id	sid	tid	token	lemma	pos
ap-8720	1	1	acta	acta	PROPN
ap-8720	1	2	polytechnica	polytechnica	PROPN
ap-8720	1	3	https://doi.org/10.14311/ap.2023.63.0305	https://doi.org/10.14311/ap.2023.63.0305	PROPN
ap-8720	1	4	acta	acta	PROPN
ap-8720	1	5	polytechnica	polytechnica	PROPN
ap-8720	1	6	63(5):305–319	63(5):305–319	PROPN
ap-8720	1	7	,	,	PUNCT
ap-8720	1	8	2023	2023	NUM
ap-8720	1	9	©	©	ADP
ap-8720	1	10	2023	2023	NUM
ap-8720	1	11	the	the	DET
ap-8720	1	12	author(s	author(s	NOUN
ap-8720	1	13	)	)	PUNCT
ap-8720	1	14	.	.	PUNCT
ap-8720	2	1	licensed	license	VERB
ap-8720	2	2	under	under	ADP
ap-8720	2	3	a	a	DET
ap-8720	2	4	cc	cc	NOUN
ap-8720	2	5	-	-	PUNCT
ap-8720	2	6	by	by	ADP
ap-8720	2	7	4.0	4.0	NUM
ap-8720	2	8	licence	licence	NOUN
ap-8720	2	9	published	publish	VERB
ap-8720	2	10	by	by	ADP
ap-8720	2	11	the	the	DET
ap-8720	2	12	czech	czech	PROPN
ap-8720	2	13	technical	technical	PROPN
ap-8720	2	14	university	university	PROPN
ap-8720	2	15	in	in	ADP
ap-8720	2	16	prague	prague	PROPN
ap-8720	2	17	fake	fake	ADJ
ap-8720	2	18	face	face	NOUN
ap-8720	2	19	detection	detection	NOUN
ap-8720	2	20	based	base	VERB
ap-8720	2	21	on	on	ADP
ap-8720	2	22	a	a	DET
ap-8720	2	23	multi	multi	ADJ
ap-8720	2	24	discriminator	discriminator	NOUN
ap-8720	2	25	deep	deep	ADJ
ap-8720	2	26	cnn	cnn	PROPN
ap-8720	2	27	architecture	architecture	NOUN
ap-8720	2	28	(	(	PUNCT
ap-8720	2	29	mdd	mdd	PROPN
ap-8720	2	30	-	-	PUNCT
ap-8720	2	31	cnn	cnn	PROPN
ap-8720	2	32	)	)	PUNCT
ap-8720	2	33	chemesse	chemesse	NOUN
ap-8720	2	34	ennehar	ennehar	PROPN
ap-8720	2	35	bencherieta,∗	bencherieta,∗	PROPN
ap-8720	2	36	,	,	PUNCT
ap-8720	2	37	hiba	hiba	PROPN
ap-8720	2	38	abdelmoumèneb	abdelmoumèneb	PROPN
ap-8720	2	39	,	,	PUNCT
ap-8720	2	40	abdennour	abdennour	ADJ
ap-8720	2	41	sebbaghc	sebbaghc	NOUN
ap-8720	2	42	,	,	PUNCT
ap-8720	2	43	abdennour	abdennour	VERB
ap-8720	2	44	yahiyaouid	yahiyaouid	ADJ
ap-8720	2	45	,	,	PUNCT
ap-8720	2	46	zahra	zahra	NOUN
ap-8720	2	47	tabad	tabad	VERB
ap-8720	2	48	a	a	DET
ap-8720	2	49	8	8	NUM
ap-8720	2	50	mai	mai	PROPN
ap-8720	2	51	1945	1945	NUM
ap-8720	2	52	-	-	PUNCT
ap-8720	2	53	guelma	guelma	PROPN
ap-8720	2	54	university	university	NOUN
ap-8720	2	55	,	,	PUNCT
ap-8720	2	56	computer	computer	NOUN
ap-8720	2	57	science	science	PROPN
ap-8720	2	58	department	department	PROPN
ap-8720	2	59	/	/	SYM
ap-8720	2	60	laig	laig	PROPN
ap-8720	2	61	laboratory	laboratory	NOUN
ap-8720	2	62	,	,	PUNCT
ap-8720	2	63	24000	24000	NUM
ap-8720	2	64	guelma	guelma	ADJ
ap-8720	2	65	,	,	PUNCT
ap-8720	2	66	algeria	algeria	PROPN
ap-8720	2	67	b	b	PROPN
ap-8720	2	68	8	8	NUM
ap-8720	2	69	mai	mai	PROPN
ap-8720	2	70	1945	1945	NUM
ap-8720	2	71	-	-	PUNCT
ap-8720	2	72	guelma	guelma	PROPN
ap-8720	2	73	university	university	NOUN
ap-8720	2	74	,	,	PUNCT
ap-8720	2	75	computer	computer	NOUN
ap-8720	2	76	science	science	PROPN
ap-8720	2	77	department	department	PROPN
ap-8720	2	78	/	/	SYM
ap-8720	2	79	labged	labged	ADJ
ap-8720	2	80	laboratory	laboratory	NOUN
ap-8720	2	81	,	,	PUNCT
ap-8720	2	82	24000	24000	NUM
ap-8720	2	83	guelma	guelma	ADJ
ap-8720	2	84	,	,	PUNCT
ap-8720	2	85	algeria	algeria	PROPN
ap-8720	2	86	c	c	PROPN
ap-8720	2	87	8	8	NUM
ap-8720	2	88	mai	mai	PROPN
ap-8720	2	89	1945	1945	NUM
ap-8720	2	90	-	-	PUNCT
ap-8720	2	91	guelma	guelma	PROPN
ap-8720	2	92	university	university	NOUN
ap-8720	2	93	,	,	PUNCT
ap-8720	2	94	department	department	NOUN
ap-8720	2	95	of	of	ADP
ap-8720	2	96	electrical	electrical	ADJ
ap-8720	2	97	atomatic	atomatic	ADJ
ap-8720	2	98	engineering	engineering	NOUN
ap-8720	2	99	/	/	SYM
ap-8720	2	100	laig	laig	PROPN
ap-8720	2	101	laboratory	laboratory	NOUN
ap-8720	2	102	,	,	PUNCT
ap-8720	2	103	24000	24000	NUM
ap-8720	2	104	guelma	guelma	ADJ
ap-8720	2	105	,	,	PUNCT
ap-8720	2	106	algeria	algeria	PROPN
ap-8720	2	107	d	d	PROPN
ap-8720	2	108	8	8	NUM
ap-8720	2	109	mai	mai	PROPN
ap-8720	2	110	1945	1945	NUM
ap-8720	2	111	-	-	PUNCT
ap-8720	2	112	guelma	guelma	PROPN
ap-8720	2	113	university	university	NOUN
ap-8720	2	114	,	,	PUNCT
ap-8720	2	115	computer	computer	NOUN
ap-8720	2	116	science	science	PROPN
ap-8720	2	117	department	department	PROPN
ap-8720	2	118	,	,	PUNCT
ap-8720	2	119	24000	24000	NUM
ap-8720	2	120	guelma	guelma	ADJ
ap-8720	2	121	,	,	PUNCT
ap-8720	2	122	algeria	algeria	PROPN
ap-8720	2	123	∗	∗	NOUN
ap-8720	2	124	corresponding	correspond	VERB
ap-8720	2	125	author	author	NOUN
ap-8720	2	126	:	:	PUNCT
ap-8720	2	127	cbencheriet@yahoo.fr	cbencheriet@yahoo.fr	NOUN
ap-8720	2	128	abstract	abstract	NOUN
ap-8720	2	129	.	.	PUNCT
ap-8720	3	1	due	due	ADP
ap-8720	3	2	to	to	ADP
ap-8720	3	3	the	the	DET
ap-8720	3	4	robustness	robustness	NOUN
ap-8720	3	5	of	of	ADP
ap-8720	3	6	the	the	DET
ap-8720	3	7	deep	deep	ADJ
ap-8720	3	8	learning	learning	NOUN
ap-8720	3	9	tools	tool	NOUN
ap-8720	3	10	used	use	VERB
ap-8720	3	11	to	to	PART
ap-8720	3	12	design	design	VERB
ap-8720	3	13	these	these	DET
ap-8720	3	14	applications	application	NOUN
ap-8720	3	15	,	,	PUNCT
ap-8720	3	16	fakes	fake	NOUN
ap-8720	3	17	are	be	AUX
ap-8720	3	18	becoming	become	VERB
ap-8720	3	19	increasingly	increasingly	ADV
ap-8720	3	20	common	common	ADJ
ap-8720	3	21	as	as	SCONJ
ap-8720	3	22	these	these	DET
ap-8720	3	23	applications	application	NOUN
ap-8720	3	24	become	become	VERB
ap-8720	3	25	more	more	ADV
ap-8720	3	26	widely	widely	ADV
ap-8720	3	27	available	available	ADJ
ap-8720	3	28	and	and	CCONJ
ap-8720	3	29	accessible	accessible	ADJ
ap-8720	3	30	to	to	ADP
ap-8720	3	31	the	the	DET
ap-8720	3	32	general	general	ADJ
ap-8720	3	33	public	public	NOUN
ap-8720	3	34	.	.	PUNCT
ap-8720	4	1	these	these	DET
ap-8720	4	2	fakes	fake	NOUN
ap-8720	4	3	are	be	AUX
ap-8720	4	4	typically	typically	ADV
ap-8720	4	5	fake	fake	ADJ
ap-8720	4	6	faces	face	NOUN
ap-8720	4	7	or	or	CCONJ
ap-8720	4	8	even	even	ADV
ap-8720	4	9	fake	fake	ADJ
ap-8720	4	10	people	people	NOUN
ap-8720	4	11	,	,	PUNCT
ap-8720	4	12	which	which	PRON
ap-8720	4	13	are	be	AUX
ap-8720	4	14	difficult	difficult	ADJ
ap-8720	4	15	to	to	PART
ap-8720	4	16	distinguish	distinguish	VERB
ap-8720	4	17	from	from	ADP
ap-8720	4	18	real	real	ADJ
ap-8720	4	19	individuals	individual	NOUN
ap-8720	4	20	.	.	PUNCT
ap-8720	5	1	therefore	therefore	ADV
ap-8720	5	2	,	,	PUNCT
ap-8720	5	3	we	we	PRON
ap-8720	5	4	need	need	VERB
ap-8720	5	5	more	more	ADV
ap-8720	5	6	efficient	efficient	ADJ
ap-8720	5	7	applications	application	NOUN
ap-8720	5	8	for	for	ADP
ap-8720	5	9	fraud	fraud	NOUN
ap-8720	5	10	detection	detection	NOUN
ap-8720	5	11	.	.	PUNCT
ap-8720	6	1	in	in	ADP
ap-8720	6	2	this	this	DET
ap-8720	6	3	work	work	NOUN
ap-8720	6	4	,	,	PUNCT
ap-8720	6	5	we	we	PRON
