420	arxiv:2501.13420	NOUN
easat-5356	342	34	,	,	PUNCT
easat-5356	342	35	2025	2025	NUM
easat-5356	342	36	.	.	PUNCT
easat-5356	343	1	[	[	X
easat-5356	343	2	5	5	NUM
easat-5356	343	3	]	]	X
easat-5356	343	4	n.	n.	PROPN
easat-5356	343	5	algaraawi	algaraawi	PROPN
easat-5356	343	6	,	,	PUNCT
easat-5356	343	7	t.	t.	PROPN
easat-5356	343	8	morris	morris	PROPN
easat-5356	343	9	,	,	PUNCT
easat-5356	343	10	and	and	CCONJ
easat-5356	343	11	t.	t.	PROPN
easat-5356	343	12	f.	f.	PROPN
easat-5356	343	13	cootes	cootes	PROPN
easat-5356	343	14	,	,	PUNCT
easat-5356	343	15	"	"	PUNCT
easat-5356	343	16	facial	facial	ADJ
easat-5356	343	17	feature	feature	NOUN
easat-5356	343	18	point	point	NOUN
easat-5356	343	19	detection	detection	NOUN
easat-5356	343	20	under	under	ADP
easat-5356	343	21	large	large	ADJ
easat-5356	343	22	range	range	NOUN
easat-5356	343	23	of	of	ADP
easat-5356	343	24	face	face	NOUN
easat-5356	343	25	deformations	deformation	NOUN
easat-5356	343	26	,	,	PUNCT
easat-5356	343	27	"	"	PUNCT
easat-5356	343	28	journal	journal	NOUN
easat-5356	343	29	of	of	ADP
easat-5356	343	30	visual	visual	ADJ
easat-5356	343	31	communication	communication	NOUN
easat-5356	343	32	and	and	CCONJ
easat-5356	343	33	image	image	NOUN
easat-5356	343	34	representation	representation	NOUN
easat-5356	343	35	,	,	PUNCT
easat-5356	343	36	vol	vol	NOUN
easat-5356	343	37	.	.	PROPN
easat-5356	343	38	103	103	NUM
easat-5356	343	39	,	,	PUNCT
easat-5356	343	40	p.	p.	NOUN
easat-5356	343	41	104264	104264	NUM
easat-5356	343	42	,	,	PUNCT
easat-5356	343	43	2024	2024	NUM
easat-5356	343	44	.	.	PUNCT
easat-5356	344	1	https://doi.org/10.1016/j.jvcir.2024.104264	https://doi.org/10.1016/j.jvcir.2024.104264	ADJ
easat-5356	344	2	[	[	X
easat-5356	344	3	6	6	NUM
easat-5356	344	4	]	]	PUNCT
easat-5356	344	5	a.	a.	NOUN
easat-5356	344	6	nogales	nogales	PROPN
easat-5356	344	7	,	,	PUNCT
easat-5356	344	8	m.	m.	NOUN
easat-5356	344	9	rodríguez	rodríguez	NOUN
easat-5356	344	10	-	-	PUNCT
easat-5356	344	11	aragón	aragón	PROPN
easat-5356	344	12	,	,	PUNCT
easat-5356	344	13	and	and	CCONJ
easat-5356	344	14	á	á	X
easat-5356	344	15	.	.	PUNCT
easat-5356	345	1	j.	j.	PROPN
easat-5356	345	2	garcía	garcía	PROPN
easat-5356	345	3	-	-	PUNCT
easat-5356	345	4	tejedor	tejedor	NOUN
easat-5356	345	5	,	,	PUNCT
easat-5356	345	6	"	"	PUNCT
easat-5356	345	7	a	a	DET
easat-5356	345	8	systematic	systematic	ADJ
easat-5356	345	9	review	review	NOUN
easat-5356	345	10	of	of	ADP
easat-5356	345	11	the	the	DET
easat-5356	345	12	application	application	NOUN
easat-5356	345	13	of	of	ADP
easat-5356	345	14	deep	deep	ADJ
easat-5356	345	15	learning	learning	NOUN
easat-5356	345	16	techniques	technique	NOUN
easat-5356	345	17	in	in	ADP
easat-5356	345	18	the	the	DET
easat-5356	345	19	physiotherapeutic	physiotherapeutic	ADJ
easat-5356	345	20	therapy	therapy	NOUN
easat-5356	345	21	of	of	ADP
easat-5356	345	22	musculoskeletal	musculoskeletal	ADJ
easat-5356	345	23	pathologies	pathology	NOUN
easat-5356	345	24	,	,	PUNCT
easat-5356	345	25	"	"	PUNCT
easat-5356	345	26	computers	computer	NOUN
easat-5356	345	27	in	in	ADP
easat-5356	345	28	biology	biology	NOUN
easat-5356	345	29	and	and	CCONJ
easat-5356	345	30	medicine	medicine	NOUN
easat-5356	345	31	,	,	PUNCT
easat-5356	345	32	vol	vol	NOUN
easat-5356	345	33	.	.	PUNCT
easat-5356	345	34	172	172	NUM
easat-5356	345	35	,	,	PUNCT
easat-5356	345	36	p.	p.	NOUN
easat-5356	345	37	108082	108082	NUM
easat-5356	345	38	,	,	PUNCT
easat-5356	345	39	2024	2024	NUM
easat-5356	345	40	.	.	PUNCT
easat-5356	346	1	https://doi.org/10.1016/j.compbiomed.2024.108082	https://doi.org/10.1016/j.compbiomed.2024.108082	NOUN
easat-5356	346	2	[	[	X
easat-5356	346	3	7	7	NUM
easat-5356	346	4	]	]	X
easat-5356	346	5	m.	m.	PROPN
easat-5356	346	6	jani	jani	PROPN
easat-5356	346	7	,	,	PUNCT
easat-5356	346	8	j.	j.	PROPN
easat-5356	346	9	fayyad	fayyad	PROPN
easat-5356	346	10	,	,	PUNCT
easat-5356	346	11	y.	y.	PROPN
easat-5356	346	12	al	al	PROPN
easat-5356	346	13	-	-	PUNCT
easat-5356	346	14	younes	younes	PROPN
easat-5356	346	15	,	,	PUNCT
easat-5356	346	16	and	and	CCONJ
easat-5356	346	17	h.	h.	PROPN
easat-5356	346	18	najjaran	najjaran	PROPN
easat-5356	346	19	,	,	PUNCT
easat-5356	346	20	"	"	PUNCT
easat-5356	346	21	model	model	NOUN
easat-5356	346	22	compression	compression	NOUN
easat-5356	346	23	methods	method	NOUN
easat-5356	346	24	for	for	ADP
easat-5356	346	25	yolov5	yolov5	NOUN
easat-5356	346	26	:	:	PUNCT
easat-5356	346	27	a	a	DET
easat-5356	346	28	review	review	NOUN
easat-5356	346	29	,	,	PUNCT
easat-5356	346	30	"	"	PUNCT
easat-5356	346	31	arxiv	arxiv	PROPN
easat-5356	346	32	preprint	preprint	NOUN
easat-5356	346	33	arxiv:2307.11904	arxiv:2307.11904	NOUN
easat-5356	346	34	,	,	PUNCT
easat-5356	346	35	2023	2023	NUM
easat-5356	346	36	.	.	PUNCT
easat-5356	347	1	[	[	X
easat-5356	347	2	8	8	NUM
easat-5356	347	3	]	]	X
easat-5356	347	4	b.	b.	PROPN
easat-5356	347	5	b.	b.	PROPN
easat-5356	347	6	topal	topal	PROPN
easat-5356	347	7	,	,	PUNCT
easat-5356	347	8	"	"	PUNCT
easat-5356	347	9	reproducibility	reproducibility	NOUN
easat-5356	347	10	report	report	NOUN
easat-5356	347	11	retinaface	retinaface	NOUN
easat-5356	347	12	:	:	PUNCT
easat-5356	347	13	single	single	ADJ
easat-5356	347	14	-	-	PUNCT
easat-5356	347	15	shot	shoot	VERB
easat-5356	347	16	multi	multi	ADJ
easat-5356	347	17	-	-	ADJ
easat-5356	347	18	level	level	ADJ
easat-5356	347	19	face	face	NOUN
easat-5356	347	20	localization	localization	NOUN
easat-5356	347	21	in	in	ADP
easat-5356	347	22	the	the	DET
easat-5356	347	23	wild	wild	NOUN
easat-5356	347	24	.	.	PUNCT
easat-5356	347	25	"	"	PUNCT
easat-5356	348	1	https://doi.org/10.1109/cvpr42600.2020.00525	https://doi.org/10.1109/cvpr42600.2020.00525	PROPN
easat-5356	349	1	[	[	X
easat-5356	349	2	9	9	NUM
easat-5356	349	3	]	]	PUNCT
easat-5356	349	4	m.	m.	NOUN
easat-5356	349	5	i.	i.	PROPN
easat-5356	349	6	hossain	hossain	PROPN
easat-5356	349	7	and	and	CCONJ
easat-5356	349	8	h.	h.	PROPN
easat-5356	349	9	kabir	kabir	PROPN
easat-5356	349	10	,	,	PUNCT
easat-5356	349	11	"	"	PUNCT
easat-5356	349	12	an	an	DET
easat-5356	349	13	efficient	efficient	ADJ
easat-5356	349	14	way	way	NOUN
easat-5356	349	15	to	to	PART
easat-5356	349	16	recognize	recognize	VERB
easat-5356	349	17	faces	face	NOUN
easat-5356	349	18	using	use	VERB
easat-5356	349	19	mean	mean	ADJ
easat-5356	349	20	embeddings	embedding	NOUN
easat-5356	349	21	,	,	PUNCT
easat-5356	349	22	"	"	PUNCT
easat-5356	349	23	in	in	ADP
easat-5356	349	24	2021	2021	NUM
easat-5356	349	25	international	international	ADJ
easat-5356	349	26	conference	conference	NOUN
easat-5356	349	27	on	on	ADP
easat-5356	349	28	advances	advance	NOUN
easat-5356	349	29	in	in	ADP
easat-5356	349	30	electrical	electrical	ADJ
easat-5356	349	31	,	,	PUNCT
easat-5356	349	32	computing	computing	NOUN
easat-5356	349	33	,	,	PUNCT
easat-5356	349	34	communication	communication	NOUN
easat-5356	349	35	and	and	CCONJ
easat-5356	349	36	sustainable	sustainable	ADJ
easat-5356	349	37	technologies	technology	NOUN
easat-5356	349	38	(	(	PUNCT
easat-5356	349	39	icaect	icaect	NOUN
easat-5356	349	40	)	)	PUNCT
easat-5356	349	41	,	,	PUNCT
easat-5356	349	42	2021	2021	NUM
easat-5356	349	43	:	:	PUNCT
easat-5356	349	44	ieee	ieee	NOUN
easat-5356	349	45	,	,	PUNCT
easat-5356	349	46	pp	pp	ADJ
easat-5356	349	47	.	.	PUNCT
easat-5356	350	1	1	1	NUM
easat-5356	350	2	-	-	SYM
easat-5356	350	3	10	10	NUM
easat-5356	350	4	,	,	PUNCT
easat-5356	350	5	doi	doi	NOUN
easat-5356	350	6	:	:	PUNCT
easat-5356	350	7	https://doi.org/10.1109/icaect49130.2021.9392401	https://doi.org/10.1109/icaect49130.2021.9392401	PROPN
easat-5356	350	8	.	.	PUNCT
easat-5356	351	1	[	[	X
easat-5356	351	2	10	10	NUM
easat-5356	351	3	]	]	PUNCT
easat-5356	351	4	s.	s.	PROPN
easat-5356	351	5	qi	qi	PROPN
easat-5356	351	6	,	,	PUNCT
easat-5356	351	7	j.	j.	PROPN
easat-5356	351	8	zhang	zhang	PROPN
easat-5356	351	9	,	,	PUNCT
easat-5356	351	10	z.	z.	PROPN
easat-5356	351	11	bao	bao	PROPN
easat-5356	351	12	,	,	PUNCT