ap-8720	6	6	propose	propose	VERB
ap-8720	6	7	a	a	DET
ap-8720	6	8	new	new	ADJ
ap-8720	6	9	multi	multi	ADJ
ap-8720	6	10	-	-	ADJ
ap-8720	6	11	discriminator	discriminator	ADJ
ap-8720	6	12	architecture	architecture	NOUN
ap-8720	6	13	to	to	PART
ap-8720	6	14	distinguish	distinguish	VERB
ap-8720	6	15	fake	fake	ADJ
ap-8720	6	16	faces	face	NOUN
ap-8720	6	17	from	from	ADP
ap-8720	6	18	real	real	ADJ
ap-8720	6	19	ones	one	NOUN
ap-8720	6	20	.	.	PUNCT
ap-8720	7	1	the	the	DET
ap-8720	7	2	architecture	architecture	NOUN
ap-8720	7	3	consists	consist	VERB
ap-8720	7	4	of	of	ADP
ap-8720	7	5	three	three	NUM
ap-8720	7	6	deep	deep	ADJ
ap-8720	7	7	networks	network	NOUN
ap-8720	7	8	(	(	PUNCT
ap-8720	7	9	discriminators	discriminator	NOUN
ap-8720	7	10	)	)	PUNCT
ap-8720	7	11	competing	compete	VERB
ap-8720	7	12	with	with	ADP
ap-8720	7	13	each	each	DET
ap-8720	7	14	other	other	ADJ
ap-8720	7	15	,	,	PUNCT
ap-8720	7	16	each	each	PRON
ap-8720	7	17	trained	train	VERB
ap-8720	7	18	differently	differently	ADV
ap-8720	7	19	.	.	PUNCT
ap-8720	8	1	the	the	DET
ap-8720	8	2	final	final	ADJ
ap-8720	8	3	decision	decision	NOUN
ap-8720	8	4	is	be	AUX
ap-8720	8	5	made	make	VERB
ap-8720	8	6	by	by	ADP
ap-8720	8	7	voting	voting	NOUN
ap-8720	8	8	based	base	VERB
ap-8720	8	9	on	on	ADP
ap-8720	8	10	the	the	DET
ap-8720	8	11	decisions	decision	NOUN
ap-8720	8	12	of	of	ADP
ap-8720	8	13	the	the	DET
ap-8720	8	14	three	three	NUM
ap-8720	8	15	discriminators	discriminator	NOUN
ap-8720	8	16	.	.	PUNCT
ap-8720	9	1	the	the	DET
ap-8720	9	2	core	core	NOUN
ap-8720	9	3	element	element	NOUN
ap-8720	9	4	of	of	ADP
ap-8720	9	5	our	our	PRON
ap-8720	9	6	architecture	architecture	NOUN
ap-8720	9	7	is	be	AUX
ap-8720	9	8	the	the	DET
ap-8720	9	9	proposed	propose	VERB
ap-8720	9	10	new	new	ADJ
ap-8720	9	11	adversarial	adversarial	ADJ
ap-8720	9	12	deep	deep	ADJ
ap-8720	9	13	network	network	NOUN
ap-8720	9	14	discriminator	discriminator	NOUN
ap-8720	9	15	(	(	PUNCT
ap-8720	9	16	ndgan	ndgan	PROPN
ap-8720	9	17	)	)	PUNCT
ap-8720	9	18	,	,	PUNCT
ap-8720	9	19	which	which	PRON
ap-8720	9	20	is	be	AUX
ap-8720	9	21	trained	train	VERB
ap-8720	9	22	in	in	ADP
ap-8720	9	23	three	three	NUM
ap-8720	9	24	different	different	ADJ
ap-8720	9	25	ways	way	NOUN
ap-8720	9	26	,	,	PUNCT
ap-8720	9	27	resulting	result	VERB
ap-8720	9	28	in	in	ADP
ap-8720	9	29	three	three	NUM
ap-8720	9	30	distinct	distinct	ADJ
ap-8720	9	31	discriminators	discriminator	NOUN
ap-8720	9	32	.	.	PUNCT
ap-8720	10	1	discriminator	discriminator	NOUN
ap-8720	10	2	1	1	NUM
ap-8720	10	3	undergoes	undergo	VERB
ap-8720	10	4	adversarial	adversarial	ADJ
ap-8720	10	5	training	training	NOUN
ap-8720	10	6	,	,	PUNCT
ap-8720	10	7	discriminator	discriminator	NOUN
ap-8720	10	8	2	2	NUM
ap-8720	10	9	is	be	AUX
ap-8720	10	10	trained	train	VERB
ap-8720	10	11	using	use	VERB
ap-8720	10	12	transfer	transfer	NOUN
ap-8720	10	13	learning	learning	NOUN
ap-8720	10	14	,	,	PUNCT
ap-8720	10	15	and	and	CCONJ
ap-8720	10	16	the	the	DET
ap-8720	10	17	third	third	ADJ
ap-8720	10	18	discriminator	discriminator	NOUN
ap-8720	10	19	undergoes	undergoe	NOUN
ap-8720	10	20	supervised	supervise	VERB
ap-8720	10	21	training	training	NOUN
ap-8720	10	22	with	with	ADP
ap-8720	10	23	a	a	DET
ap-8720	10	24	standard	standard	ADJ
ap-8720	10	25	cnn	cnn	NOUN
ap-8720	10	26	using	use	VERB
ap-8720	10	27	examples	example	NOUN
ap-8720	10	28	and	and	CCONJ
ap-8720	10	29	counterexamples	counterexample	NOUN
ap-8720	10	30	.	.	PUNCT
ap-8720	11	1	training	training	NOUN
ap-8720	11	2	and	and	CCONJ
ap-8720	11	3	testing	testing	NOUN
ap-8720	11	4	were	be	AUX
ap-8720	11	5	performed	perform	VERB
ap-8720	11	6	on	on	ADP
ap-8720	11	7	70	70	NUM
ap-8720	11	8	000	000	NUM
ap-8720	11	9	real	real	ADJ
ap-8720	11	10	faces	face	NOUN
ap-8720	11	11	from	from	ADP
ap-8720	11	12	the	the	DET
ap-8720	11	13	flickr	flickr	PROPN
ap-8720	11	14	-	-	PUNCT
ap-8720	11	15	face	face	NOUN
ap-8720	11	16	-	-	PUNCT
ap-8720	11	17	hq	hq	NOUN
ap-8720	11	18	(	(	PUNCT
ap-8720	11	19	ffhq	ffhq	NOUN
ap-8720	11	20	)	)	PUNCT
ap-8720	11	21	dataset	dataset	NOUN
ap-8720	11	22	,	,	PUNCT
ap-8720	11	23	while	while	SCONJ
ap-8720	11	24	70	70	NUM
ap-8720	11	25	000	000	NUM
ap-8720	11	26	fake	fake	ADJ
ap-8720	11	27	faces	face	NOUN
ap-8720	11	28	were	be	AUX
ap-8720	11	29	generated	generate	VERB
ap-8720	11	30	using	use	VERB
ap-8720	11	31	nvidia	nvidia	PROPN
ap-8720	11	32	’s	’s	PART
ap-8720	11	33	stylegan	stylegan	NOUN
ap-8720	11	34	.	.	PUNCT
ap-8720	12	1	the	the	DET
ap-8720	12	2	tests	test	NOUN
ap-8720	12	3	conducted	conduct	VERB
ap-8720	12	4	on	on	ADP
ap-8720	12	5	the	the	DET
ap-8720	12	6	three	three	NUM
ap-8720	12	7	networks	network	NOUN
ap-8720	12	8	produced	produce	VERB
ap-8720	12	9	significant	significant	ADJ
ap-8720	12	10	results	result	NOUN
ap-8720	12	11	,	,	PUNCT
ap-8720	12	12	with	with	ADP
ap-8720	12	13	accuracy	accuracy	NOUN
ap-8720	12	14	ranging	range	VERB
ap-8720	12	15	from	from	ADP
ap-8720	12	16	79	79	NUM
ap-8720	12	17	%	%	NOUN
ap-8720	12	18	to	to	ADP
ap-8720	12	19	98	98	NUM
ap-8720	12	20	%	%	NOUN
ap-8720	12	21	for	for	ADP
ap-8720	12	22	fake	fake	ADJ
ap-8720	12	23	faces	face	NOUN
ap-8720	12	24	,	,	PUNCT
ap-8720	12	25	and	and	CCONJ
ap-8720	12	26	from	from	ADP
ap-8720	12	27	80	80	NUM
ap-8720	12	28	%	%	NOUN
ap-8720	12	29	to	to	ADP
ap-8720	12	30	98	98	NUM
ap-8720	12	31	%	%	NOUN
ap-8720	12	32	for	for	ADP
ap-8720	12	33	real	real	ADJ
ap-8720	12	34	faces	face	NOUN
ap-8720	12	35	.	.	PUNCT
ap-8720	13	1	the	the	DET
ap-8720	13	2	reliability	reliability	NOUN
ap-8720	13	3	of	of	ADP
ap-8720	13	4	the	the	DET
ap-8720	13	5	discriminators	discriminator	NOUN
ap-8720	13	6	contributes	contribute	VERB
ap-8720	13	7	significantly	significantly	ADV
ap-8720	13	8	to	to	ADP
ap-8720	13	9	the	the	DET
ap-8720	13	10	overall	overall	ADJ
ap-8720	13	11	performance	performance	NOUN
ap-8720	13	12	of	of	ADP
ap-8720	13	13	the	the	DET
ap-8720	13	14	multi	multi	ADJ
ap-8720	13	15	-	-	ADJ
ap-8720	13	16	discriminator	discriminator	ADJ
ap-8720	13	17	system	system	NOUN
ap-8720	13	18	,	,	PUNCT
ap-8720	13	19	achieving	achieve	VERB
ap-8720	13	20	an	an	DET
ap-8720	13	21	accuracy	accuracy	NOUN
ap-8720	13	22	of	of	ADP
ap-8720	13	23	96	96	NUM
ap-8720	13	24	%	%	NOUN
ap-8720	13	25	for	for	ADP
ap-8720	13	26	fake	fake	ADJ
ap-8720	13	27	faces	face	NOUN
ap-8720	13	28	and	and	CCONJ
ap-8720	13	29	98	98	NUM
ap-8720	13	30	%	%	NOUN
ap-8720	13	31	for	for	ADP
ap-8720	13	32	real	real	ADJ
ap-8720	13	33	faces	face	NOUN
ap-8720	13	34	.	.	PUNCT
ap-8720	14	1	keywords	keyword	NOUN
ap-8720	14	2	:	:	PUNCT
ap-8720	14	3	fake	fake	ADJ
ap-8720	14	4	face	face	NOUN
ap-8720	14	5	,	,	PUNCT
ap-8720	14	6	real	real	ADJ
ap-8720	14	7	face	face	NOUN
ap-8720	14	8	,	,	PUNCT
ap-8720	14	9	discriminator	discriminator	NOUN
ap-8720	14	10	,	,	PUNCT
ap-8720	14	11	mdd	mdd	PROPN
ap-8720	14	12	-	-	PUNCT
ap-8720	14	13	cnn	cnn	PROPN
ap-8720	14	14	architecture	architecture	NOUN
ap-8720	14	15	adversarial	adversarial	ADJ
ap-8720	14	16	training	training	NOUN
ap-8720	14	17	,	,	PUNCT
ap-8720	14	18	transfer	transfer	NOUN
ap-8720	14	19	learning	learning	NOUN
ap-8720	14	20	,	,	PUNCT
ap-8720	14	21	deep	deep	ADJ
ap-8720	14	22	learning	learning	NOUN
ap-8720	14	23	.	.	PUNCT
ap-8720	15	1	1	1	X
ap-8720	15	2	.	.	X
ap-8720	15	3	introduction	introduction	NOUN
ap-8720	15	4	false	false	ADJ
ap-8720	15	5	images	image	NOUN
ap-8720	15	6	have	have	AUX
ap-8720	15	7	become	become	VERB
ap-8720	15	8	a	a	DET
ap-8720	15	9	major	major	ADJ
ap-8720	15	10	problem	problem	NOUN
ap-8720	15	11	in	in	ADP
ap-8720	15	12	recent	recent	ADJ
ap-8720	15	13	years	year	NOUN
ap-8720	15	14	.	.	PUNCT
ap-8720	16	1	they	they	PRON
ap-8720	16	2	can	can	AUX
ap-8720	16	3	easily	easily	ADV
ap-8720	16	4	deceive	deceive	VERB
ap-8720	16	5	people	people	NOUN
ap-8720	16	6	and	and	CCONJ
ap-8720	16	7	inevitably	inevitably	ADV
ap-8720	16	8	lead	lead	VERB
ap-8720	16	9	to	to	ADP
ap-8720	16	10	serious	serious	ADJ
ap-8720	16	11	social	social	ADJ
ap-8720	16	12	risks	risk	NOUN
ap-8720	16	13	,	,	PUNCT
ap-8720	16	14	such	such	ADJ
ap-8720	16	15	as	as	ADP
ap-8720	16	16	false	false	ADJ
ap-8720	16	17	evidence	evidence	NOUN
ap-8720	16	18	and	and	CCONJ
ap-8720	16	19	false	false	ADJ
ap-8720	16	20	information	information	NOUN
ap-8720	16	21	.	.	PUNCT
ap-8720	17	1	they	they	PRON
ap-8720	17	2	can	can	AUX
ap-8720	17	3	also	also	ADV
ap-8720	17	4	damage	damage	VERB
ap-8720	17	5	reputations	reputation	NOUN
ap-8720	17	6	and	and	CCONJ
ap-8720	17	7	harm	harm	VERB
ap-8720	17	8	people	people	NOUN
ap-8720	17	9	.	.	PUNCT
ap-8720	18	1	today	today	NOUN
ap-8720	18	2	,	,	PUNCT
ap-8720	18	3	fake	fake	ADJ
ap-8720	18	4	faces	face	NOUN
ap-8720	18	5	can	can	AUX
ap-8720	18	6	be	be	AUX
ap-8720	18	7	generated	generate	VERB
ap-8720	18	8	with	with	ADP
ap-8720	18	9	minimal	minimal	ADJ
ap-8720	18	10	effort	effort	NOUN
ap-8720	18	11	and	and	CCONJ
ap-8720	18	12	by	by	ADP
ap-8720	18	13	anyone	anyone	PRON
ap-8720	18	14	with	with	ADP
ap-8720	18	15	sufficient	sufficient	ADJ
ap-8720	18	16	equipment	equipment	NOUN
ap-8720	18	17	.	.	PUNCT
ap-8720	19	1	detecting	detect	VERB
ap-8720	19	2	deppfakes	deppfake	NOUN
ap-8720	19	3	is	be	AUX
ap-8720	19	4	tberefore	tberefore	NOUN
ap-8720	19	5	becoming	become	VERB
ap-8720	19	6	increasingly	increasingly	ADV
ap-8720	19	7	important	important	ADJ
ap-8720	19	8	.	.	PUNCT
ap-8720	20	1	the	the	DET
ap-8720	20	2	act	act	NOUN
ap-8720	20	3	of	of	ADP
ap-8720	20	4	image	image	NOUN
ap-8720	20	5	forgery	forgery	NOUN
ap-8720	20	6	is	be	AUX
ap-8720	20	7	not	not	PART
ap-8720	20	8	new	new	ADJ
ap-8720	20	9	,	,	PUNCT
ap-8720	20	10	in	in	ADP
ap-8720	20	11	1997	1997	NUM
ap-8720	20	12	[	[	X
ap-8720	20	13	1	1	NUM
ap-8720	20	14	]	]	PUNCT
ap-8720	20	15	,	,	PUNCT
ap-8720	20	16	a	a	DET
ap-8720	20	17	project	project	NOUN
ap-8720	20	18	called	call	VERB
ap-8720	20	19	“	"	PUNCT
ap-8720	20	20	video	video	NOUN
ap-8720	20	21	rewrite	rewrite	NOUN
ap-8720	20	22	”	"	PUNCT
ap-8720	20	23	altered	alter	VERB
ap-8720	20	24	existing	exist	VERB
ap-8720	20	25	video	video	NOUN
ap-8720	20	26	footage	footage	NOUN
ap-8720	20	27	of	of	ADP
ap-8720	20	28	people	people	NOUN
ap-8720	20	29	to	to	PART
ap-8720	20	30	make	make	VERB
ap-8720	20	31	them	they	PRON
ap-8720	20	32	appear	appear	VERB
ap-8720	20	33	to	to	PART
ap-8720	20	34	be	be	AUX
ap-8720	20	35	saying	say	VERB
ap-8720	20	36	words	word	NOUN
ap-8720	20	37	that	that	PRON
ap-8720	20	38	appeared	appear	VERB
ap-8720	20	39	on	on	ADP
ap-8720	20	40	an	an	DET
ap-8720	20	41	altered	altered	ADJ
ap-8720	20	42	audio	audio	NOUN
ap-8720	20	43	track	track	NOUN
ap-8720	20	44	:	:	PUNCT
ap-8720	20	45	essentially	essentially	ADV
ap-8720	20	46	,	,	PUNCT
ap-8720	20	47	putting	put	VERB
ap-8720	20	48	words	word	NOUN
ap-8720	20	49	in	in	ADP
ap-8720	20	50	their	their	PRON
ap-8720	20	51	mouths	mouth	NOUN
ap-8720	20	52	.	.	PUNCT
ap-8720	21	1	the	the	DET
ap-8720	21	2	early	early	ADJ
ap-8720	21	3	2000s	2000s	NUM
ap-8720	21	4	were	be	AUX
ap-8720	21	5	relatively	relatively	ADV
ap-8720	21	6	quiet	quiet	ADJ
ap-8720	21	7	as	as	SCONJ
ap-8720	21	8	computer	computer	NOUN
ap-8720	21	9	development	development	NOUN
ap-8720	21	10	moved	move	VERB
ap-8720	21	11	deeper	deeply	ADV
ap-8720	21	12	into	into	ADP
ap-8720	21	13	the	the	DET
ap-8720	21	14	world	world	NOUN
ap-8720	21	15	of	of	ADP
ap-8720	21	16	facial	facial	ADJ
ap-8720	21	17	recognition	recognition	NOUN
ap-8720	21	18	.	.	PUNCT
ap-8720	22	1	developments	development	NOUN
ap-8720	22	2	in	in	ADP
ap-8720	22	3	this	this	DET
ap-8720	22	4	field	field	NOUN
ap-8720	22	5	led	lead	VERB
ap-8720	22	6	to	to	ADP
ap-8720	22	7	dramatic	dramatic	ADJ
ap-8720	22	8	improvements	improvement	NOUN
ap-8720	22	9	in	in	ADP
ap-8720	22	10	motion	motion	NOUN
ap-8720	22	11	tracking	tracking	NOUN
ap-8720	22	12	that	that	PRON
ap-8720	22	13	makes	make	VERB
ap-8720	22	14	today	today	NOUN
ap-8720	22	15	’s	’s	PART
ap-8720	22	16	deepfakes	deepfake	VERB
ap-8720	22	17	more	more	ADV
ap-8720	22	18	convincing	convincing	ADJ
ap-8720	22	19	.	.	PUNCT
ap-8720	23	1	in	in	ADP
ap-8720	23	2	2001	2001	NUM
ap-8720	23	3	[	[	X
ap-8720	23	4	2	2	NUM
ap-8720	23	5	]	]	PUNCT
ap-8720	23	6	,	,	PUNCT
ap-8720	23	7	the	the	DET
ap-8720	23	8	active	active	ADJ
ap-8720	23	9	appearance	appearance	NOUN
ap-8720	23	10	models	model	NOUN
ap-8720	23	11	algorithm	algorithm	PROPN
ap-8720	23	12	made	make	VERB
ap-8720	23	13	its	its	PRON
ap-8720	23	14	debut	debut	NOUN
ap-8720	23	15	.	.	PUNCT
ap-8720	24	1	using	use	VERB
ap-8720	24	2	a	a	DET
ap-8720	24	3	thorough	thorough	ADJ
ap-8720	24	4	statistical	statistical	ADJ
ap-8720	24	5	model	model	NOUN
ap-8720	24	6	to	to	PART
ap-8720	24	7	match	match	VERB
ap-8720	24	8	a	a	DET
ap-8720	24	9	shape	shape	NOUN
ap-8720	24	10	to	to	ADP
ap-8720	24	11	an	an	DET
ap-8720	24	12	image	image	NOUN
ap-8720	24	13	proved	prove	VERB
ap-8720	24	14	to	to	PART
ap-8720	24	15	be	be	AUX
ap-8720	24	16	a	a	DET
ap-8720	24	17	big	big	ADJ
ap-8720	24	18	step	step	NOUN
ap-8720	24	19	forward	forward	ADV
ap-8720	24	20	.	.	PUNCT
ap-8720	25	1	this	this	PRON
ap-8720	25	2	made	make	VERB
ap-8720	25	3	face	face	NOUN
ap-8720	25	4	matching	match	VERB
ap-8720	25	5	and	and	CCONJ
ap-8720	25	6	tracking	track	VERB
ap-8720	25	7	much	much	ADV
ap-8720	25	8	more	more	ADV