easat-5356	351	13	and	and	CCONJ
easat-5356	351	14	d.	d.	PROPN
easat-5356	351	15	li	li	PROPN
easat-5356	351	16	,	,	PUNCT
easat-5356	351	17	"	"	PUNCT
easat-5356	351	18	making	make	VERB
easat-5356	351	19	faceboxes	faceboxe	NOUN
easat-5356	351	20	better	well	ADV
easat-5356	351	21	and	and	CCONJ
easat-5356	351	22	faster	fast	ADV
easat-5356	351	23	by	by	ADP
easat-5356	351	24	reparameterization	reparameterization	NOUN
easat-5356	351	25	,	,	PUNCT
easat-5356	351	26	"	"	PUNCT
easat-5356	351	27	in	in	ADP
easat-5356	351	28	third	third	ADJ
easat-5356	351	29	international	international	ADJ
easat-5356	351	30	conference	conference	NOUN
easat-5356	351	31	on	on	ADP
easat-5356	351	32	control	control	NOUN
easat-5356	351	33	and	and	CCONJ
easat-5356	351	34	intelligent	intelligent	ADJ
easat-5356	351	35	robotics	robotic	NOUN
easat-5356	351	36	(	(	PUNCT
easat-5356	351	37	iccir	iccir	NOUN
easat-5356	351	38	2023	2023	NUM
easat-5356	351	39	)	)	PUNCT
easat-5356	351	40	,	,	PUNCT
easat-5356	351	41	2023	2023	NUM
easat-5356	351	42	,	,	PUNCT
easat-5356	351	43	vol	vol	NOUN
easat-5356	351	44	.	.	NOUN
easat-5356	351	45	12940	12940	NUM
easat-5356	351	46	:	:	PUNCT
easat-5356	351	47	spie	spie	ADJ
easat-5356	351	48	,	,	PUNCT
easat-5356	351	49	pp	pp	ADJ
easat-5356	351	50	.	.	PUNCT
easat-5356	352	1	726	726	NUM
easat-5356	352	2	-	-	SYM
easat-5356	352	3	729	729	NUM
easat-5356	352	4	.	.	PUNCT
easat-5356	353	1	[	[	X
easat-5356	353	2	11	11	NUM
easat-5356	353	3	]	]	PUNCT
easat-5356	353	4	j.	j.	PROPN
easat-5356	353	5	zhang	zhang	PROPN
easat-5356	353	6	,	,	PUNCT
easat-5356	353	7	c.	c.	PROPN
easat-5356	353	8	hou	hou	PROPN
easat-5356	353	9	,	,	PUNCT
easat-5356	353	10	x.	x.	PROPN
easat-5356	353	11	yang	yang	PROPN
easat-5356	353	12	,	,	PUNCT
easat-5356	353	13	x.	x.	PROPN
easat-5356	353	14	yang	yang	PROPN
easat-5356	353	15	,	,	PUNCT
easat-5356	353	16	w.	w.	PROPN
easat-5356	353	17	yang	yang	PROPN
easat-5356	353	18	,	,	PUNCT
easat-5356	353	19	and	and	CCONJ
easat-5356	353	20	h.	h.	PROPN
easat-5356	353	21	cui	cui	PROPN
easat-5356	353	22	,	,	PUNCT
easat-5356	353	23	"	"	PUNCT
easat-5356	353	24	advancing	advance	VERB
easat-5356	353	25	face	face	NOUN
easat-5356	353	26	detection	detection	NOUN
easat-5356	353	27	efficiency	efficiency	NOUN
easat-5356	353	28	:	:	PUNCT
easat-5356	353	29	utilizing	utilize	VERB
easat-5356	353	30	classification	classification	NOUN
easat-5356	353	31	networks	network	NOUN
easat-5356	353	32	for	for	ADP
easat-5356	353	33	lowering	lower	VERB
easat-5356	353	34	false	false	ADJ
easat-5356	353	35	positive	positive	ADJ
easat-5356	353	36	incidences	incidence	NOUN
easat-5356	353	37	,	,	PUNCT
easat-5356	353	38	"	"	PUNCT
easat-5356	353	39	array	array	NOUN
easat-5356	353	40	,	,	PUNCT
easat-5356	353	41	vol	vol	NOUN
easat-5356	353	42	.	.	PROPN
easat-5356	353	43	22	22	NUM
easat-5356	353	44	,	,	PUNCT
easat-5356	353	45	p.	p.	NOUN
easat-5356	353	46	100347	100347	NUM
easat-5356	353	47	,	,	PUNCT
easat-5356	353	48	2024	2024	NUM
easat-5356	353	49	.	.	PUNCT
easat-5356	354	1	https://doi.org/10.1016/j.array.2024.100347	https://doi.org/10.1016/j.array.2024.100347	ADJ
easat-5356	354	2	[	[	X
easat-5356	354	3	12	12	NUM
easat-5356	354	4	]	]	PUNCT
easat-5356	354	5	q.	q.	PROPN
easat-5356	354	6	wu	wu	PROPN
easat-5356	354	7	,	,	PUNCT
easat-5356	354	8	x.	x.	PROPN
easat-5356	354	9	wang	wang	PROPN
easat-5356	354	10	,	,	PUNCT
easat-5356	354	11	n.	n.	PROPN
easat-5356	354	12	li	li	PROPN
easat-5356	354	13	,	,	PUNCT
easat-5356	354	14	s.	s.	PROPN
easat-5356	354	15	fong	fong	PROPN
easat-5356	354	16	,	,	PUNCT
easat-5356	354	17	l.	l.	PROPN
easat-5356	354	18	zhang	zhang	PROPN
easat-5356	354	19	,	,	PUNCT
easat-5356	354	20	and	and	CCONJ
easat-5356	354	21	j.	j.	PROPN
easat-5356	354	22	yang	yang	PROPN
easat-5356	354	23	,	,	PUNCT
easat-5356	354	24	"	"	PUNCT
easat-5356	354	25	real	real	ADJ
easat-5356	354	26	-	-	PUNCT
easat-5356	354	27	time	time	NOUN
easat-5356	354	28	face	face	NOUN
easat-5356	354	29	and	and	CCONJ
easat-5356	354	30	facial	facial	ADJ
easat-5356	354	31	landmark	landmark	NOUN
easat-5356	354	32	joint	joint	ADJ
easat-5356	354	33	detection	detection	NOUN
easat-5356	354	34	based	base	VERB
easat-5356	354	35	on	on	ADP
easat-5356	354	36	end	end	NOUN
easat-5356	354	37	-	-	PUNCT
easat-5356	354	38	to	to	ADP
easat-5356	354	39	-	-	PUNCT
easat-5356	354	40	end	end	NOUN
easat-5356	354	41	deep	deep	ADJ
easat-5356	354	42	network	network	NOUN
easat-5356	354	43	,	,	PUNCT
easat-5356	354	44	"	"	PUNCT
easat-5356	354	45	ieee	ieee	NOUN
easat-5356	354	46	transactions	transaction	NOUN
easat-5356	354	47	on	on	ADP
easat-5356	354	48	instrumentation	instrumentation	NOUN
easat-5356	354	49	and	and	CCONJ
easat-5356	354	50	measurement	measurement	NOUN
easat-5356	354	51	,	,	PUNCT
easat-5356	354	52	2025	2025	NUM
easat-5356	354	53	.	.	PUNCT
easat-5356	355	1	https://doi.org/10.1109/tim.2025.3541698	https://doi.org/10.1109/tim.2025.3541698	NOUN
easat-5356	356	1	[	[	X
easat-5356	356	2	13	13	NUM
easat-5356	356	3	]	]	PUNCT
easat-5356	356	4	s.	s.	PROPN
easat-5356	356	5	huang	huang	PROPN
easat-5356	356	6	,	,	PUNCT
easat-5356	356	7	z.	z.	PROPN
easat-5356	356	8	lu	lu	PROPN
easat-5356	356	9	,	,	PUNCT
easat-5356	356	10	r.	r.	PROPN
easat-5356	356	11	cheng	cheng	PROPN
easat-5356	356	12	,	,	PUNCT
easat-5356	356	13	and	and	CCONJ
easat-5356	356	14	c.	c.	PROPN
easat-5356	356	15	he	he	PRON
easat-5356	356	16	,	,	PUNCT
easat-5356	356	17	"	"	PUNCT
easat-5356	356	18	fapn	fapn	NOUN
easat-5356	356	19	:	:	PUNCT
easat-5356	356	20	feature	feature	NOUN
easat-5356	356	21	-	-	PUNCT
easat-5356	356	22	aligned	align	VERB
easat-5356	356	23	pyramid	pyramid	NOUN
easat-5356	356	24	network	network	NOUN
easat-5356	356	25	for	for	ADP
easat-5356	356	26	dense	dense	ADJ
easat-5356	356	27	image	image	NOUN
easat-5356	356	28	prediction	prediction	NOUN
easat-5356	356	29	,	,	PUNCT
easat-5356	356	30	"	"	PUNCT
easat-5356	356	31	in	in	ADP
easat-5356	356	32	proceedings	proceeding	NOUN
easat-5356	356	33	of	of	ADP
easat-5356	356	34	the	the	DET
easat-5356	356	35	ieee	ieee	NOUN
easat-5356	356	36	/	/	SYM
easat-5356	356	37	cvf	cvf	NOUN
easat-5356	356	38	international	international	ADJ
easat-5356	356	39	conference	conference	NOUN
easat-5356	356	40	on	on	ADP
easat-5356	356	41	computer	computer	NOUN
easat-5356	356	42	vision	vision	NOUN
easat-5356	356	43	,	,	PUNCT
easat-5356	356	44	2021	2021	NUM
easat-5356	356	45	,	,	PUNCT
easat-5356	356	46	pp	pp	PROPN
easat-5356	356	47	.	.	PUNCT
easat-5356	356	48	864	864	NUM
easat-5356	356	49	-	-	SYM
easat-5356	356	50	873	873	NUM
easat-5356	356	51	,	,	PUNCT
easat-5356	356	52	doi	doi	NOUN
easat-5356	356	53	:	:	PUNCT
easat-5356	356	54	https://doi.org/10.1109/iccv48922.2021.00090	https://doi.org/10.1109/iccv48922.2021.00090	NOUN
easat-5356	356	55	.	.	PUNCT
easat-5356	357	1	[	[	X
easat-5356	357	2	14	14	NUM
easat-5356	357	3	]	]	PUNCT
easat-5356	357	4	j.	j.	PROPN
easat-5356	357	5	terven	terven	PROPN
easat-5356	357	6	,	,	PUNCT
easat-5356	357	7	d.-m	d.-m	PROPN
easat-5356	357	8	.	.	PUNCT
easat-5356	358	1	córdova	córdova	PROPN
easat-5356	358	2	-	-	PUNCT
easat-5356	358	3	esparza	esparza	PROPN
easat-5356	358	4	,	,	PUNCT
easat-5356	358	5	and	and	CCONJ
easat-5356	358	6	j.-a	j.-a	NOUN
easat-5356	358	7	.	.	PUNCT
easat-5356	359	1	romero	romero	PROPN
easat-5356	359	2	-	-	PUNCT
easat-5356	359	3	gonzález	gonzález	PROPN
easat-5356	359	4	,	,	PUNCT
easat-5356	359	5	"	"	PUNCT
easat-5356	359	6	a	a	DET
easat-5356	359	7	comprehensive	comprehensive	ADJ
easat-5356	359	8	review	review	NOUN
easat-5356	359	9	of	of	ADP
easat-5356	359	10	yolo	yolo	ADJ
easat-5356	359	11	architectures	architecture	NOUN
easat-5356	359	12	in	in	ADP
easat-5356	359	13	computer	computer	NOUN
easat-5356	359	14	vision	vision	NOUN
easat-5356	359	15	:	:	PUNCT
easat-5356	359	16	from	from	ADP
easat-5356	359	17	yolov1	yolov1	NOUN
easat-5356	359	18	to	to	ADP
easat-5356	359	19	yolov8	yolov8	PROPN
easat-5356	359	20	and	and	CCONJ
easat-5356	359	21	yolo	yolo	PROPN
easat-5356	359	22	-	-	PUNCT