ap-8720	25	9	efficient	efficient	ADJ
ap-8720	25	10	.	.	PUNCT
ap-8720	26	1	face	face	NOUN
ap-8720	26	2	manipulation	manipulation	NOUN
ap-8720	26	3	is	be	AUX
ap-8720	26	4	generally	generally	ADV
ap-8720	26	5	divided	divide	VERB
ap-8720	26	6	into	into	ADP
ap-8720	26	7	two	two	NUM
ap-8720	26	8	categories	category	NOUN
ap-8720	26	9	,	,	PUNCT
ap-8720	26	10	faceswap	faceswap	NOUN
ap-8720	26	11	(	(	PUNCT
ap-8720	26	12	facial	facial	ADJ
ap-8720	26	13	change	change	NOUN
ap-8720	26	14	)	)	PUNCT
ap-8720	26	15	,	,	PUNCT
ap-8720	26	16	and	and	CCONJ
ap-8720	26	17	facial	facial	ADJ
ap-8720	26	18	expression	expression	NOUN
ap-8720	26	19	reconstruction	reconstruction	NOUN
ap-8720	26	20	.	.	PUNCT
ap-8720	27	1	faceswap	faceswap	NOUN
ap-8720	27	2	and	and	CCONJ
ap-8720	27	3	deepfakes	deepfake	NOUN
ap-8720	27	4	belong	belong	VERB
ap-8720	27	5	to	to	ADP
ap-8720	27	6	the	the	DET
ap-8720	27	7	same	same	ADJ
ap-8720	27	8	category	category	NOUN
ap-8720	27	9	:	:	PUNCT
ap-8720	27	10	a	a	DET
ap-8720	27	11	target	target	NOUN
ap-8720	27	12	face	face	NOUN
ap-8720	27	13	replaces	replace	VERB
ap-8720	27	14	the	the	DET
ap-8720	27	15	face	face	NOUN
ap-8720	27	16	in	in	ADP
ap-8720	27	17	the	the	DET
ap-8720	27	18	source	source	NOUN
ap-8720	27	19	image	image	NOUN
ap-8720	27	20	(	(	PUNCT
ap-8720	27	21	see	see	VERB
ap-8720	27	22	figure	figure	NOUN
ap-8720	27	23	1	1	NUM
ap-8720	27	24	)	)	PUNCT
ap-8720	27	25	.	.	PUNCT
ap-8720	28	1	faceswap	faceswap	PROPN
ap-8720	28	2	is	be	AUX
ap-8720	28	3	a	a	DET
ap-8720	28	4	graphical	graphical	ADJ
ap-8720	28	5	approach	approach	NOUN
ap-8720	28	6	,	,	PUNCT
ap-8720	28	7	while	while	SCONJ
ap-8720	28	8	deepfakes	deepfake	NOUN
ap-8720	28	9	is	be	AUX
ap-8720	28	10	based	base	VERB
ap-8720	28	11	on	on	ADP
ap-8720	28	12	gans	gan	NOUN
ap-8720	28	13	.	.	PUNCT
ap-8720	29	1	mainly	mainly	ADV
ap-8720	29	2	the	the	DET
ap-8720	29	3	face2face	face2face	NOUN
ap-8720	29	4	and	and	CCONJ
ap-8720	29	5	neural	neural	ADJ
ap-8720	29	6	textures	texture	NOUN
ap-8720	29	7	methods	method	NOUN
ap-8720	29	8	transfer	transfer	VERB
ap-8720	29	9	facial	facial	ADJ
ap-8720	29	10	expressions	expression	NOUN
ap-8720	29	11	from	from	ADP
ap-8720	29	12	the	the	DET
ap-8720	29	13	target	target	NOUN
ap-8720	29	14	image	image	NOUN
ap-8720	29	15	to	to	ADP
ap-8720	29	16	the	the	DET
ap-8720	29	17	source	source	NOUN
ap-8720	29	18	image	image	NOUN
ap-8720	29	19	.	.	PUNCT
ap-8720	30	1	the	the	DET
ap-8720	30	2	face2face	face2face	NOUN
ap-8720	30	3	method	method	NOUN
ap-8720	30	4	transfers	transfer	VERB
ap-8720	30	5	the	the	DET
ap-8720	30	6	expression	expression	NOUN
ap-8720	30	7	from	from	ADP
ap-8720	30	8	the	the	DET
ap-8720	30	9	source	source	NOUN
ap-8720	30	10	video	video	NOUN
ap-8720	30	11	to	to	ADP
ap-8720	30	12	the	the	DET
ap-8720	30	13	target	target	NOUN
ap-8720	30	14	video	video	NOUN
ap-8720	30	15	using	use	VERB
ap-8720	30	16	blend	blend	NOUN
ap-8720	30	17	shape	shape	NOUN
ap-8720	30	18	coefficients	coefficient	NOUN
ap-8720	30	19	.	.	PUNCT
ap-8720	31	1	in	in	ADP
ap-8720	31	2	contrast	contrast	NOUN
ap-8720	31	3	,	,	PUNCT
ap-8720	31	4	the	the	DET
ap-8720	31	5	neural	neural	ADJ
ap-8720	31	6	textures	texture	NOUN
ap-8720	31	7	approach	approach	NOUN
ap-8720	31	8	uses	use	VERB
ap-8720	31	9	gans	gan	NOUN
ap-8720	31	10	to	to	PART
ap-8720	31	11	learn	learn	VERB
ap-8720	31	12	the	the	DET
ap-8720	31	13	texture	texture	NOUN
ap-8720	31	14	of	of	ADP
ap-8720	31	15	the	the	DET
ap-8720	31	16	target	target	NOUN
ap-8720	31	17	person	person	NOUN
ap-8720	31	18	[	[	X
ap-8720	31	19	3	3	NUM
ap-8720	31	20	]	]	PUNCT
ap-8720	31	21	.	.	PUNCT
ap-8720	32	1	new	new	ADJ
ap-8720	32	2	technological	technological	ADJ
ap-8720	32	3	advances	advance	NOUN
ap-8720	32	4	made	make	VERB
ap-8720	32	5	it	it	PRON
ap-8720	32	6	easier	easy	ADJ
ap-8720	32	7	to	to	PART
ap-8720	32	8	create	create	VERB
ap-8720	32	9	deepfakes	deepfake	NOUN
ap-8720	32	10	,	,	PUNCT
ap-8720	32	11	which	which	PRON
ap-8720	32	12	are	be	AUX
ap-8720	32	13	hyper	hyper	ADJ
ap-8720	32	14	-	-	ADJ
ap-8720	32	15	realistic	realistic	ADJ
ap-8720	32	16	videos	video	NOUN
ap-8720	32	17	or	or	CCONJ
ap-8720	32	18	images	image	NOUN
ap-8720	32	19	using	use	VERB
ap-8720	32	20	face	face	NOUN
ap-8720	32	21	swaps	swap	NOUN
ap-8720	32	22	that	that	PRON
ap-8720	32	23	leave	leave	VERB
ap-8720	32	24	little	little	ADJ
ap-8720	32	25	trace	trace	NOUN
ap-8720	32	26	of	of	ADP
ap-8720	32	27	manipulation	manipulation	NOUN
ap-8720	32	28	(	(	PUNCT
ap-8720	32	29	see	see	VERB
ap-8720	32	30	figure	figure	NOUN
ap-8720	32	31	2	2	NUM
ap-8720	32	32	)	)	PUNCT
ap-8720	32	33	.	.	PUNCT
ap-8720	33	1	these	these	DET
ap-8720	33	2	deepfakes	deepfake	NOUN
ap-8720	33	3	are	be	AUX
ap-8720	33	4	the	the	DET
ap-8720	33	5	result	result	NOUN
ap-8720	33	6	of	of	ADP
ap-8720	33	7	artificial	artificial	ADJ
ap-8720	33	8	intelligence	intelligence	NOUN
ap-8720	33	9	(	(	PUNCT
ap-8720	33	10	ai	ai	NOUN
ap-8720	33	11	)	)	PUNCT
ap-8720	33	12	applications	application	NOUN
ap-8720	33	13	that	that	PRON
ap-8720	33	14	mix	mix	VERB
ap-8720	33	15	and	and	CCONJ
ap-8720	33	16	match	match	VERB
ap-8720	33	17	images	image	NOUN
ap-8720	33	18	and	and	CCONJ
ap-8720	33	19	video	video	NOUN
ap-8720	33	20	clips	clip	NOUN
ap-8720	33	21	to	to	PART
ap-8720	33	22	create	create	VERB
ap-8720	33	23	fake	fake	ADJ
ap-8720	33	24	videos	video	NOUN
ap-8720	33	25	that	that	PRON
ap-8720	33	26	appear	appear	VERB
ap-8720	33	27	real	real	ADJ
ap-8720	33	28	and	and	CCONJ
ap-8720	33	29	natural	natural	ADJ
ap-8720	33	30	.	.	PUNCT
ap-8720	34	1	this	this	DET
ap-8720	34	2	technology	technology	NOUN
ap-8720	34	3	can	can	AUX
ap-8720	34	4	generate	generate	VERB
ap-8720	34	5	,	,	PUNCT
ap-8720	34	6	for	for	ADP
ap-8720	34	7	example	example	NOUN
ap-8720	34	8	,	,	PUNCT
ap-8720	34	9	a	a	DET
ap-8720	34	10	humorous	humorous	ADJ
ap-8720	34	11	or	or	CCONJ
ap-8720	34	12	political	political	ADJ
ap-8720	34	13	video	video	NOUN
ap-8720	34	14	of	of	ADP
ap-8720	34	15	a	a	DET
ap-8720	34	16	person	person	NOUN
ap-8720	34	17	without	without	ADP
ap-8720	34	18	the	the	DET
ap-8720	34	19	consent	consent	NOUN
ap-8720	34	20	or	or	CCONJ
ap-8720	34	21	permission	permission	NOUN
ap-8720	34	22	of	of	ADP
ap-8720	34	23	the	the	DET
ap-8720	34	24	person	person	NOUN
ap-8720	34	25	whose	whose	DET
ap-8720	34	26	image	image	NOUN
ap-8720	34	27	and	and	CCONJ
ap-8720	34	28	voice	voice	NOUN
ap-8720	34	29	are	be	AUX
ap-8720	34	30	involved	involve	VERB
ap-8720	34	31	.	.	PUNCT
ap-8720	35	1	deepfakes	deepfake	NOUN
ap-8720	35	2	can	can	AUX
ap-8720	35	3	now	now	ADV
ap-8720	35	4	be	be	AUX
ap-8720	35	5	created	create	VERB