easat-5356	359	23	nas	nas	PROPN
easat-5356	359	24	,	,	PUNCT
easat-5356	359	25	"	"	PUNCT
easat-5356	359	26	machine	machine	NOUN
easat-5356	359	27	learning	learning	NOUN
easat-5356	359	28	and	and	CCONJ
easat-5356	359	29	knowledge	knowledge	NOUN
easat-5356	359	30	extraction	extraction	NOUN
easat-5356	359	31	,	,	PUNCT
easat-5356	359	32	vol	vol	NOUN
easat-5356	359	33	.	.	PROPN
easat-5356	359	34	5	5	NUM
easat-5356	359	35	,	,	PUNCT
easat-5356	359	36	no	no	INTJ
easat-5356	359	37	.	.	NOUN
easat-5356	359	38	4	4	NUM
easat-5356	359	39	,	,	PUNCT
easat-5356	359	40	pp	pp	ADJ
easat-5356	359	41	.	.	PUNCT
easat-5356	360	1	1680	1680	NUM
easat-5356	360	2	-	-	SYM
easat-5356	360	3	1716	1716	NUM
easat-5356	360	4	,	,	PUNCT
easat-5356	360	5	2023	2023	NUM
easat-5356	360	6	.	.	PUNCT
easat-5356	361	1	https://doi.org/10.3390/make5040083	https://doi.org/10.3390/make5040083	PROPN
easat-5356	362	1	[	[	X
easat-5356	362	2	15	15	NUM
easat-5356	362	3	]	]	PUNCT
easat-5356	362	4	z.	z.	PROPN
easat-5356	362	5	wang	wang	PROPN
easat-5356	362	6	,	,	PUNCT
easat-5356	362	7	x.	x.	PROPN
easat-5356	362	8	xie	xie	PROPN
easat-5356	362	9	,	,	PUNCT
easat-5356	362	10	j.	j.	PROPN
easat-5356	362	11	yang	yang	PROPN
easat-5356	362	12	,	,	PUNCT
easat-5356	362	13	and	and	CCONJ
easat-5356	362	14	g.	g.	PROPN
easat-5356	362	15	shi	shi	PROPN
easat-5356	362	16	,	,	PUNCT
easat-5356	362	17	"	"	PUNCT
easat-5356	362	18	soft	soft	ADJ
easat-5356	362	19	focal	focal	ADJ
easat-5356	362	20	loss	loss	NOUN
easat-5356	362	21	:	:	PUNCT
easat-5356	362	22	evaluating	evaluate	VERB
easat-5356	362	23	sample	sample	NOUN
easat-5356	362	24	quality	quality	NOUN
easat-5356	362	25	for	for	ADP
easat-5356	362	26	dense	dense	ADJ
easat-5356	362	27	object	object	NOUN
easat-5356	362	28	detection	detection	NOUN
easat-5356	362	29	,	,	PUNCT
easat-5356	362	30	"	"	PUNCT
easat-5356	362	31	neurocomputing	neurocomputing	NOUN
easat-5356	362	32	,	,	PUNCT
easat-5356	362	33	vol	vol	NOUN
easat-5356	362	34	.	.	PROPN
easat-5356	362	35	480	480	NUM
easat-5356	362	36	,	,	PUNCT
easat-5356	362	37	pp	pp	ADJ
easat-5356	362	38	.	.	PUNCT
easat-5356	363	1	271	271	NUM
easat-5356	363	2	-	-	SYM
easat-5356	363	3	280	280	NUM
easat-5356	363	4	,	,	PUNCT
easat-5356	363	5	2022	2022	NUM
easat-5356	363	6	.	.	PUNCT
easat-5356	364	1	https://doi.org/10.1016/j.neucom.2021.12.102	https://doi.org/10.1016/j.neucom.2021.12.102	NUM
easat-5356	364	2	[	[	X
easat-5356	364	3	16	16	NUM
easat-5356	364	4	]	]	PUNCT
easat-5356	364	5	t.-y	t.-y	NOUN
easat-5356	364	6	.	.	PUNCT
easat-5356	365	1	lin	lin	PROPN
easat-5356	365	2	,	,	PUNCT
easat-5356	365	3	p.	p.	PROPN
easat-5356	365	4	goyal	goyal	PROPN
easat-5356	365	5	,	,	PUNCT
easat-5356	365	6	r.	r.	PROPN
easat-5356	365	7	girshick	girshick	PROPN
easat-5356	365	8	,	,	PUNCT
easat-5356	365	9	k.	k.	PROPN
easat-5356	365	10	he	he	PRON
easat-5356	365	11	,	,	PUNCT
easat-5356	365	12	and	and	CCONJ
easat-5356	365	13	p.	p.	NOUN
easat-5356	365	14	dollár	dollár	NOUN
easat-5356	365	15	,	,	PUNCT
easat-5356	365	16	"	"	PUNCT
easat-5356	365	17	focal	focal	ADJ
easat-5356	365	18	loss	loss	NOUN
easat-5356	365	19	for	for	ADP
easat-5356	365	20	dense	dense	ADJ
easat-5356	365	21	object	object	NOUN
easat-5356	365	22	detection	detection	NOUN
easat-5356	365	23	,	,	PUNCT
easat-5356	365	24	"	"	PUNCT
easat-5356	365	25	in	in	ADP
easat-5356	365	26	proceedings	proceeding	NOUN
easat-5356	365	27	of	of	ADP
easat-5356	365	28	the	the	DET
easat-5356	365	29	ieee	ieee	NOUN
easat-5356	365	30	international	international	PROPN
easat-5356	365	31	conference	conference	NOUN
easat-5356	365	32	on	on	ADP
easat-5356	365	33	computer	computer	NOUN
easat-5356	365	34	vision	vision	NOUN
easat-5356	365	35	,	,	PUNCT
easat-5356	365	36	2017	2017	NUM
easat-5356	365	37	,	,	PUNCT
easat-5356	365	38	pp	pp	ADJ
easat-5356	365	39	.	.	PUNCT
easat-5356	366	1	2980	2980	NUM
easat-5356	366	2	-	-	SYM
easat-5356	366	3	2988	2988	NUM
easat-5356	366	4	,	,	PUNCT
easat-5356	366	5	doi	doi	NOUN
easat-5356	366	6	:	:	PUNCT
easat-5356	366	7	https://doi.org/10.1109/iccv.2017.324	https://doi.org/10.1109/iccv.2017.324	ADJ
easat-5356	366	8	.	.	PUNCT
easat-5356	367	1	[	[	X
easat-5356	367	2	17	17	NUM
easat-5356	367	3	]	]	PUNCT
easat-5356	367	4	p.	p.	NOUN
easat-5356	367	5	song	song	NOUN
easat-5356	367	6	,	,	PUNCT
easat-5356	367	7	p.	p.	PROPN
easat-5356	367	8	li	li	PROPN
easat-5356	367	9	,	,	PUNCT
easat-5356	367	10	l.	l.	PROPN
easat-5356	367	11	dai	dai	PROPN
easat-5356	367	12	,	,	PUNCT
easat-5356	367	13	t.	t.	PROPN
easat-5356	367	14	wang	wang	PROPN
easat-5356	367	15	,	,	PUNCT
easat-5356	367	16	and	and	CCONJ
easat-5356	367	17	z.	z.	PROPN
easat-5356	367	18	chen	chen	PROPN
easat-5356	367	19	,	,	PUNCT
easat-5356	367	20	"	"	PUNCT
easat-5356	367	21	boosting	boost	VERB
easat-5356	367	22	r	r	NOUN
easat-5356	367	23	-	-	PUNCT
easat-5356	367	24	cnn	cnn	PROPN
easat-5356	367	25	:	:	PUNCT
easat-5356	367	26	reweighting	reweighte	VERB
easat-5356	367	27	r	r	X
easat-5356	367	28	-	-	PUNCT
easat-5356	367	29	cnn	cnn	PROPN
easat-5356	367	30	samples	sample	NOUN
easat-5356	367	31	by	by	ADP
easat-5356	367	32	rpn	rpn	PROPN
easat-5356	367	33	’s	’s	PART
easat-5356	367	34	error	error	NOUN
easat-5356	367	35	for	for	ADP
easat-5356	367	36	underwater	underwater	ADJ
easat-5356	367	37	object	object	NOUN
easat-5356	367	38	detection	detection	NOUN
easat-5356	367	39	,	,	PUNCT
easat-5356	367	40	"	"	PUNCT
easat-5356	367	41	neurocomputing	neurocomputing	NOUN
easat-5356	367	42	,	,	PUNCT
easat-5356	367	43	vol	vol	NOUN
easat-5356	367	44	.	.	PROPN
easat-5356	367	45	530	530	NUM
easat-5356	367	46	,	,	PUNCT
easat-5356	367	47	pp	pp	ADJ
easat-5356	367	48	.	.	PUNCT
easat-5356	368	1	150	150	NUM
easat-5356	368	2	-	-	SYM
easat-5356	368	3	164	164	NUM
easat-5356	368	4	,	,	PUNCT
easat-5356	368	5	2023	2023	NUM
easat-5356	368	6	.	.	PUNCT
easat-5356	369	1	https://creativecommons.org/licenses/by/4.0/	https://creativecommons.org/licenses/by/4.0/	PROPN
easat-5356	369	2	https://doi.org/10.1007/s13370-022-01031-7	https://doi.org/10.1007/s13370-022-01031-7	NUM
easat-5356	369	3	https://doi.org/10.1145/3507902	https://doi.org/10.1145/3507902	VERB
easat-5356	369	4	https://doi.org/10.1016/j.jvcir.2024.104264	https://doi.org/10.1016/j.jvcir.2024.104264	PROPN
easat-5356	369	5	https://doi.org/10.1016/j.compbiomed.2024.108082	https://doi.org/10.1016/j.compbiomed.2024.108082	PROPN
easat-5356	369	6	https://doi.org/10.1109/cvpr42600.2020.00525	https://doi.org/10.1109/cvpr42600.2020.00525	PROPN
easat-5356	369	7	https://doi.org/10.1109/icaect49130.2021.9392401	https://doi.org/10.1109/icaect49130.2021.9392401	PROPN
easat-5356	369	8	https://doi.org/10.1016/j.array.2024.100347	https://doi.org/10.1016/j.array.2024.100347	PROPN
easat-5356	369	9	https://doi.org/10.1109/tim.2025.3541698	https://doi.org/10.1109/tim.2025.3541698	NOUN
easat-5356	369	10	https://doi.org/10.1109/iccv48922.2021.00090	https://doi.org/10.1109/iccv48922.2021.00090	NOUN
easat-5356	369	11	https://doi.org/10.3390/make5040083	https://doi.org/10.3390/make5040083	NOUN
easat-5356	369	12	https://doi.org/10.1016/j.neucom.2021.12.102	https://doi.org/10.1016/j.neucom.2021.12.102	NOUN
easat-5356	369	13	https://doi.org/10.1109/iccv.2017.324	https://doi.org/10.1109/iccv.2017.324	PROPN
easat-5356	369	14	855	855	NUM
easat-5356	369	15	edelweiss	edelweiss	PROPN
easat-5356	369	16	applied	apply	VERB
easat-5356	369	17	science	science	NOUN
easat-5356	369	18	and	and	CCONJ
easat-5356	369	19	technology	technology	NOUN
easat-5356	369	20	issn	issn	PROPN
easat-5356	369	21	:	:	PUNCT
easat-5356	369	22	2576	2576	NUM
easat-5356	369	23	-	-	SYM
easat-5356	369	24	8484	8484	NUM
easat-5356	369	25	vol	vol	NOUN
easat-5356	369	26	.	.	PROPN
easat-5356	370	1	9	9	NUM
easat-5356	370	2	,	,	PUNCT
easat-5356	370	3	no	no	INTJ
easat-5356	370	4	.	.	NOUN
easat-5356	371	1	3	3	NUM
easat-5356	371	2	:	:	PUNCT
easat-5356	371	3	840	840	NUM
easat-5356	371	4	-	-	SYM
easat-5356	371	5	855	855	NUM
easat-5356	371	6	,	,	PUNCT
easat-5356	371	7	2025	2025	NUM
easat-5356	371	8	doi	doi	NOUN
easat-5356	371	9	:	:	PUNCT
easat-5356	371	10	10.55214/25768484.v9i3.5356	10.55214/25768484.v9i3.5356	NUM
easat-5356	371	11	©	©	PROPN
easat-5356	371	12	2025	2025	NUM
easat-5356	371	13	by	by	ADP
easat-5356	371	14	the	the	DET
easat-5356	371	15	authors	author	NOUN
easat-5356	371	16	;	;	PUNCT
easat-5356	371	17	licensee	licensee	PROPN
easat-5356	371	18	learning	learning	NOUN
easat-5356	371	19	gate	gate	NOUN
easat-5356	372	1	[	[	X
easat-5356	372	2	18	18	NUM
easat-5356	372	3	]	]	PUNCT
easat-5356	372	4	q.	q.	PROPN
easat-5356	372	5	zhang	zhang	PROPN
easat-5356	372	6	,	,	PUNCT
easat-5356	372	7	y.	y.	PROPN
easat-5356	372	8	xu	xu	PROPN
easat-5356	372	9	,	,	PUNCT
easat-5356	372	10	j.	j.	PROPN
easat-5356	372	11	zhang	zhang	PROPN
easat-5356	372	12	,	,	PUNCT
easat-5356	372	13	and	and	CCONJ
easat-5356	372	14	d.	d.	PROPN
easat-5356	372	15	tao	tao	PROPN
easat-5356	372	16	,	,	PUNCT
easat-5356	372	17	"	"	PUNCT
easat-5356	372	18	vitaev2	vitaev2	PROPN
easat-5356	372	19	:	:	PUNCT
easat-5356	372	20	vision	vision	NOUN
easat-5356	372	21	transformer	transformer	NOUN
easat-5356	372	22	advanced	advance	VERB
easat-5356	372	23	by	by	ADP
easat-5356	372	24	exploring	explore	VERB
easat-5356	372	25	inductive	inductive	ADJ
easat-5356	372	26	bias	bias	NOUN
easat-5356	372	27	for	for	ADP
easat-5356	372	28	image	image	NOUN
easat-5356	372	29	recognition	recognition	NOUN
easat-5356	372	30	and	and	CCONJ
easat-5356	372	31	beyond	beyond	ADP
easat-5356	372	32	,	,	PUNCT
easat-5356	372	33	"	"	PUNCT
easat-5356	372	34	international	international	ADJ
easat-5356	372	35	journal	journal	NOUN
easat-5356	372	36	of	of	ADP
easat-5356	372	37	computer	computer	NOUN
easat-5356	372	38	vision	vision	NOUN
easat-5356	372	39	,	,	PUNCT
easat-5356	372	40	vol	vol	NOUN
easat-5356	372	41	.	.	PROPN
easat-5356	373	1	131	131	NUM
easat-5356	373	2	,	,	PUNCT
easat-5356	373	3	no	no	INTJ
easat-5356	373	4	.	.	NOUN
easat-5356	373	5	5	5	NUM
easat-5356	373	6	,	,	PUNCT
easat-5356	373	7	pp	pp	ADJ
easat-5356	373	8	.	.	PUNCT
easat-5356	374	1	1141	1141	NUM
easat-5356	374	2	-	-	SYM
easat-5356	374	3	1162	1162	NUM
easat-5356	374	4	,	,	PUNCT
easat-5356	374	5	2023	2023	NUM
easat-5356	374	6	.	.	PUNCT
easat-5356	375	1	https://doi.org/10.1007/s11263-022-01739-w	https://doi.org/10.1007/s11263-022-01739-w	PRON
easat-5356	376	1	[	[	X
easat-5356	376	2	19	19	NUM
easat-5356	376	3	]	]	PUNCT
easat-5356	376	4	z.	z.	PROPN
easat-5356	376	5	tong	tong	PROPN
easat-5356	376	6	,	,	PUNCT
easat-5356	376	7	y.	y.	PROPN
easat-5356	376	8	chen	chen	PROPN
easat-5356	376	9	,	,	PUNCT
easat-5356	376	10	z.	z.	PROPN
easat-5356	376	11	xu	xu	PROPN
easat-5356	376	12	,	,	PUNCT
easat-5356	376	13	and	and	CCONJ
easat-5356	376	14	r.	r.	PROPN
easat-5356	376	15	yu	yu	PROPN
easat-5356	376	16	,	,	PUNCT
easat-5356	376	17	"	"	PUNCT
easat-5356	376	18	wise	wise	ADJ
easat-5356	376	19	-	-	PUNCT
easat-5356	376	20	iou	iou	NOUN
easat-5356	376	21	:	:	PUNCT
easat-5356	376	22	bounding	bound	VERB
easat-5356	376	23	box	box	NOUN
easat-5356	376	24	regression	regression	NOUN
easat-5356	376	25	loss	loss	NOUN
easat-5356	376	26	with	with	ADP
easat-5356	376	27	dynamic	dynamic	ADJ
easat-5356	376	28	focusing	focus	VERB
easat-5356	376	29	mechanism	mechanism	NOUN
easat-5356	376	30	,	,	PUNCT
easat-5356	376	31	"	"	PUNCT
easat-5356	376	32	arxiv	arxiv	PROPN
easat-5356	376	33	preprint	preprint	NOUN
easat-5356	376	34	arxiv:2301.10051	arxiv:2301.10051	PROPN
easat-5356	376	35	,	,	PUNCT
easat-5356	376	36	2023	2023	NUM
easat-5356	376	37	.	.	PUNCT
easat-5356	377	1	https://doi.org/10.1016/j.patcog.2022.109256	https://doi.org/10.1016/j.patcog.2022.109256	PROPN
easat-5356	378	1	[	[	X
easat-5356	378	2	20	20	NUM
easat-5356	378	3	]	]	PUNCT
easat-5356	378	4	h.	h.	PROPN
easat-5356	378	5	k.	k.	PROPN
easat-5356	379	1	alrammahi	alrammahi	PROPN
easat-5356	379	2	and	and	CCONJ
easat-5356	379	3	k.	k.	PROPN
easat-5356	379	4	a.	a.	PROPN
easat-5356	379	5	jabbar	jabbar	PROPN
easat-5356	379	6	,	,	PUNCT
easat-5356	379	7	"	"	PUNCT
easat-5356	379	8	automatic	automatic	ADJ
easat-5356	379	9	facial	facial	ADJ
easat-5356	379	10	expression	expression	NOUN
easat-5356	379	11	detection	detection	NOUN
easat-5356	379	12	using	use	VERB
easat-5356	379	13	genetic	genetic	ADJ
easat-5356	379	14	algorithm	algorithm	NOUN
easat-5356	379	15	with	with	ADP
easat-5356	379	16	svm	svm	PROPN
easat-5356	379	17	and	and	CCONJ
easat-5356	379	18	haar	haar	PROPN
easat-5356	379	19	wavelet	wavelet	NOUN
easat-5356	379	20	features	feature	NOUN
easat-5356	379	21	,	,	PUNCT
easat-5356	379	22	"	"	PUNCT
easat-5356	379	23	in	in	ADP
easat-5356	379	24	international	international	ADJ
easat-5356	379	25	conference	conference	NOUN
easat-5356	379	26	on	on	ADP
easat-5356	379	27	machine	machine	NOUN
easat-5356	379	28	learning	learning	NOUN
easat-5356	379	29	,	,	PUNCT
easat-5356	379	30	iot	iot	PROPN
easat-5356	379	31	and	and	CCONJ
easat-5356	379	32	big	big	ADJ
easat-5356	379	33	data	datum	NOUN
easat-5356	379	34	,	,	PUNCT
easat-5356	379	35	2024	2024	NUM
easat-5356	379	36	,	,	PUNCT
easat-5356	379	37	pp	pp	ADV
easat-5356	379	38	.	.	PUNCT
easat-5356	379	39	329	329	NUM
easat-5356	379	40	-	-	SYM
easat-5356	379	41	348	348	NUM
easat-5356	379	42	,	,	PUNCT
easat-5356	379	43	doi	doi	NOUN
easat-5356	379	44	:	:	PUNCT
easat-5356	379	45	https://doi.org/10.1007/978-981-97-8160-7_26	https://doi.org/10.1007/978-981-97-8160-7_26	NUM
easat-5356	379	46	.	.	PUNCT
easat-5356	380	1	[	[	X
easat-5356	380	2	21	21	NUM
easat-5356	380	3	]	]	X
easat-5356	380	4	s.	s.	PROPN
easat-5356	380	5	i.	i.	PROPN
easat-5356	380	6	venieris	venieris	PROPN
easat-5356	380	7	,	,	PUNCT
easat-5356	380	8	j.	j.	PROPN
easat-5356	380	9	fernandez	fernandez	PROPN
easat-5356	380	10	-	-	PUNCT
easat-5356	380	11	marques	marques	PROPN
easat-5356	380	12	,	,	PUNCT
easat-5356	380	13	and	and	CCONJ
easat-5356	380	14	n.	n.	PROPN
easat-5356	380	15	d.	d.	PROPN
easat-5356	380	16	lane	lane	PROPN
easat-5356	380	17	,	,	PUNCT
easat-5356	380	18	"	"	PUNCT
easat-5356	380	19	mitigating	mitigate	VERB
easat-5356	380	20	memory	memory	NOUN
easat-5356	380	21	wall	wall	NOUN
easat-5356	380	22	effects	effect	NOUN
easat-5356	380	23	in	in	ADP
easat-5356	380	24	cnn	cnn	PROPN
easat-5356	380	25	engines	engine	NOUN
easat-5356	380	26	with	with	ADP
easat-5356	380	27	on	on	ADP
easat-5356	380	28	-	-	PUNCT
easat-5356	380	29	thefly	thefly	NOUN
easat-5356	380	30	weights	weight	NOUN
easat-5356	380	31	generation	generation	NOUN
easat-5356	380	32	,	,	PUNCT
easat-5356	380	33	"	"	PUNCT
easat-5356	380	34	acm	acm	NOUN
easat-5356	380	35	transactions	transaction	NOUN
easat-5356	380	36	on	on	ADP
easat-5356	380	37	design	design	NOUN
easat-5356	380	38	automation	automation	NOUN
easat-5356	380	39	of	of	ADP
easat-5356	380	40	electronic	electronic	ADJ
easat-5356	380	41	systems	system	NOUN
easat-5356	380	42	,	,	PUNCT
easat-5356	380	43	vol	vol	NOUN
easat-5356	380	44	.	.	PROPN
easat-5356	380	45	28	28	NUM
easat-5356	380	46	,	,	PUNCT
easat-5356	380	47	no	no	INTJ
easat-5356	380	48	.	.	NOUN
easat-5356	380	49	6	6	NUM
easat-5356	380	50	,	,	PUNCT
easat-5356	380	51	pp	pp	ADJ
easat-5356	380	52	.	.	PUNCT
easat-5356	381	1	1	1	NUM
easat-5356	381	2	-	-	SYM
easat-5356	381	3	31	31	NUM
easat-5356	381	4	,	,	PUNCT
easat-5356	381	5	2023	2023	NUM
easat-5356	381	6	.	.	PUNCT
easat-5356	381	7	https://doi.org/10.1145/3611673	https://doi.org/10.1145/3611673	NOUN
easat-5356	382	1	[	[	X
easat-5356	382	2	22	22	NUM
easat-5356	382	3	]	]	PUNCT
easat-5356	382	4	s.	s.	PROPN
easat-5356	382	5	s.	s.	PROPN
easat-5356	382	6	khan	khan	PROPN
easat-5356	382	7	,	,	PUNCT
easat-5356	382	8	d.	d.	PROPN
easat-5356	382	9	sengupta	sengupta	PROPN
easat-5356	382	10	,	,	PUNCT
easat-5356	382	11	a.	a.	PROPN
easat-5356	382	12	ghosh	ghosh	PROPN
easat-5356	382	13	,	,	PUNCT
easat-5356	382	14	and	and	CCONJ
easat-5356	382	15	a.	a.	NOUN
easat-5356	382	16	chaudhuri	chaudhuri	PROPN
easat-5356	382	17	,	,	PUNCT
easat-5356	382	18	"	"	PUNCT