ap-8720	35	6	305	305	NUM
ap-8720	35	7	https://doi.org/10.14311/ap.2023.63.0305	https://doi.org/10.14311/ap.2023.63.0305	PROPN
ap-8720	35	8	https://creativecommons.org/licenses/by/4.0/	https://creativecommons.org/licenses/by/4.0/	PROPN
ap-8720	35	9	https://www.cvut.cz/en	https://www.cvut.cz/en	PROPN
ap-8720	35	10	ch	ch	PROPN
ap-8720	35	11	.	.	PUNCT
ap-8720	36	1	e.	e.	PROPN
ap-8720	36	2	bencheriet	bencheriet	PROPN
ap-8720	36	3	,	,	PUNCT
ap-8720	36	4	h.	h.	PROPN
ap-8720	36	5	abdelmoumène	abdelmoumène	PROPN
ap-8720	36	6	,	,	PUNCT
ap-8720	36	7	a.	a.	PROPN
ap-8720	36	8	sebbagh	sebbagh	PROPN
ap-8720	36	9	et	et	PROPN
ap-8720	36	10	al	al	PROPN
ap-8720	36	11	.	.	PROPN
ap-8720	36	12	acta	acta	PROPN
ap-8720	36	13	polytechnica	polytechnica	PROPN
ap-8720	36	14	figure	figure	NOUN
ap-8720	36	15	1	1	NUM
ap-8720	36	16	.	.	NOUN
ap-8720	36	17	example	example	NOUN
ap-8720	36	18	of	of	ADP
ap-8720	36	19	facial	facial	ADJ
ap-8720	36	20	reconstitution	reconstitution	NOUN
ap-8720	36	21	.	.	PUNCT
ap-8720	37	1	(	(	PUNCT
ap-8720	37	2	a	a	X
ap-8720	37	3	)	)	PUNCT
ap-8720	37	4	.	.	PUNCT
ap-8720	38	1	real	real	ADJ
ap-8720	38	2	face	face	NOUN
ap-8720	38	3	images	image	NOUN
ap-8720	38	4	.	.	PUNCT
ap-8720	39	1	(	(	PUNCT
ap-8720	39	2	b	b	NOUN
ap-8720	39	3	)	)	PUNCT
ap-8720	39	4	.	.	PUNCT
ap-8720	40	1	fake	fake	ADJ
ap-8720	40	2	face	face	NOUN
ap-8720	40	3	images	image	NOUN
ap-8720	40	4	.	.	PUNCT
ap-8720	41	1	figure	figure	NOUN
ap-8720	41	2	2	2	NUM
ap-8720	41	3	.	.	PUNCT
ap-8720	41	4	deepfake	deepfake	VERB
ap-8720	41	5	example	example	NOUN
ap-8720	42	1	[	[	X
ap-8720	42	2	12	12	NUM
ap-8720	42	3	]	]	PUNCT
ap-8720	42	4	.	.	PUNCT
ap-8720	43	1	using	use	VERB
ap-8720	43	2	consumer	consumer	NOUN
ap-8720	43	3	-	-	PUNCT
ap-8720	43	4	grade	grade	NOUN
ap-8720	43	5	hardware	hardware	NOUN
ap-8720	43	6	as	as	SCONJ
ap-8720	43	7	demonstrated	demonstrate	VERB
ap-8720	43	8	by	by	ADP
ap-8720	43	9	two	two	NUM
ap-8720	43	10	papers	paper	NOUN
ap-8720	43	11	published	publish	VERB
ap-8720	43	12	in	in	ADP
ap-8720	43	13	2016	2016	NUM
ap-8720	43	14	and	and	CCONJ
ap-8720	43	15	2017	2017	NUM
ap-8720	44	1	[	[	SYM
ap-8720	44	2	4	4	NUM
ap-8720	44	3	,	,	PUNCT
ap-8720	44	4	5	5	NUM
ap-8720	44	5	]	]	PUNCT
ap-8720	44	6	.	.	PUNCT
ap-8720	45	1	the	the	DET
ap-8720	45	2	face2face	face2face	NOUN
ap-8720	45	3	project	project	NOUN
ap-8720	45	4	from	from	ADP
ap-8720	45	5	the	the	DET
ap-8720	45	6	technical	technical	ADJ
ap-8720	45	7	university	university	PROPN
ap-8720	45	8	of	of	ADP
ap-8720	45	9	munich	munich	PROPN
ap-8720	45	10	and	and	CCONJ
ap-8720	45	11	the	the	DET
ap-8720	45	12	synthesizing	synthesize	VERB
ap-8720	45	13	obama	obama	PROPN
ap-8720	45	14	project	project	NOUN
ap-8720	45	15	from	from	ADP
ap-8720	45	16	the	the	DET
ap-8720	45	17	university	university	PROPN
ap-8720	45	18	of	of	ADP
ap-8720	45	19	washington	washington	PROPN
ap-8720	45	20	established	establish	VERB
ap-8720	45	21	the	the	DET
ap-8720	45	22	feasibility	feasibility	NOUN
ap-8720	45	23	of	of	ADP
ap-8720	45	24	generating	generate	VERB
ap-8720	45	25	deepfakes	deepfake	NOUN
ap-8720	45	26	.	.	PUNCT
ap-8720	46	1	they	they	PRON
ap-8720	46	2	have	have	AUX
ap-8720	46	3	improved	improve	VERB
ap-8720	46	4	computing	computing	NOUN
ap-8720	46	5	and	and	CCONJ
ap-8720	46	6	rendering	rendering	NOUN
ap-8720	46	7	times	time	NOUN
ap-8720	46	8	while	while	SCONJ
ap-8720	46	9	updating	update	VERB
ap-8720	46	10	graphical	graphical	ADJ
ap-8720	46	11	fidelity	fidelity	NOUN
ap-8720	46	12	for	for	ADP
ap-8720	46	13	a	a	DET
ap-8720	46	14	photorealistic	photorealistic	ADJ
ap-8720	46	15	look	look	NOUN
ap-8720	46	16	.	.	PUNCT
ap-8720	47	1	the	the	DET
ap-8720	47	2	game	game	NOUN
ap-8720	47	3	-	-	PUNCT
ap-8720	47	4	changing	change	VERB
ap-8720	47	5	factors	factor	NOUN
ap-8720	47	6	for	for	ADP
ap-8720	47	7	forgeries	forgery	NOUN
ap-8720	47	8	are	be	AUX
ap-8720	47	9	the	the	DET
ap-8720	47	10	scope	scope	NOUN
ap-8720	47	11	,	,	PUNCT
ap-8720	47	12	scale	scale	NOUN
ap-8720	47	13	,	,	PUNCT
ap-8720	47	14	and	and	CCONJ
ap-8720	47	15	complexity	complexity	NOUN
ap-8720	47	16	of	of	ADP
ap-8720	47	17	the	the	DET
ap-8720	47	18	technology	technology	NOUN
ap-8720	47	19	involved	involve	VERB
ap-8720	47	20	,	,	PUNCT
ap-8720	47	21	as	as	SCONJ
ap-8720	47	22	today	today	NOUN
ap-8720	47	23	anyone	anyone	PRON
ap-8720	47	24	with	with	ADP
ap-8720	47	25	a	a	DET
ap-8720	47	26	computer	computer	NOUN
ap-8720	47	27	can	can	AUX
ap-8720	47	28	create	create	VERB
ap-8720	47	29	fake	fake	ADJ
ap-8720	47	30	videos	video	NOUN
ap-8720	47	31	that	that	PRON
ap-8720	47	32	are	be	AUX
ap-8720	47	33	almost	almost	ADV
ap-8720	47	34	indistinguishable	indistinguishable	ADJ
ap-8720	47	35	from	from	ADP
ap-8720	47	36	real	real	ADJ
ap-8720	47	37	videos	video	NOUN
ap-8720	47	38	[	[	X
ap-8720	47	39	6	6	NUM
ap-8720	47	40	]	]	PUNCT
ap-8720	47	41	.	.	PUNCT
ap-8720	48	1	the	the	DET
ap-8720	48	2	gan	gan	PROPN
ap-8720	48	3	and	and	CCONJ
ap-8720	48	4	autoencoder	autoencoder	NOUN
ap-8720	48	5	networks	network	NOUN
ap-8720	48	6	have	have	AUX
ap-8720	48	7	become	become	VERB
ap-8720	48	8	very	very	ADV
ap-8720	48	9	famous	famous	ADJ
ap-8720	48	10	for	for	ADP
ap-8720	48	11	their	their	PRON
ap-8720	48	12	ability	ability	NOUN
ap-8720	48	13	to	to	PART
ap-8720	48	14	produce	produce	VERB
ap-8720	48	15	forgeries	forgery	NOUN
ap-8720	48	16	of	of	ADP
ap-8720	48	17	remarkable	remarkable	ADJ
ap-8720	48	18	quality	quality	NOUN
ap-8720	48	19	and	and	CCONJ
ap-8720	48	20	undetectable	undetectable	ADJ
ap-8720	48	21	by	by	ADP
ap-8720	48	22	human	human	ADJ
ap-8720	48	23	beings	being	NOUN
ap-8720	48	24	.	.	PUNCT
ap-8720	49	1	recently	recently	ADV
ap-8720	49	2	,	,	PUNCT
ap-8720	49	3	many	many	ADJ
ap-8720	49	4	architectures	architecture	NOUN
ap-8720	49	5	have	have	AUX
ap-8720	49	6	been	be	AUX
ap-8720	49	7	developed	develop	VERB
ap-8720	49	8	to	to	PART
ap-8720	49	9	generate	generate	VERB
ap-8720	49	10	fakes	fake	NOUN
ap-8720	49	11	.	.	PUNCT
ap-8720	50	1	a	a	DET
ap-8720	50	2	severe	severe	ADJ
ap-8720	50	3	competition	competition	NOUN
ap-8720	50	4	between	between	ADP
ap-8720	50	5	these	these	DET
ap-8720	50	6	architectures	architecture	NOUN
ap-8720	50	7	was	be	AUX
ap-8720	50	8	measured	measure	VERB
ap-8720	50	9	by	by	ADP
ap-8720	50	10	the	the	DET
ap-8720	50	11	quality	quality	NOUN
ap-8720	50	12	of	of	ADP
ap-8720	50	13	the	the	DET
ap-8720	50	14	false	false	ADJ
ap-8720	50	15	images	image	NOUN
ap-8720	50	16	generated	generate	VERB
ap-8720	50	17	by	by	ADP
ap-8720	50	18	each	each	DET
ap-8720	50	19	one	one	NUM
ap-8720	50	20	.	.	PUNCT