easat-5356	382	19	mtcnn++	mtcnn++	NOUN
easat-5356	382	20	:	:	PUNCT
easat-5356	382	21	a	a	DET
easat-5356	382	22	cnn	cnn	PROPN
easat-5356	382	23	-	-	PUNCT
easat-5356	382	24	based	base	VERB
easat-5356	382	25	face	face	NOUN
easat-5356	382	26	detection	detection	NOUN
easat-5356	382	27	algorithm	algorithm	NOUN
easat-5356	382	28	inspired	inspire	VERB
easat-5356	382	29	by	by	ADP
easat-5356	382	30	mtcnn	mtcnn	NOUN
easat-5356	382	31	,	,	PUNCT
easat-5356	382	32	"	"	PUNCT
easat-5356	382	33	the	the	DET
easat-5356	382	34	visual	visual	ADJ
easat-5356	382	35	computer	computer	NOUN
easat-5356	382	36	,	,	PUNCT
easat-5356	382	37	vol	vol	NOUN
easat-5356	382	38	.	.	PROPN
easat-5356	382	39	40	40	NUM
easat-5356	382	40	,	,	PUNCT
easat-5356	382	41	no	no	INTJ
easat-5356	382	42	.	.	NOUN
easat-5356	382	43	2	2	NUM
easat-5356	382	44	,	,	PUNCT
easat-5356	382	45	pp	pp	ADJ
easat-5356	382	46	.	.	PUNCT
easat-5356	383	1	899	899	NUM
easat-5356	383	2	-	-	SYM
easat-5356	383	3	917	917	NUM
easat-5356	383	4	,	,	PUNCT
easat-5356	383	5	2024	2024	NUM
easat-5356	383	6	.	.	PUNCT
easat-5356	384	1	https://doi.org/10.1007/s00371-023-02822-0	https://doi.org/10.1007/s00371-023-02822-0	X
easat-5356	385	1	[	[	X
easat-5356	385	2	23	23	NUM
easat-5356	385	3	]	]	X
easat-5356	385	4	h.	h.	PROPN
easat-5356	385	5	kaewkorn	kaewkorn	PROPN
easat-5356	385	6	,	,	PUNCT
easat-5356	385	7	l.	l.	PROPN
easat-5356	385	8	zhou	zhou	PROPN
easat-5356	385	9	,	,	PUNCT
easat-5356	385	10	and	and	CCONJ
easat-5356	385	11	w.	w.	PROPN
easat-5356	385	12	li	li	PROPN
easat-5356	385	13	,	,	PUNCT
easat-5356	385	14	"	"	PUNCT
easat-5356	385	15	rp	rp	NOUN
easat-5356	385	16	-	-	PUNCT
easat-5356	385	17	net	net	NOUN
easat-5356	385	18	:	:	PUNCT
easat-5356	385	19	a	a	DET
easat-5356	385	20	robust	robust	ADJ
easat-5356	385	21	polar	polar	ADJ
easat-5356	385	22	transformation	transformation	NOUN
easat-5356	385	23	network	network	NOUN
easat-5356	385	24	for	for	ADP
easat-5356	385	25	rotation	rotation	NOUN
easat-5356	385	26	-	-	PUNCT
easat-5356	385	27	invariant	invariant	ADJ
easat-5356	385	28	face	face	NOUN
easat-5356	385	29	detection	detection	NOUN
easat-5356	385	30	,	,	PUNCT
easat-5356	385	31	"	"	PUNCT
easat-5356	385	32	pattern	pattern	NOUN
easat-5356	385	33	recognition	recognition	NOUN
easat-5356	385	34	,	,	PUNCT
easat-5356	385	35	vol	vol	NOUN
easat-5356	385	36	.	.	PROPN
easat-5356	385	37	158	158	NUM
easat-5356	385	38	,	,	PUNCT
easat-5356	385	39	p.	p.	NOUN
easat-5356	385	40	111044	111044	NUM
easat-5356	385	41	,	,	PUNCT
easat-5356	385	42	2025	2025	NUM
easat-5356	385	43	.	.	PUNCT
easat-5356	386	1	https://doi.org/10.1016/j.patcog.2024.111044	https://doi.org/10.1016/j.patcog.2024.111044	PROPN
easat-5356	386	2	[	[	X
easat-5356	386	3	24	24	NUM
easat-5356	386	4	]	]	X
easat-5356	386	5	d.	d.	PROPN
easat-5356	386	6	m.	m.	PROPN
easat-5356	386	7	abdulhussien	abdulhussien	PROPN
easat-5356	386	8	and	and	CCONJ
easat-5356	386	9	l.	l.	PROPN
easat-5356	386	10	j.	j.	PROPN
easat-5356	386	11	saud	saud	PROPN
easat-5356	386	12	,	,	PUNCT
easat-5356	386	13	"	"	PUNCT
easat-5356	386	14	an	an	DET
easat-5356	386	15	evaluation	evaluation	NOUN
easat-5356	386	16	study	study	NOUN
easat-5356	386	17	of	of	ADP
easat-5356	386	18	face	face	NOUN
easat-5356	386	19	detection	detection	NOUN
easat-5356	386	20	by	by	ADP
easat-5356	386	21	viola	viola	PROPN
easat-5356	386	22	-	-	PUNCT
easat-5356	386	23	jones	jones	PROPN
easat-5356	386	24	algorithm	algorithm	PROPN
easat-5356	386	25	,	,	PUNCT
easat-5356	386	26	"	"	PUNCT
easat-5356	386	27	international	international	ADJ
easat-5356	386	28	journal	journal	NOUN
easat-5356	386	29	of	of	ADP
easat-5356	386	30	health	health	PROPN
easat-5356	386	31	sciences	sciences	PROPN
easat-5356	386	32	,	,	PUNCT
easat-5356	386	33	vol	vol	NOUN
easat-5356	386	34	.	.	PROPN
easat-5356	386	35	6	6	NUM
easat-5356	386	36	,	,	PUNCT
easat-5356	386	37	no	no	INTJ
easat-5356	386	38	.	.	PUNCT
easat-5356	386	39	s8	s8	PROPN
easat-5356	386	40	,	,	PUNCT
easat-5356	386	41	pp	pp	PROPN
easat-5356	386	42	.	.	PUNCT
easat-5356	387	1	4174	4174	NUM
easat-5356	387	2	-	-	SYM
easat-5356	387	3	4182	4182	NUM
easat-5356	387	4	,	,	PUNCT
easat-5356	387	5	2022	2022	NUM
easat-5356	387	6	.	.	PUNCT
easat-5356	388	1	https://doi.org/10.53730/ijhs.v6ns8.13127	https://doi.org/10.53730/ijhs.v6ns8.13127	PROPN
easat-5356	388	2	[	[	X
easat-5356	388	3	25	25	NUM
easat-5356	388	4	]	]	PUNCT
easat-5356	388	5	a.	a.	NOUN
easat-5356	388	6	aggarwal	aggarwal	NOUN
easat-5356	388	7	,	,	PUNCT
easat-5356	388	8	v.	v.	PROPN
easat-5356	388	9	kumar	kumar	PROPN
easat-5356	388	10	,	,	PUNCT
easat-5356	388	11	and	and	CCONJ
easat-5356	388	12	r.	r.	PROPN
easat-5356	388	13	gupta	gupta	PROPN
easat-5356	388	14	,	,	PUNCT
easat-5356	388	15	"	"	PUNCT
easat-5356	388	16	object	object	NOUN
easat-5356	388	17	detection	detection	NOUN
easat-5356	388	18	based	base	VERB
easat-5356	388	19	approaches	approach	NOUN
easat-5356	388	20	in	in	ADP
easat-5356	388	21	image	image	NOUN
easat-5356	388	22	classification	classification	NOUN
easat-5356	388	23	:	:	PUNCT
easat-5356	388	24	a	a	DET
easat-5356	388	25	brief	brief	ADJ
easat-5356	388	26	overview	overview	NOUN
easat-5356	388	27	,	,	PUNCT
easat-5356	388	28	"	"	PUNCT
easat-5356	388	29	in	in	ADP
easat-5356	388	30	2023	2023	NUM
easat-5356	388	31	ieee	ieee	NOUN
easat-5356	388	32	guwahati	guwahati	PROPN
easat-5356	388	33	subsection	subsection	NOUN
easat-5356	388	34	conference	conference	NOUN
easat-5356	388	35	(	(	PUNCT
easat-5356	388	36	gcon	gcon	NOUN
easat-5356	388	37	)	)	PUNCT
easat-5356	388	38	,	,	PUNCT
easat-5356	388	39	2023	2023	NUM
easat-5356	388	40	:	:	PUNCT
easat-5356	388	41	ieee	ieee	NOUN
easat-5356	388	42	,	,	PUNCT
easat-5356	388	43	pp	pp	ADJ
easat-5356	388	44	.	.	PUNCT
easat-5356	389	1	1	1	NUM
easat-5356	389	2	-	-	SYM
easat-5356	389	3	6	6	NUM
easat-5356	389	4	,	,	PUNCT
easat-5356	389	5	doi	doi	NOUN
easat-5356	389	6	:	:	PUNCT
easat-5356	389	7	https://doi.org/10.1109/gcon58516.2023.10183609	https://doi.org/10.1109/gcon58516.2023.10183609	NOUN
easat-5356	389	8	.	.	PUNCT
easat-5356	390	1	[	[	X
easat-5356	390	2	26	26	NUM
easat-5356	390	3	]	]	X
easat-5356	390	4	g.	g.	PROPN
easat-5356	390	5	zhao	zhao	PROPN
easat-5356	390	6	,	,	PUNCT
easat-5356	390	7	w.	w.	PROPN
easat-5356	390	8	ge	ge	PROPN
easat-5356	390	9	,	,	PUNCT
easat-5356	390	10	and	and	CCONJ
easat-5356	390	11	y.	y.	PROPN
easat-5356	390	12	yu	yu	PROPN
easat-5356	390	13	,	,	PUNCT
easat-5356	390	14	"	"	PUNCT
easat-5356	390	15	graphfpn	graphfpn	NOUN
easat-5356	390	16	:	:	PUNCT
easat-5356	390	17	graph	graph	NOUN
easat-5356	390	18	feature	feature	NOUN
easat-5356	390	19	pyramid	pyramid	NOUN
easat-5356	390	20	network	network	NOUN
easat-5356	390	21	for	for	ADP
easat-5356	390	22	object	object	NOUN
easat-5356	390	23	detection	detection	NOUN
easat-5356	390	24	,	,	PUNCT
easat-5356	390	25	"	"	PUNCT
easat-5356	390	26	in	in	ADP
easat-5356	390	27	proceedings	proceeding	NOUN
easat-5356	390	28	of	of	ADP
easat-5356	390	29	the	the	DET
easat-5356	390	30	ieee	ieee	NOUN
easat-5356	390	31	/	/	SYM
easat-5356	390	32	cvf	cvf	NOUN
easat-5356	390	33	international	international	ADJ
easat-5356	390	34	conference	conference	NOUN
easat-5356	390	35	on	on	ADP
easat-5356	390	36	computer	computer	NOUN
easat-5356	390	37	vision	vision	NOUN
easat-5356	390	38	,	,	PUNCT
easat-5356	390	39	2021	2021	NUM
easat-5356	390	40	,	,	PUNCT
easat-5356	390	41	pp	pp	ADJ
easat-5356	390	42	.	.	PUNCT
easat-5356	391	1	2763	2763	NUM
easat-5356	391	2	-	-	SYM
easat-5356	391	3	2772	2772	NUM
easat-5356	391	4	,	,	PUNCT
easat-5356	391	5	doi	doi	NOUN
easat-5356	391	6	:	:	PUNCT
easat-5356	391	7	https://doi.org/10.1109/iccv48922.2021.00276	https://doi.org/10.1109/iccv48922.2021.00276	NOUN
easat-5356	391	8	.	.	PUNCT
easat-5356	392	1	[	[	X
easat-5356	392	2	27	27	NUM
easat-5356	392	3	]	]	PUNCT
easat-5356	392	4	a.	a.	NOUN
easat-5356	392	5	m.	m.	NOUN
easat-5356	392	6	ridha	ridha	PROPN
easat-5356	392	7	,	,	PUNCT
easat-5356	392	8	m.	m.	NOUN
easat-5356	392	9	j.	j.	PROPN
easat-5356	392	10	mohammed	mohammed	PROPN
easat-5356	392	11	,	,	PUNCT
easat-5356	392	12	h.	h.	PROPN
easat-5356	392	13	a.	a.	PROPN
easat-5356	392	14	saber	saber	PROPN
easat-5356	392	15	,	,	PUNCT
easat-5356	392	16	m.	m.	NOUN
easat-5356	392	17	h.	h.	PROPN
easat-5356	392	18	chyad	chyad	PROPN
easat-5356	392	19	,	,	PUNCT
easat-5356	392	20	and	and	CCONJ
easat-5356	392	21	m.	m.	PROPN
easat-5356	392	22	h.	h.	PROPN
easat-5356	392	23	abdulameer	abdulameer	PROPN
easat-5356	392	24	,	,	PUNCT
easat-5356	392	25	"	"	PUNCT
easat-5356	392	26	advanced	advanced	ADJ
easat-5356	392	27	deep	deep	ADJ
easat-5356	392	28	learning	learning	NOUN
easat-5356	392	29	approaches	approach	NOUN
easat-5356	392	30	for	for	ADP
easat-5356	392	31	early	early	ADJ
easat-5356	392	32	detection	detection	NOUN
easat-5356	392	33	and	and	CCONJ
easat-5356	392	34	localization	localization	NOUN
easat-5356	392	35	of	of	ADP
easat-5356	392	36	ocular	ocular	ADJ
easat-5356	392	37	diseases	disease	NOUN
easat-5356	392	38	,	,	PUNCT
easat-5356	392	39	"	"	PUNCT
easat-5356	392	40	edelweiss	edelweiss	PROPN
easat-5356	392	41	applied	apply	VERB
easat-5356	392	42	science	science	NOUN
easat-5356	392	43	and	and	CCONJ
easat-5356	392	44	technology	technology	NOUN
easat-5356	392	45	,	,	PUNCT
easat-5356	392	46	vol	vol	NOUN
easat-5356	392	47	.	.	PROPN
easat-5356	392	48	8	8	NUM
easat-5356	392	49	,	,	PUNCT
easat-5356	392	50	no	no	INTJ
easat-5356	392	51	.	.	NOUN
easat-5356	392	52	6	6	NUM
easat-5356	392	53	,	,	PUNCT
easat-5356	392	54	pp	pp	ADJ
easat-5356	392	55	.	.	PUNCT
easat-5356	393	1	3708	3708	NUM
easat-5356	393	2	-	-	SYM
easat-5356	393	3	3721	3721	NUM
easat-5356	393	4	,	,	PUNCT
easat-5356	393	5	2024	2024	NUM
easat-5356	393	6	.	.	PUNCT
easat-5356	394	1	https://doi.org/10.55214/25768484.v8i6.2813	https://doi.org/10.55214/25768484.v8i6.2813	X
easat-5356	395	1	[	[	X
easat-5356	395	2	28	28	NUM
easat-5356	395	3	]	]	X
easat-5356	395	4	f.	f.	PROPN
easat-5356	395	5	yang	yang	PROPN
easat-5356	395	6	,	,	PUNCT
easat-5356	395	7	l.	l.	PROPN
easat-5356	395	8	huang	huang	PROPN
easat-5356	395	9	,	,	PUNCT
easat-5356	395	10	x.	x.	PROPN
easat-5356	395	11	tan	tan	PROPN
easat-5356	395	12	,	,	PUNCT
easat-5356	395	13	and	and	CCONJ
easat-5356	395	14	y.	y.	PROPN
easat-5356	395	15	yuan	yuan	PROPN
easat-5356	395	16	,	,	PUNCT
easat-5356	395	17	"	"	PUNCT
easat-5356	395	18	fasternet	fasternet	NOUN
easat-5356	395	19	-	-	PUNCT
easat-5356	395	20	ssd	ssd	NOUN
easat-5356	395	21	:	:	PUNCT
easat-5356	395	22	a	a	DET
easat-5356	395	23	small	small	ADJ
easat-5356	395	24	object	object	NOUN
easat-5356	395	25	detection	detection	NOUN
easat-5356	395	26	method	method	NOUN
easat-5356	395	27	based	base	VERB
easat-5356	395	28	on	on	ADP
easat-5356	395	29	ssd	ssd	PROPN
easat-5356	395	30	model	model	NOUN
easat-5356	395	31	,	,	PUNCT
easat-5356	395	32	"	"	PUNCT
easat-5356	395	33	signal	signal	NOUN
easat-5356	395	34	,	,	PUNCT
easat-5356	395	35	image	image	NOUN
easat-5356	395	36	and	and	CCONJ
easat-5356	395	37	video	video	NOUN
easat-5356	395	38	processing	processing	NOUN
easat-5356	395	39	,	,	PUNCT
easat-5356	395	40	vol	vol	NOUN
easat-5356	395	41	.	.	PROPN
easat-5356	395	42	18	18	NUM
easat-5356	395	43	,	,	PUNCT
easat-5356	395	44	no	no	INTJ
easat-5356	395	45	.	.	NOUN
easat-5356	395	46	1	1	NUM
easat-5356	395	47	,	,	PUNCT
easat-5356	395	48	pp	pp	ADJ
easat-5356	395	49	.	.	PUNCT
easat-5356	396	1	173	173	NUM
easat-5356	396	2	-	-	SYM
easat-5356	396	3	180	180	NUM
easat-5356	396	4	,	,	PUNCT
easat-5356	396	5	2024	2024	NUM
easat-5356	396	6	.	.	PUNCT
easat-5356	397	1	https://doi.org/10.1007/s11760-023-02726-5	https://doi.org/10.1007/s11760-023-02726-5	PROPN
easat-5356	398	1	[	[	X
easat-5356	398	2	29	29	NUM
easat-5356	398	3	]	]	X
easat-5356	398	4	g.	g.	PROPN
easat-5356	398	5	han	han	PROPN
easat-5356	398	6	,	,	PUNCT
easat-5356	398	7	s.	s.	PROPN
easat-5356	398	8	huang	huang	PROPN
easat-5356	398	9	,	,	PUNCT
easat-5356	398	10	j.	j.	PROPN
easat-5356	398	11	ma	ma	PROPN
easat-5356	398	12	,	,	PUNCT
easat-5356	398	13	y.	y.	PROPN
easat-5356	398	14	he	he	PRON
easat-5356	398	15	,	,	PUNCT
easat-5356	398	16	and	and	CCONJ
easat-5356	398	17	s.-f	s.-f	PROPN
easat-5356	398	18	.	.	PUNCT
easat-5356	399	1	chang	chang	PROPN
easat-5356	399	2	,	,	PUNCT
easat-5356	399	3	"	"	PUNCT
easat-5356	399	4	meta	meta	X
easat-5356	399	5	faster	fast	ADJ
easat-5356	399	6	r	r	NOUN
easat-5356	399	7	-	-	PUNCT
easat-5356	399	8	cnn	cnn	NOUN
easat-5356	399	9	:	:	PUNCT
easat-5356	399	10	towards	towards	ADP
easat-5356	399	11	accurate	accurate	ADJ
easat-5356	399	12	few	few	ADJ
easat-5356	399	13	-	-	PUNCT
easat-5356	399	14	shot	shot	NOUN
easat-5356	399	15	object	object	NOUN
easat-5356	399	16	detection	detection	NOUN
easat-5356	399	17	with	with	ADP
easat-5356	399	18	attentive	attentive	ADJ
easat-5356	399	19	feature	feature	NOUN
easat-5356	399	20	alignment	alignment	NOUN
easat-5356	399	21	,	,	PUNCT
easat-5356	399	22	"	"	PUNCT
easat-5356	399	23	in	in	ADP
easat-5356	399	24	proceedings	proceeding	NOUN
easat-5356	399	25	of	of	ADP
easat-5356	399	26	the	the	DET
easat-5356	399	27	aaai	aaai	PROPN
easat-5356	399	28	conference	conference	NOUN
easat-5356	399	29	on	on	ADP
easat-5356	399	30	artificial	artificial	ADJ
easat-5356	399	31	intelligence	intelligence	NOUN
easat-5356	399	32	,	,	PUNCT
easat-5356	399	33	2022	2022	NUM
easat-5356	399	34	,	,	PUNCT
easat-5356	399	35	vol	vol	NOUN
easat-5356	399	36	.	.	PROPN
easat-5356	399	37	36	36	NUM
easat-5356	399	38	,	,	PUNCT
easat-5356	399	39	no	no	INTJ
easat-5356	399	40	.	.	NOUN
easat-5356	399	41	1	1	NUM
easat-5356	399	42	,	,	PUNCT
easat-5356	399	43	pp	pp	ADJ
easat-5356	399	44	.	.	PUNCT
easat-5356	400	1	780	780	NUM
easat-5356	400	2	-	-	SYM
easat-5356	400	3	789	789	NUM
easat-5356	400	4	,	,	PUNCT
easat-5356	400	5	doi	doi	NOUN
easat-5356	400	6	:	:	PUNCT
easat-5356	400	7	https://doi.org/10.1609/aaai.v36i1.19959	https://doi.org/10.1609/aaai.v36i1.19959	X
easat-5356	400	8	.	.	PUNCT
easat-5356	401	1	[	[	X
easat-5356	401	2	30	30	NUM
easat-5356	401	3	]	]	X
easat-5356	401	4	y.	y.	PROPN
easat-5356	401	5	du	du	PROPN
easat-5356	401	6	,	,	PUNCT
easat-5356	401	7	h.	h.	PROPN
easat-5356	401	8	chen	chen	PROPN
easat-5356	401	9	,	,	PUNCT
easat-5356	401	10	y.	y.	PROPN
easat-5356	401	11	fu	fu	PROPN
easat-5356	401	12	,	,	PUNCT
easat-5356	401	13	j.	j.	PROPN
easat-5356	401	14	zhu	zhu	PROPN
easat-5356	401	15	,	,	PUNCT
easat-5356	401	16	and	and	CCONJ
easat-5356	401	17	h.	h.	PROPN
easat-5356	401	18	zeng	zeng	PROPN
easat-5356	401	19	,	,	PUNCT
easat-5356	401	20	"	"	PUNCT
easat-5356	401	21	aff	aff	NOUN
easat-5356	401	22	-	-	PUNCT
easat-5356	401	23	net	net	NOUN
easat-5356	401	24	:	:	PUNCT
easat-5356	401	25	a	a	DET
easat-5356	401	26	strip	strip	NOUN
easat-5356	401	27	steel	steel	NOUN
easat-5356	401	28	surface	surface	NOUN
easat-5356	401	29	defect	defect	NOUN
easat-5356	401	30	detection	detection	NOUN
easat-5356	401	31	network	network	NOUN
easat-5356	401	32	via	via	ADP
easat-5356	401	33	adaptive	adaptive	ADJ
easat-5356	401	34	focusing	focus	VERB
easat-5356	401	35	features	feature	NOUN
easat-5356	401	36	,	,	PUNCT
easat-5356	401	37	"	"	PUNCT
easat-5356	401	38	ieee	ieee	NOUN
easat-5356	401	39	transactions	transaction	NOUN
easat-5356	401	40	on	on	ADP
easat-5356	401	41	instrumentation	instrumentation	NOUN
easat-5356	401	42	and	and	CCONJ
easat-5356	401	43	measurement	measurement	NOUN
easat-5356	401	44	,	,	PUNCT
easat-5356	401	45	2024	2024	NUM
easat-5356	401	46	.	.	PUNCT
easat-5356	401	47	https://doi.org/10.1109/tim.2024.3398131	https://doi.org/10.1109/tim.2024.3398131	PUNCT
easat-5356	402	1	[	[	X
easat-5356	402	2	31	31	NUM
easat-5356	402	3	]	]	PUNCT
easat-5356	402	4	q.	q.	PROPN
easat-5356	402	5	yang	yang	PROPN
easat-5356	402	6	,	,	PUNCT
easat-5356	402	7	x.	x.	PROPN
easat-5356	402	8	shixiang	shixiang	PROPN
easat-5356	402	9	,	,	PUNCT
easat-5356	402	10	l.	l.	PROPN
easat-5356	402	11	yue	yue	PROPN
easat-5356	402	12	,	,	PUNCT
easat-5356	402	13	and	and	CCONJ