ap-8720	51	1	several	several	ADJ
ap-8720	51	2	fake	fake	ADJ
ap-8720	51	3	datasets	dataset	NOUN
ap-8720	51	4	consisting	consist	VERB
ap-8720	51	5	mainly	mainly	ADV
ap-8720	51	6	of	of	ADP
ap-8720	51	7	fake	fake	ADJ
ap-8720	51	8	faces	face	NOUN
ap-8720	51	9	have	have	AUX
ap-8720	51	10	been	be	AUX
ap-8720	51	11	developed	develop	VERB
ap-8720	51	12	,	,	PUNCT
ap-8720	51	13	such	such	ADJ
ap-8720	51	14	as	as	ADP
ap-8720	51	15	the	the	DET
ap-8720	51	16	gan	gan	PROPN
ap-8720	51	17	collection	collection	NOUN
ap-8720	52	1	[	[	X
ap-8720	52	2	7	7	NUM
ap-8720	52	3	]	]	PUNCT
ap-8720	52	4	,	,	PUNCT
ap-8720	52	5	faceforensics++	faceforensics++	PROPN
ap-8720	52	6	[	[	X
ap-8720	52	7	8	8	NUM
ap-8720	52	8	]	]	PUNCT
ap-8720	52	9	,	,	PUNCT
ap-8720	52	10	celeb	celeb	PROPN
ap-8720	52	11	-	-	PUNCT
ap-8720	52	12	df	df	PROPN
ap-8720	53	1	[	[	X
ap-8720	53	2	9	9	NUM
ap-8720	53	3	]	]	PUNCT
ap-8720	53	4	,	,	PUNCT
ap-8720	53	5	defacto	defacto	X
ap-8720	54	1	[	[	X
ap-8720	54	2	10	10	NUM
ap-8720	54	3	]	]	PUNCT
ap-8720	54	4	,	,	PUNCT
ap-8720	54	5	faceswap	faceswap	ADJ
ap-8720	55	1	[	[	X
ap-8720	55	2	11	11	NUM
ap-8720	55	3	]	]	PUNCT
ap-8720	55	4	.	.	PUNCT
ap-8720	56	1	in	in	ADP
ap-8720	56	2	this	this	DET
ap-8720	56	3	paper	paper	NOUN
ap-8720	56	4	,	,	PUNCT
ap-8720	56	5	we	we	PRON
ap-8720	56	6	introduce	introduce	VERB
ap-8720	56	7	a	a	DET
ap-8720	56	8	new	new	ADJ
ap-8720	56	9	architecture	architecture	NOUN
ap-8720	56	10	called	call	VERB
ap-8720	56	11	ndcgan	ndcgan	NOUN
ap-8720	56	12	,	,	PUNCT
ap-8720	56	13	which	which	PRON
ap-8720	56	14	stands	stand	VERB
ap-8720	56	15	for	for	ADP
ap-8720	56	16	“	"	PUNCT
ap-8720	56	17	neural	neural	ADJ
ap-8720	56	18	discriminatorgenerator	discriminatorgenerator	NOUN
ap-8720	56	19	confrontation	confrontation	NOUN
ap-8720	56	20	generative	generative	VERB
ap-8720	56	21	adversarial	adversarial	ADJ
ap-8720	56	22	network	network	NOUN
ap-8720	56	23	”	"	PUNCT
ap-8720	56	24	.	.	PUNCT
ap-8720	57	1	this	this	DET
ap-8720	57	2	architecture	architecture	NOUN
ap-8720	57	3	consists	consist	VERB
ap-8720	57	4	of	of	ADP
ap-8720	57	5	three	three	NUM
ap-8720	57	6	discriminator	discriminator	NOUN
ap-8720	57	7	modules	module	NOUN
ap-8720	57	8	that	that	PRON
ap-8720	57	9	work	work	VERB
ap-8720	57	10	together	together	ADV
ap-8720	57	11	to	to	PART
ap-8720	57	12	distinguish	distinguish	VERB
ap-8720	57	13	between	between	ADP
ap-8720	57	14	real	real	ADJ
ap-8720	57	15	and	and	CCONJ
ap-8720	57	16	fake	fake	ADJ
ap-8720	57	17	faces	face	NOUN
ap-8720	57	18	.	.	PUNCT
ap-8720	58	1	the	the	DET
ap-8720	58	2	first	first	ADJ
ap-8720	58	3	module	module	NOUN
ap-8720	58	4	,	,	PUNCT
ap-8720	58	5	discriminator	discriminator	NOUN
ap-8720	58	6	1	1	NUM
ap-8720	58	7	,	,	PUNCT
ap-8720	58	8	is	be	AUX
ap-8720	58	9	trained	train	VERB
ap-8720	58	10	against	against	ADP
ap-8720	58	11	its	its	PRON
ap-8720	58	12	opponent	opponent	NOUN
ap-8720	58	13	,	,	PUNCT
ap-8720	58	14	the	the	DET
ap-8720	58	15	generator	generator	NOUN
ap-8720	58	16	,	,	PUNCT
ap-8720	58	17	to	to	PART
ap-8720	58	18	outperform	outperform	VERB
ap-8720	58	19	it	it	PRON
ap-8720	58	20	and	and	CCONJ
ap-8720	58	21	accurately	accurately	ADV
ap-8720	58	22	identify	identify	VERB
ap-8720	58	23	the	the	DET
ap-8720	58	24	fake	fake	ADJ
ap-8720	58	25	faces	face	NOUN
ap-8720	58	26	generated	generate	VERB
ap-8720	58	27	by	by	ADP
ap-8720	58	28	the	the	DET
ap-8720	58	29	generator	generator	NOUN
ap-8720	58	30	.	.	PUNCT
ap-8720	59	1	the	the	DET
ap-8720	59	2	second	second	ADJ
ap-8720	59	3	module	module	NOUN
ap-8720	59	4	,	,	PUNCT
ap-8720	59	5	discriminator	discriminator	NOUN
ap-8720	59	6	2	2	NUM
ap-8720	59	7	,	,	PUNCT
ap-8720	59	8	is	be	AUX
ap-8720	59	9	the	the	DET
ap-8720	59	10	ndcgan	ndcgan	ADJ
ap-8720	59	11	discriminator	discriminator	NOUN
ap-8720	59	12	(	(	PUNCT
ap-8720	59	13	discriminator	discriminator	NOUN
ap-8720	59	14	1	1	NUM
ap-8720	59	15	)	)	PUNCT
ap-8720	59	16	trained	train	VERB
ap-8720	59	17	using	use	VERB
ap-8720	59	18	transfer	transfer	NOUN
ap-8720	59	19	learning	learning	NOUN
ap-8720	59	20	techniques	technique	NOUN
ap-8720	59	21	.	.	PUNCT
ap-8720	60	1	this	this	DET
ap-8720	60	2	module	module	NOUN
ap-8720	60	3	leverages	leverage	VERB
ap-8720	60	4	the	the	DET
ap-8720	60	5	pre	pre	ADJ
ap-8720	60	6	-	-	ADJ
ap-8720	60	7	trained	train	VERB
ap-8720	60	8	knowledge	knowledge	NOUN
ap-8720	60	9	of	of	ADP
ap-8720	60	10	discriminator	discriminator	NOUN
ap-8720	60	11	1	1	NUM
ap-8720	60	12	to	to	PART
ap-8720	60	13	enhance	enhance	VERB
ap-8720	60	14	its	its	PRON
ap-8720	60	15	ability	ability	NOUN
ap-8720	60	16	to	to	PART
ap-8720	60	17	distinguish	distinguish	VERB
ap-8720	60	18	between	between	ADP
ap-8720	60	19	real	real	ADJ
ap-8720	60	20	and	and	CCONJ
ap-8720	60	21	fake	fake	ADJ
ap-8720	60	22	faces	face	NOUN
ap-8720	60	23	.	.	PUNCT
ap-8720	61	1	the	the	DET
ap-8720	61	2	third	third	ADJ
ap-8720	61	3	module	module	NOUN
ap-8720	61	4	,	,	PUNCT
ap-8720	61	5	discriminator	discriminator	NOUN
ap-8720	61	6	3	3	NUM
ap-8720	61	7	,	,	PUNCT
ap-8720	61	8	is	be	AUX
ap-8720	61	9	a	a	DET
ap-8720	61	10	separate	separate	ADJ
ap-8720	61	11	discriminator	discriminator	NOUN
ap-8720	61	12	trained	train	VERB
ap-8720	61	13	from	from	ADP
ap-8720	61	14	scratch	scratch	NOUN
ap-8720	61	15	on	on	ADP
ap-8720	61	16	both	both	CCONJ
ap-8720	61	17	real	real	ADJ
ap-8720	61	18	and	and	CCONJ
ap-8720	61	19	fake	fake	ADJ
ap-8720	61	20	faces	face	NOUN
ap-8720	61	21	.	.	PUNCT
ap-8720	62	1	it	it	PRON
ap-8720	62	2	is	be	AUX
ap-8720	62	3	designed	design	VERB
ap-8720	62	4	to	to	PART
ap-8720	62	5	work	work	VERB
ap-8720	62	6	independently	independently	ADV
ap-8720	62	7	of	of	ADP
ap-8720	62	8	its	its	PRON
ap-8720	62	9	associated	associated	ADJ
ap-8720	62	10	generator	generator	NOUN
ap-8720	62	11	,	,	PUNCT
ap-8720	62	12	focusing	focus	VERB
ap-8720	62	13	solely	solely	ADV
ap-8720	62	14	on	on	ADP
ap-8720	62	15	accurately	accurately	ADV
ap-8720	62	16	classifying	classify	VERB
ap-8720	62	17	faces	face	NOUN
ap-8720	62	18	.	.	PUNCT
ap-8720	63	1	in	in	ADP
ap-8720	63	2	the	the	DET
ap-8720	63	3	final	final	ADJ
ap-8720	63	4	decision	decision	NOUN
ap-8720	63	5	-	-	PUNCT
ap-8720	63	6	making	make	VERB
ap-8720	63	7	process	process	NOUN
ap-8720	63	8	,	,	PUNCT
ap-8720	63	9	the	the	DET
ap-8720	63	10	three	three	NUM
ap-8720	63	11	discriminators	discriminator	NOUN
ap-8720	63	12	vote	vote	VERB
ap-8720	63	13	together	together	ADV
ap-8720	63	14	on	on	ADP
ap-8720	63	15	whether	whether	SCONJ
ap-8720	63	16	a	a	DET
ap-8720	63	17	given	give	VERB
ap-8720	63	18	face	face	NOUN
ap-8720	63	19	belongs	belong	VERB
ap-8720	63	20	to	to	ADP
ap-8720	63	21	the	the	DET
ap-8720	63	22	real	real	ADJ
ap-8720	63	23	face	face	NOUN
ap-8720	63	24	class	class	NOUN
ap-8720	63	25	or	or	CCONJ
ap-8720	63	26	the	the	DET
ap-8720	63	27	fake	fake	ADJ
ap-8720	63	28	face	face	NOUN
ap-8720	63	29	class	class	NOUN
ap-8720	63	30	.	.	PUNCT
ap-8720	64	1	this	this	DET
ap-8720	64	2	voting	voting	NOUN
ap-8720	64	3	mechanism	mechanism	NOUN
ap-8720	64	4	ensures	ensure	VERB
ap-8720	64	5	a	a	DET
ap-8720	64	6	more	more	ADV
ap-8720	64	7	robust	robust	ADJ
ap-8720	64	8	and	and	CCONJ
ap-8720	64	9	reliable	reliable	ADJ
ap-8720	64	10	determination	determination	NOUN
ap-8720	64	11	of	of	ADP
ap-8720	64	12	a	a	DET
ap-8720	64	13	face	face	NOUN
ap-8720	64	14	’s	’s	PART
ap-8720	64	15	authenticity	authenticity	NOUN
ap-8720	64	16	.	.	PUNCT
ap-8720	65	1	through	through	ADP
ap-8720	65	2	the	the	DET
ap-8720	65	3	utilisation	utilisation	NOUN
ap-8720	65	4	of	of	ADP
ap-8720	65	5	ndcgan	ndcgan	NOUN
ap-8720	65	6	and	and	CCONJ
ap-8720	65	7	the	the	DET
ap-8720	65	8	collaboration	collaboration	NOUN
ap-8720	65	9	of	of	ADP
ap-8720	65	10	these	these	DET
ap-8720	65	11	three	three	NUM
ap-8720	65	12	discriminator	discriminator	NOUN
ap-8720	65	13	modules	module	NOUN
ap-8720	65	14	,	,	PUNCT
ap-8720	65	15	our	our	PRON
ap-8720	65	16	proposed	propose	VERB
ap-8720	65	17	architecture	architecture	NOUN
ap-8720	65	18	offers	offer	VERB
ap-8720	65	19	improved	improved	ADJ
ap-8720	65	20	performance	performance	NOUN
ap-8720	65	21	and	and	CCONJ
ap-8720	65	22	accuracy	accuracy	NOUN
ap-8720	65	23	in	in	ADP
ap-8720	65	24	distinguishing	distinguish	VERB
ap-8720	65	25	between	between	ADP
ap-8720	65	26	real	real	ADJ
ap-8720	65	27	and	and	CCONJ
ap-8720	65	28	fake	fake	ADJ
ap-8720	65	29	faces	face	NOUN
ap-8720	65	30	.	.	PUNCT
ap-8720	66	1	the	the	DET
ap-8720	66	2	rest	rest	NOUN
ap-8720	66	3	of	of	ADP
ap-8720	66	4	the	the	DET
ap-8720	66	5	paper	paper	NOUN
ap-8720	66	6	is	be	AUX
ap-8720	66	7	organised	organise	VERB
ap-8720	66	8	as	as	SCONJ
ap-8720	66	9	follows	follow	VERB
ap-8720	66	10	:	:	PUNCT
ap-8720	66	11	section	section	NOUN
ap-8720	66	12	2	2	NUM
ap-8720	66	13	presents	present	VERB
ap-8720	66	14	existing	exist	VERB
ap-8720	66	15	work	work	NOUN
ap-8720	66	16	regarding	regard	VERB
ap-8720	66	17	fake	fake	ADJ
ap-8720	66	18	detection	detection	NOUN
ap-8720	66	19	based	base	VERB
ap-8720	66	20	on	on	ADP
ap-8720	66	21	deep	deep	ADJ
ap-8720	66	22	neural	neural	ADJ
ap-8720	66	23	networks	network	NOUN
ap-8720	66	24	.	.	PUNCT
ap-8720	67	1	section	section	NOUN
ap-8720	67	2	3	3	NUM
ap-8720	67	3	provides	provide	VERB
ap-8720	67	4	an	an	DET
ap-8720	67	5	overview	overview	NOUN
ap-8720	67	6	of	of	ADP
ap-8720	67	7	the	the	DET
ap-8720	67	8	proposed	propose	VERB
ap-8720	67	9	architecture	architecture	NOUN
ap-8720	67	10	,	,	PUNCT
ap-8720	67	11	along	along	ADP
ap-8720	67	12	with	with	ADP
ap-8720	67	13	a	a	DET
ap-8720	67	14	brief	brief	ADJ
ap-8720	67	15	explanation	explanation	NOUN
ap-8720	67	16	of	of	ADP
ap-8720	67	17	the	the	DET
ap-8720	67	18	baseline	baseline	NOUN
ap-8720	67	19	modules	module	NOUN
ap-8720	67	20	.	.	PUNCT
ap-8720	68	1	in	in	ADP
ap-8720	68	2	section	section	NOUN
ap-8720	68	3	4	4	NUM
ap-8720	68	4	,	,	PUNCT
ap-8720	68	5	we	we	PRON
ap-8720	68	6	present	present	VERB
ap-8720	68	7	the	the	DET
ap-8720	68	8	proposed	propose	VERB
ap-8720	68	9	model	model	NOUN
ap-8720	68	10	,	,	PUNCT
ap-8720	68	11	along	along	ADP
ap-8720	68	12	with	with	ADP
ap-8720	68	13	the	the	DET
ap-8720	68	14	training	training	NOUN
ap-8720	68	15	parameters	parameter	NOUN
ap-8720	68	16	,	,	PUNCT
ap-8720	68	17	implementation	implementation	NOUN
ap-8720	68	18	details	detail	NOUN
ap-8720	68	19	,	,	PUNCT
ap-8720	68	20	classification	classification	NOUN
ap-8720	68	21	results	result	NOUN
ap-8720	68	22	,	,	PUNCT
ap-8720	68	23	and	and	CCONJ
ap-8720	68	24	discussions	discussion	NOUN
ap-8720	68	25	.	.	PUNCT
ap-8720	69	1	finally	finally	ADV
ap-8720	69	2	,	,	PUNCT
ap-8720	69	3	section	section	NOUN
ap-8720	69	4	5	5	NUM
ap-8720	69	5	presents	present	VERB
ap-8720	69	6	the	the	DET
ap-8720	69	7	conclusions	conclusion	NOUN
ap-8720	69	8	and	and	CCONJ
ap-8720	69	9	discusses	discuss	VERB
ap-8720	69	10	future	future	ADJ
ap-8720	69	11	work	work	NOUN
ap-8720	69	12	.	.	PUNCT
ap-8720	70	1	2	2	X
ap-8720	70	2	.	.	X
ap-8720	70	3	related	relate	VERB
ap-8720	70	4	work	work	NOUN
ap-8720	70	5	although	although	SCONJ
ap-8720	70	6	deepfakes	deepfake	NOUN
ap-8720	70	7	were	be	AUX
ap-8720	70	8	initially	initially	ADV
ap-8720	70	9	created	create	VERB
ap-8720	70	10	for	for	ADP
ap-8720	70	11	entertainment	entertainment	NOUN
ap-8720	70	12	purposes	purpose	NOUN
ap-8720	70	13	,	,	PUNCT
ap-8720	70	14	such	such	ADJ
ap-8720	70	15	as	as	ADP
ap-8720	70	16	allowing	allow	VERB
ap-8720	70	17	people	people	NOUN
ap-8720	70	18	to	to	PART
ap-8720	70	19	experience	experience	VERB
ap-8720	70	20	things	thing	NOUN
ap-8720	70	21	that	that	PRON
ap-8720	70	22	no	no	ADV
ap-8720	70	23	longer	long	ADV
ap-8720	70	24	exist	exist	VERB
ap-8720	70	25	or	or	CCONJ
ap-8720	70	26	to	to	PART
ap-8720	70	27	attract	attract	VERB
ap-8720	70	28	extraordinary	extraordinary	ADJ
ap-8720	70	29	attention	attention	NOUN
ap-8720	70	30	from	from	ADP
ap-8720	70	31	online	online	ADJ
ap-8720	70	32	audiences	audience	NOUN
ap-8720	70	33	,	,	PUNCT
ap-8720	70	34	making	make	VERB
ap-8720	70	35	a	a	DET
ap-8720	70	36	website	website	NOUN
ap-8720	70	37	popular	popular	ADJ
ap-8720	70	38	in	in	ADP
ap-8720	70	39	search	search	NOUN
ap-8720	70	40	engines	engine	NOUN
ap-8720	70	41	.	.	PUNCT
ap-8720	71	1	later	later	ADV
ap-8720	71	2	,	,	PUNCT
ap-8720	71	3	their	their	PRON
ap-8720	71	4	applications	application	NOUN
ap-8720	71	5	were	be	AUX
ap-8720	71	6	exploited	exploit	VERB
ap-8720	71	7	for	for	ADP
ap-8720	71	8	counterfeiting	counterfeit	VERB
ap-8720	71	9	and	and	CCONJ
ap-8720	71	10	spreading	spread	VERB
ap-8720	71	11	misinformation	misinformation	NOUN
ap-8720	71	12	,	,	PUNCT
ap-8720	71	13	including	include	VERB
ap-8720	71	14	the	the	DET
ap-8720	71	15	creation	creation	NOUN
ap-8720	71	16	of	of	ADP
ap-8720	71	17	false	false	ADJ
ap-8720	71	18	news	news	NOUN
ap-8720	71	19	and	and	CCONJ
ap-8720	71	20	deceptive	deceptive	ADJ
ap-8720	71	21	advertising	advertising	NOUN
ap-8720	71	22	,	,	PUNCT
ap-8720	71	23	as	as	ADV
ap-8720	71	24	well	well	ADV
ap-8720	71	25	as	as	ADP
ap-8720	71	26	the	the	DET
ap-8720	71	27	malicious	malicious	ADJ
ap-8720	71	28	use	use	NOUN
ap-8720	71	29	of	of	ADP
ap-8720	71	30	video	video	NOUN
ap-8720	71	31	deepfakes	deepfake	NOUN
ap-8720	71	32	for	for	ADP
ap-8720	71	33	blackmail	blackmail	NOUN