easat-5356	402	14	j.	j.	PROPN
easat-5356	402	15	jianbin	jianbin	PROPN
easat-5356	402	16	,	,	PUNCT
easat-5356	402	17	"	"	PUNCT
easat-5356	402	18	skeleton	skeleton	NOUN
easat-5356	402	19	detection	detection	NOUN
easat-5356	402	20	based	base	VERB
easat-5356	402	21	on	on	ADP
easat-5356	402	22	local	local	ADJ
easat-5356	402	23	short	short	ADJ
easat-5356	402	24	-	-	PUNCT
easat-5356	402	25	connection	connection	NOUN
easat-5356	402	26	unidirectional	unidirectional	ADJ
easat-5356	402	27	fusion	fusion	NOUN
easat-5356	402	28	networks	network	NOUN
easat-5356	402	29	,	,	PUNCT
easat-5356	402	30	"	"	PUNCT
easat-5356	402	31	journal	journal	NOUN
easat-5356	402	32	of	of	ADP
easat-5356	402	33	university	university	PROPN
easat-5356	402	34	of	of	ADP
easat-5356	402	35	chinese	chinese	PROPN
easat-5356	402	36	academy	academy	PROPN
easat-5356	402	37	of	of	ADP
easat-5356	402	38	sciences	sciences	PROPN
easat-5356	402	39	,	,	PUNCT
easat-5356	402	40	vol	vol	NOUN
easat-5356	402	41	.	.	PROPN
easat-5356	402	42	40	40	NUM
easat-5356	402	43	,	,	PUNCT
easat-5356	402	44	no	no	INTJ
easat-5356	402	45	.	.	NOUN
easat-5356	402	46	2	2	NUM
easat-5356	402	47	,	,	PUNCT
easat-5356	402	48	p.	p.	NOUN
easat-5356	402	49	250	250	NUM
easat-5356	402	50	,	,	PUNCT
easat-5356	402	51	2023	2023	NUM
easat-5356	402	52	.	.	PUNCT
easat-5356	403	1	https://doi.org/10.1016/j.neunet.2023.07.051	https://doi.org/10.1016/j.neunet.2023.07.051	NUM
easat-5356	404	1	[	[	X
easat-5356	404	2	32	32	NUM
easat-5356	404	3	]	]	PUNCT
easat-5356	404	4	y.-l	y.-l	NOUN
easat-5356	404	5	.	.	PUNCT
easat-5356	405	1	li	li	PROPN
easat-5356	405	2	,	,	PUNCT
easat-5356	405	3	y.	y.	PROPN
easat-5356	405	4	feng	feng	PROPN
easat-5356	405	5	,	,	PUNCT
easat-5356	405	6	m.-l	m.-l	PROPN
easat-5356	405	7	.	.	PUNCT
easat-5356	406	1	zhou	zhou	PROPN
easat-5356	406	2	,	,	PUNCT
easat-5356	406	3	x.-c	x.-c	PROPN
easat-5356	406	4	.	.	PUNCT
easat-5356	407	1	xiong	xiong	PROPN
easat-5356	407	2	,	,	PUNCT
easat-5356	407	3	y.-h	y.-h	PROPN
easat-5356	407	4	.	.	PUNCT
easat-5356	408	1	wang	wang	PROPN
easat-5356	408	2	,	,	PUNCT
easat-5356	408	3	and	and	CCONJ
easat-5356	408	4	b.-h	b.-h	NOUN
easat-5356	408	5	.	.	PUNCT
easat-5356	409	1	qiang	qiang	PROPN
easat-5356	409	2	,	,	PUNCT
easat-5356	409	3	"	"	PUNCT
easat-5356	409	4	dma	dma	PROPN
easat-5356	409	5	-	-	PUNCT
easat-5356	409	6	yolo	yolo	ADJ
easat-5356	409	7	:	:	PUNCT
easat-5356	409	8	multi	multi	ADJ
easat-5356	409	9	-	-	ADJ
easat-5356	409	10	scale	scale	ADJ
easat-5356	409	11	object	object	NOUN
easat-5356	409	12	detection	detection	NOUN
easat-5356	409	13	method	method	NOUN
easat-5356	409	14	with	with	ADP
easat-5356	409	15	attention	attention	NOUN
easat-5356	409	16	mechanism	mechanism	NOUN
easat-5356	409	17	for	for	ADP
easat-5356	409	18	aerial	aerial	ADJ
easat-5356	409	19	images	image	NOUN
easat-5356	409	20	,	,	PUNCT
easat-5356	409	21	"	"	PUNCT
easat-5356	409	22	the	the	DET
easat-5356	409	23	visual	visual	ADJ
easat-5356	409	24	computer	computer	NOUN
easat-5356	409	25	,	,	PUNCT
easat-5356	409	26	vol	vol	NOUN
easat-5356	409	27	.	.	PROPN
easat-5356	409	28	40	40	NUM
easat-5356	409	29	,	,	PUNCT
easat-5356	409	30	no	no	INTJ
easat-5356	409	31	.	.	NOUN
easat-5356	409	32	6	6	NUM
easat-5356	409	33	,	,	PUNCT
easat-5356	409	34	pp	pp	ADJ
easat-5356	409	35	.	.	PUNCT
easat-5356	410	1	4505	4505	NUM
easat-5356	410	2	-	-	SYM
easat-5356	410	3	4518	4518	NUM
easat-5356	410	4	,	,	PUNCT
easat-5356	410	5	2024	2024	NUM
easat-5356	410	6	.	.	PUNCT
easat-5356	411	1	https://doi.org/10.1007/s00371-023-03095-3	https://doi.org/10.1007/s00371-023-03095-3	X
easat-5356	412	1	[	[	X
easat-5356	412	2	33	33	NUM
easat-5356	412	3	]	]	PUNCT
easat-5356	412	4	m.	m.	NOUN
easat-5356	412	5	kavitha	kavitha	PROPN
easat-5356	412	6	and	and	CCONJ
easat-5356	412	7	a.	a.	NOUN
easat-5356	412	8	rajivkannan	rajivkannan	PROPN
easat-5356	412	9	,	,	PUNCT
easat-5356	412	10	"	"	PUNCT
easat-5356	412	11	hybrid	hybrid	ADJ
easat-5356	412	12	convolutional	convolutional	ADJ
easat-5356	412	13	neural	neural	ADJ
easat-5356	412	14	network	network	NOUN
easat-5356	412	15	and	and	CCONJ
easat-5356	412	16	long	long	ADJ
easat-5356	412	17	short	short	ADJ
easat-5356	412	18	-	-	PUNCT
easat-5356	412	19	term	term	NOUN
easat-5356	412	20	memory	memory	NOUN
easat-5356	412	21	approach	approach	NOUN
easat-5356	412	22	for	for	ADP
easat-5356	412	23	facial	facial	ADJ
easat-5356	412	24	expression	expression	NOUN
easat-5356	412	25	recognition	recognition	NOUN
easat-5356	412	26	,	,	PUNCT
easat-5356	412	27	"	"	PUNCT
easat-5356	412	28	intelligent	intelligent	ADJ
easat-5356	412	29	automation	automation	NOUN
easat-5356	412	30	&	&	CCONJ
easat-5356	412	31	soft	soft	ADJ
easat-5356	412	32	computing	computing	NOUN
easat-5356	412	33	,	,	PUNCT
easat-5356	412	34	vol	vol	NOUN
easat-5356	412	35	.	.	PROPN
easat-5356	412	36	35	35	NUM
easat-5356	412	37	,	,	PUNCT
easat-5356	412	38	no	no	INTJ
easat-5356	412	39	.	.	NOUN
easat-5356	412	40	1	1	NUM
easat-5356	412	41	,	,	PUNCT
easat-5356	412	42	2023	2023	NUM
easat-5356	412	43	.	.	PUNCT
easat-5356	413	1	https://doi.org/10.32604/iasc.2023.025437	https://doi.org/10.32604/iasc.2023.025437	PROPN
easat-5356	413	2	[	[	X
easat-5356	413	3	34	34	NUM
easat-5356	413	4	]	]	X
easat-5356	413	5	g.	g.	PROPN
easat-5356	413	6	wang	wang	PROPN
easat-5356	413	7	,	,	PUNCT
easat-5356	413	8	j.	j.	PROPN
easat-5356	413	9	li	li	PROPN
easat-5356	413	10	,	,	PUNCT
easat-5356	413	11	z.	z.	PROPN
easat-5356	413	12	wu	wu	PROPN
easat-5356	413	13	,	,	PUNCT
easat-5356	413	14	j.	j.	PROPN
easat-5356	413	15	xu	xu	PROPN
easat-5356	413	16	,	,	PUNCT
easat-5356	413	17	j.	j.	PROPN
easat-5356	413	18	shen	shen	PROPN
easat-5356	413	19	,	,	PUNCT
easat-5356	413	20	and	and	CCONJ
easat-5356	413	21	w.	w.	PROPN
easat-5356	413	22	yang	yang	PROPN
easat-5356	413	23	,	,	PUNCT
easat-5356	413	24	"	"	PUNCT
easat-5356	413	25	efficientface	efficientface	NOUN
easat-5356	413	26	:	:	PUNCT
easat-5356	413	27	an	an	DET
easat-5356	413	28	efficient	efficient	ADJ
easat-5356	413	29	deep	deep	ADJ
easat-5356	413	30	network	network	NOUN
easat-5356	413	31	with	with	ADP
easat-5356	413	32	feature	feature	NOUN
easat-5356	413	33	enhancement	enhancement	NOUN
easat-5356	413	34	for	for	ADP
easat-5356	413	35	accurate	accurate	ADJ
easat-5356	413	36	face	face	NOUN
easat-5356	413	37	detection	detection	NOUN
easat-5356	413	38	,	,	PUNCT
easat-5356	413	39	"	"	PUNCT
easat-5356	413	40	multimedia	multimedia	NOUN
easat-5356	413	41	systems	system	NOUN
easat-5356	413	42	,	,	PUNCT
easat-5356	413	43	vol	vol	NOUN
easat-5356	413	44	.	.	PROPN
easat-5356	413	45	29	29	NUM
easat-5356	413	46	,	,	PUNCT
easat-5356	413	47	no	no	INTJ
easat-5356	413	48	.	.	NOUN
easat-5356	413	49	5	5	NUM
easat-5356	413	50	,	,	PUNCT
easat-5356	413	51	pp	pp	ADJ
easat-5356	413	52	.	.	PUNCT
easat-5356	414	1	2825	2825	NUM
easat-5356	414	2	-	-	SYM
easat-5356	414	3	2839	2839	NUM
easat-5356	414	4	,	,	PUNCT
easat-5356	414	5	2023	2023	NUM
easat-5356	414	6	.	.	PUNCT
easat-5356	415	1	https://doi.org/10.1007/s00530-023-01134-6	https://doi.org/10.1007/s00530-023-01134-6	PROPN
easat-5356	416	1	[	[	X
easat-5356	416	2	35	35	NUM
easat-5356	416	3	]	]	PUNCT
easat-5356	416	4	z.	z.	PROPN
easat-5356	416	5	cao	cao	PROPN
easat-5356	416	6	,	,	PUNCT
easat-5356	416	7	y.	y.	PROPN
easat-5356	416	8	qin	qin	PROPN
easat-5356	416	9	,	,	PUNCT
easat-5356	416	10	y.	y.	PROPN
easat-5356	416	11	li	li	PROPN
easat-5356	416	12	,	,	PUNCT
easat-5356	416	13	z.	z.	PROPN
easat-5356	416	14	xie	xie	PROPN
easat-5356	416	15	,	,	PUNCT
easat-5356	416	16	j.	j.	PROPN
easat-5356	416	17	guo	guo	PROPN
easat-5356	416	18	,	,	PUNCT
easat-5356	416	19	and	and	CCONJ
easat-5356	416	20	l.	l.	PROPN
easat-5356	416	21	jia	jia	PROPN
easat-5356	416	22	,	,	PUNCT
easat-5356	416	23	"	"	PUNCT
easat-5356	416	24	face	face	VERB
easat-5356	416	25	detection	detection	NOUN
easat-5356	416	26	for	for	ADP
easat-5356	416	27	rail	rail	NOUN
easat-5356	416	28	transit	transit	NOUN
easat-5356	416	29	passengers	passenger	NOUN
easat-5356	416	30	based	base	VERB
easat-5356	416	31	on	on	ADP
easat-5356	416	32	single	single	ADJ
easat-5356	416	33	shot	shot	NOUN
easat-5356	416	34	detector	detector	NOUN
easat-5356	416	35	and	and	CCONJ
easat-5356	416	36	active	active	ADJ
easat-5356	416	37	learning	learning	NOUN
easat-5356	416	38	,	,	PUNCT
easat-5356	416	39	"	"	PUNCT
easat-5356	416	40	multimedia	multimedia	NOUN
easat-5356	416	41	tools	tool	NOUN
easat-5356	416	42	and	and	CCONJ
easat-5356	416	43	applications	application	NOUN
easat-5356	416	44	,	,	PUNCT
easat-5356	416	45	vol	vol	NOUN
easat-5356	416	46	.	.	PROPN
easat-5356	417	1	81	81	NUM
easat-5356	417	2	,	,	PUNCT
easat-5356	417	3	no	no	INTJ
easat-5356	417	4	.	.	NOUN
easat-5356	417	5	29	29	NUM
easat-5356	417	6	,	,	PUNCT
easat-5356	417	7	pp	pp	ADJ
easat-5356	417	8	.	.	PUNCT
easat-5356	418	1	42433	42433	NUM
easat-5356	418	2	-	-	SYM
easat-5356	418	3	42456	42456	NUM
easat-5356	418	4	,	,	PUNCT
easat-5356	418	5	2022	2022	NUM
easat-5356	418	6	.	.	PUNCT
easat-5356	419	1	https://doi.org/10.1007/s11042-022-13491-x	https://doi.org/10.1007/s11042-022-13491-x	PROPN
easat-5356	419	2	[	[	X
easat-5356	419	3	36	36	NUM
easat-5356	419	4	]	]	PUNCT
easat-5356	419	5	j.	j.	PROPN
easat-5356	419	6	hao	hao	PROPN
easat-5356	419	7	et	et	PROPN
easat-5356	419	8	al	al	PROPN
easat-5356	419	9	.	.	PROPN
easat-5356	419	10	,	,	PUNCT
easat-5356	419	11	"	"	PUNCT
easat-5356	419	12	sheep	sheep	NOUN
easat-5356	419	13	face	face	VERB
easat-5356	419	14	detection	detection	NOUN
easat-5356	419	15	based	base	VERB
easat-5356	419	16	on	on	ADP
easat-5356	419	17	an	an	DET
easat-5356	419	18	improved	improved	ADJ
easat-5356	419	19	retinaface	retinaface	NOUN
easat-5356	419	20	algorithm	algorithm	NOUN
easat-5356	419	21	,	,	PUNCT
easat-5356	419	22	"	"	PUNCT
easat-5356	419	23	animals	animal	NOUN
easat-5356	419	24	,	,	PUNCT
easat-5356	419	25	vol	vol	NOUN
easat-5356	419	26	.	.	PROPN
easat-5356	419	27	13	13	NUM
easat-5356	419	28	,	,	PUNCT
easat-5356	419	29	no	no	INTJ
easat-5356	419	30	.	.	NOUN
easat-5356	419	31	15	15	NUM
easat-5356	419	32	,	,	PUNCT
easat-5356	419	33	p.	p.	NOUN
easat-5356	419	34	2458	2458	NUM
easat-5356	419	35	,	,	PUNCT
easat-5356	419	36	2023	2023	NUM
easat-5356	419	37	.	.	PUNCT
easat-5356	420	1	https://doi.org/10.3390/ani13152458	https://doi.org/10.3390/ani13152458	PROPN
easat-5356	421	1	[	[	X
easat-5356	421	2	37	37	NUM
easat-5356	421	3	]	]	PUNCT
easat-5356	421	4	s.	s.	PROPN
easat-5356	421	5	wang	wang	PROPN
easat-5356	421	6	,	,	PUNCT
easat-5356	421	7	"	"	PUNCT
easat-5356	421	8	measurement	measurement	NOUN
easat-5356	421	9	of	of	ADP
easat-5356	421	10	the	the	DET
easat-5356	421	11	robustness	robustness	NOUN
easat-5356	421	12	and	and	CCONJ
easat-5356	421	13	sustainability	sustainability	NOUN
easat-5356	421	14	of	of	ADP
easat-5356	421	15	china	china	PROPN
easat-5356	421	16	's	's	PART
easat-5356	421	17	economic	economic	ADJ
easat-5356	421	18	development	development	NOUN
easat-5356	421	19	based	base	VERB
easat-5356	421	20	on	on	ADP
easat-5356	421	21	the	the	DET
easat-5356	421	22	shock	shock	NOUN
easat-5356	421	23	of	of	ADP
easat-5356	421	24	the	the	DET
easat-5356	421	25	covid-19	covid-19	PROPN
easat-5356	421	26	epidemic	epidemic	NOUN
easat-5356	421	27	,	,	PUNCT
easat-5356	421	28	"	"	PUNCT
easat-5356	421	29	edelweiss	edelweiss	PROPN
easat-5356	421	30	applied	apply	VERB
easat-5356	421	31	science	science	NOUN
easat-5356	421	32	and	and	CCONJ
easat-5356	421	33	technology	technology	NOUN
easat-5356	421	34	,	,	PUNCT
easat-5356	421	35	vol	vol	NOUN
easat-5356	421	36	.	.	PROPN
easat-5356	422	1	9	9	NUM
easat-5356	422	2	,	,	PUNCT
easat-5356	422	3	no	no	INTJ
easat-5356	422	4	.	.	NOUN
easat-5356	422	5	1	1	NUM
easat-5356	422	6	,	,	PUNCT
easat-5356	422	7	pp	pp	ADJ
easat-5356	422	8	.	.	PUNCT
easat-5356	423	1	173	173	NUM
easat-5356	423	2	-	-	SYM
easat-5356	423	3	197	197	NUM
easat-5356	423	4	,	,	PUNCT
easat-5356	423	5	2025	2025	NUM
easat-5356	423	6	.	.	PUNCT
easat-5356	424	1	https://doi.org/10.55214/25768484.v9i1.4022	https://doi.org/10.55214/25768484.v9i1.4022	X
easat-5356	425	1	[	[	X
easat-5356	425	2	38	38	NUM
easat-5356	425	3	]	]	PUNCT
easat-5356	425	4	l.	l.	PROPN
easat-5356	425	5	cai	cai	PROPN
easat-5356	425	6	,	,	PUNCT
easat-5356	425	7	h.	h.	PROPN
easat-5356	425	8	li	li	PROPN
easat-5356	425	9	,	,	PUNCT
easat-5356	425	10	w.	w.	PROPN
easat-5356	425	11	dong	dong	PROPN
easat-5356	425	12	,	,	PUNCT
easat-5356	425	13	and	and	CCONJ
easat-5356	425	14	h.	h.	PROPN
easat-5356	425	15	fang	fang	PROPN
easat-5356	425	16	,	,	PUNCT
easat-5356	425	17	"	"	PUNCT
easat-5356	425	18	micro	micro	ADJ
easat-5356	425	19	-	-	NOUN
easat-5356	425	20	expression	expression	ADJ
easat-5356	425	21	recognition	recognition	NOUN
easat-5356	425	22	using	use	VERB
easat-5356	425	23	3d	3d	NUM
easat-5356	425	24	densenet	densenet	NOUN
easat-5356	425	25	fused	fuse	VERB
easat-5356	425	26	squeeze	squeeze	NOUN
easat-5356	425	27	-	-	PUNCT
easat-5356	425	28	and	and	CCONJ
easat-5356	425	29	-	-	PUNCT
easat-5356	425	30	excitation	excitation	NOUN
easat-5356	425	31	networks	network	NOUN
easat-5356	425	32	,	,	PUNCT
easat-5356	425	33	"	"	PUNCT
easat-5356	425	34	applied	apply	VERB
easat-5356	425	35	soft	soft	ADJ
easat-5356	425	36	computing	computing	NOUN
easat-5356	425	37	,	,	PUNCT
easat-5356	425	38	vol	vol	NOUN
easat-5356	425	39	.	.	PROPN
easat-5356	425	40	119	119	NUM
easat-5356	425	41	,	,	PUNCT
easat-5356	425	42	p.	p.	NOUN
easat-5356	425	43	108594	108594	NUM
easat-5356	425	44	,	,	PUNCT
easat-5356	425	45	2022	2022	NUM
easat-5356	425	46	.	.	PUNCT
easat-5356	426	1	https://doi.org/10.1016/j.asoc.2022.108594	https://doi.org/10.1016/j.asoc.2022.108594	VERB
easat-5356	426	2	https://doi.org/10.1007/s11263-022-01739-w	https://doi.org/10.1007/s11263-022-01739-w	PROPN
easat-5356	426	3	https://doi.org/10.1016/j.patcog.2022.109256	https://doi.org/10.1016/j.patcog.2022.109256	PROPN
easat-5356	426	4	https://doi.org/10.1007/978-981-97-8160-7_26	https://doi.org/10.1007/978-981-97-8160-7_26	NUM
easat-5356	426	5	https://doi.org/10.1145/3611673	https://doi.org/10.1145/3611673	NOUN
easat-5356	426	6	https://doi.org/10.1007/s00371-023-02822-0	https://doi.org/10.1007/s00371-023-02822-0	PROPN
easat-5356	427	1	https://doi.org/10.1016/j.patcog.2024.111044	https://doi.org/10.1016/j.patcog.2024.111044	PROPN
easat-5356	427	2	https://doi.org/10.53730/ijhs.v6ns8.13127	https://doi.org/10.53730/ijhs.v6ns8.13127	PROPN
easat-5356	427	3	https://doi.org/10.1109/gcon58516.2023.10183609	https://doi.org/10.1109/gcon58516.2023.10183609	PROPN
easat-5356	427	4	https://doi.org/10.1109/iccv48922.2021.00276	https://doi.org/10.1109/iccv48922.2021.00276	PROPN
easat-5356	427	5	https://doi.org/10.55214/25768484.v8i6.2813	https://doi.org/10.55214/25768484.v8i6.2813	PROPN
easat-5356	427	6	https://doi.org/10.1007/s11760-023-02726-5	https://doi.org/10.1007/s11760-023-02726-5	PROPN
easat-5356	427	7	https://doi.org/10.1609/aaai.v36i1.19959	https://doi.org/10.1609/aaai.v36i1.19959	NUM
easat-5356	427	8	https://doi.org/10.1109/tim.2024.3398131	https://doi.org/10.1109/tim.2024.3398131	PROPN
easat-5356	427	9	https://doi.org/10.1016/j.neunet.2023.07.051	https://doi.org/10.1016/j.neunet.2023.07.051	PROPN
easat-5356	427	10	https://doi.org/10.1007/s00371-023-03095-3	https://doi.org/10.1007/s00371-023-03095-3	PROPN
easat-5356	427	11	https://doi.org/10.32604/iasc.2023.025437	https://doi.org/10.32604/iasc.2023.025437	PROPN
easat-5356	427	12	https://doi.org/10.1007/s00530-023-01134-6	https://doi.org/10.1007/s00530-023-01134-6	PROPN
easat-5356	428	1	https://doi.org/10.1007/s11042-022-13491-x	https://doi.org/10.1007/s11042-022-13491-x	PROPN
easat-5356	428	2	https://doi.org/10.3390/ani13152458	https://doi.org/10.3390/ani13152458	PROPN
easat-5356	428	3	https://doi.org/10.55214/25768484.v9i1.4022	https://doi.org/10.55214/25768484.v9i1.4022	PROPN
easat-5356	428	4	https://doi.org/10.1016/j.asoc.2022.108594	https://doi.org/10.1016/j.asoc.2022.108594	VERB