ap-8720	71	34	.	.	PUNCT
ap-8720	72	1	faced	face	VERB
ap-8720	72	2	with	with	ADP
ap-8720	72	3	the	the	DET
ap-8720	72	4	emergence	emergence	NOUN
ap-8720	72	5	of	of	ADP
ap-8720	72	6	fakes	fake	NOUN
ap-8720	72	7	in	in	ADP
ap-8720	72	8	our	our	PRON
ap-8720	72	9	environment	environment	NOUN
ap-8720	72	10	that	that	PRON
ap-8720	72	11	are	be	AUX
ap-8720	72	12	undetectable	undetectable	ADJ
ap-8720	72	13	by	by	ADP
ap-8720	72	14	humans	human	NOUN
ap-8720	72	15	,	,	PUNCT
ap-8720	72	16	the	the	DET
ap-8720	72	17	interest	interest	NOUN
ap-8720	72	18	of	of	ADP
ap-8720	72	19	researchers	researcher	NOUN
ap-8720	72	20	has	have	AUX
ap-8720	72	21	subsequently	subsequently	ADV
ap-8720	72	22	been	be	AUX
ap-8720	72	23	directed	direct	VERB
ap-8720	72	24	towards	towards	ADP
ap-8720	72	25	finding	find	VERB
ap-8720	72	26	robust	robust	ADJ
ap-8720	72	27	algorithms	algorithm	NOUN
ap-8720	72	28	for	for	ADP
ap-8720	72	29	the	the	DET
ap-8720	72	30	detection	detection	NOUN
ap-8720	72	31	of	of	ADP
ap-8720	72	32	fake	fake	ADJ
ap-8720	72	33	images	image	NOUN
ap-8720	72	34	and	and	CCONJ
ap-8720	72	35	videos	video	NOUN
ap-8720	72	36	[	[	X
ap-8720	72	37	13	13	NUM
ap-8720	72	38	,	,	PUNCT
ap-8720	72	39	14	14	NUM
ap-8720	72	40	]	]	PUNCT
ap-8720	72	41	based	base	VERB
ap-8720	72	42	on	on	ADP
ap-8720	72	43	various	various	ADJ
ap-8720	72	44	deep	deep	ADJ
ap-8720	72	45	-	-	PUNCT
ap-8720	72	46	learning	learn	VERB
ap-8720	72	47	approaches	approach	NOUN
ap-8720	72	48	,	,	PUNCT
ap-8720	72	49	such	such	ADJ
ap-8720	72	50	as	as	ADP
ap-8720	72	51	long	long	ADJ
ap-8720	72	52	short	short	ADJ
ap-8720	72	53	-	-	PUNCT
ap-8720	72	54	term	term	NOUN
ap-8720	72	55	memory	memory	NOUN
ap-8720	72	56	(	(	PUNCT
ap-8720	72	57	lstm	lstm	NOUN
ap-8720	72	58	)	)	PUNCT
ap-8720	72	59	,	,	PUNCT
ap-8720	72	60	recurrent	recurrent	ADJ
ap-8720	72	61	neural	neural	ADJ
ap-8720	72	62	network	network	NOUN
ap-8720	72	63	(	(	PUNCT
ap-8720	72	64	rnn	rnn	PROPN
ap-8720	72	65	)	)	PUNCT
ap-8720	72	66	,	,	PUNCT
ap-8720	72	67	convolutional	convolutional	ADJ
ap-8720	72	68	neural	neural	ADJ
ap-8720	72	69	network	network	NOUN
ap-8720	72	70	(	(	PUNCT
ap-8720	72	71	cnn	cnn	PROPN
ap-8720	72	72	)	)	PUNCT
ap-8720	72	73	,	,	PUNCT
ap-8720	72	74	and	and	CCONJ
ap-8720	72	75	even	even	ADV
ap-8720	72	76	the	the	DET
ap-8720	72	77	hybrid	hybrid	ADJ
ap-8720	72	78	approaches	approach	NOUN
ap-8720	72	79	have	have	AUX
ap-8720	72	80	been	be	AUX
ap-8720	72	81	proposed	propose	VERB
ap-8720	72	82	.	.	PUNCT
ap-8720	73	1	guera	guera	VERB
ap-8720	73	2	and	and	CCONJ
ap-8720	73	3	delp	delp	VERB
ap-8720	73	4	[	[	X
ap-8720	73	5	15	15	NUM
ap-8720	73	6	]	]	PUNCT
ap-8720	73	7	proposed	propose	VERB
ap-8720	73	8	a	a	DET
ap-8720	73	9	video	video	NOUN
ap-8720	73	10	deepfake	deepfake	NOUN
ap-8720	73	11	detection	detection	NOUN
ap-8720	73	12	system	system	NOUN
ap-8720	73	13	that	that	PRON
ap-8720	73	14	uses	use	VERB
ap-8720	73	15	a	a	DET
ap-8720	73	16	recurrent	recurrent	ADJ
ap-8720	73	17	convolutional	convolutional	ADJ
ap-8720	73	18	neural	neural	ADJ
ap-8720	73	19	network	network	NOUN
ap-8720	73	20	to	to	PART
ap-8720	73	21	extract	extract	VERB
ap-8720	73	22	features	feature	NOUN
ap-8720	73	23	from	from	ADP
ap-8720	73	24	each	each	DET
ap-8720	73	25	frame	frame	NOUN
ap-8720	73	26	,	,	PUNCT
ap-8720	73	27	which	which	PRON
ap-8720	73	28	are	be	AUX
ap-8720	73	29	then	then	ADV
ap-8720	73	30	concatenated	concatenate	VERB
ap-8720	73	31	and	and	CCONJ
ap-8720	73	32	sent	send	VERB
ap-8720	73	33	to	to	ADP
ap-8720	73	34	the	the	DET
ap-8720	73	35	lstm	lstm	NOUN
ap-8720	73	36	for	for	ADP
ap-8720	73	37	analysis	analysis	NOUN
ap-8720	73	38	.	.	PUNCT
ap-8720	74	1	they	they	PRON
ap-8720	74	2	produce	produce	VERB
ap-8720	74	3	an	an	DET
ap-8720	74	4	estimate	estimate	NOUN
ap-8720	74	5	of	of	ADP
ap-8720	74	6	the	the	DET
ap-8720	74	7	probability	probability	NOUN
ap-8720	74	8	that	that	SCONJ
ap-8720	74	9	the	the	DET
ap-8720	74	10	sequence	sequence	NOUN
ap-8720	74	11	is	be	AUX
ap-8720	74	12	a	a	DET
ap-8720	74	13	deepfake	deepfake	NOUN
ap-8720	74	14	or	or	CCONJ
ap-8720	74	15	an	an	DET
ap-8720	74	16	unmanipulated	unmanipulated	ADJ
ap-8720	74	17	video	video	NOUN
ap-8720	74	18	.	.	PUNCT
ap-8720	75	1	the	the	DET
ap-8720	75	2	authors	author	NOUN
ap-8720	75	3	collected	collect	VERB
ap-8720	75	4	600	600	NUM
ap-8720	75	5	video	video	NOUN
ap-8720	75	6	deepfakes	deepfake	NOUN
ap-8720	75	7	,	,	PUNCT
ap-8720	75	8	306	306	NUM
ap-8720	75	9	vol	vol	NOUN
ap-8720	75	10	.	.	PUNCT
ap-8720	76	1	63	63	NUM
ap-8720	77	1	no	no	NOUN
ap-8720	77	2	.	.	PUNCT
ap-8720	78	1	5/2023	5/2023	NUM
ap-8720	78	2	fake	fake	ADJ
ap-8720	78	3	face	face	NOUN
ap-8720	78	4	detection	detection	NOUN
ap-8720	78	5	based	base	VERB
ap-8720	78	6	on	on	ADP
ap-8720	78	7	a	a	DET
ap-8720	78	8	multi	multi	ADJ
ap-8720	78	9	discriminator	discriminator	NOUN
ap-8720	78	10	deep	deep	ADJ
ap-8720	78	11	cnn	cnn	PROPN
ap-8720	78	12	architecture	architecture	NOUN
ap-8720	78	13	including	include	VERB
ap-8720	78	14	300	300	NUM
ap-8720	78	15	videos	video	NOUN
ap-8720	78	16	from	from	ADP
ap-8720	78	17	various	various	ADJ
ap-8720	78	18	websites	website	NOUN
ap-8720	78	19	and	and	CCONJ
ap-8720	78	20	another	another	DET
ap-8720	78	21	300	300	NUM
ap-8720	78	22	videos	video	NOUN
ap-8720	78	23	from	from	ADP
ap-8720	78	24	the	the	DET
ap-8720	78	25	hoha	hoha	PROPN
ap-8720	78	26	database	database	NOUN
ap-8720	78	27	[	[	X
ap-8720	78	28	16	16	NUM
ap-8720	78	29	]	]	PUNCT
ap-8720	78	30	.	.	PUNCT
ap-8720	79	1	they	they	PRON
ap-8720	79	2	presented	present	VERB
ap-8720	79	3	the	the	DET
ap-8720	79	4	performance	performance	NOUN
ap-8720	79	5	of	of	ADP
ap-8720	79	6	their	their	PRON
ap-8720	79	7	system	system	NOUN
ap-8720	79	8	in	in	ADP
ap-8720	79	9	terms	term	NOUN
ap-8720	79	10	of	of	ADP
ap-8720	79	11	accuracy	accuracy	NOUN
ap-8720	79	12	using	use	VERB
ap-8720	79	13	sub	sub	NOUN
ap-8720	79	14	-	-	NOUN
ap-8720	79	15	sequences	sequence	NOUN
ap-8720	79	16	with	with	ADP
ap-8720	79	17	lengths	length	NOUN
ap-8720	79	18	of	of	ADP
ap-8720	79	19	20	20	NUM
ap-8720	79	20	,	,	PUNCT
ap-8720	79	21	40	40	NUM
ap-8720	79	22	,	,	PUNCT
ap-8720	79	23	and	and	CCONJ
ap-8720	79	24	80	80	NUM
ap-8720	79	25	frames	frame	NOUN
ap-8720	79	26	.	.	PUNCT
ap-8720	80	1	they	they	PRON
ap-8720	80	2	achieved	achieve	VERB
ap-8720	80	3	96.7	96.7	NUM
ap-8720	80	4	%	%	NOUN
ap-8720	80	5	accuracy	accuracy	NOUN
ap-8720	80	6	for	for	ADP
ap-8720	80	7	20	20	NUM
ap-8720	80	8	frames	frame	NOUN
ap-8720	80	9	,	,	PUNCT
ap-8720	80	10	97.1	97.1	NUM
ap-8720	80	11	%	%	NOUN
ap-8720	80	12	for	for	ADP
ap-8720	80	13	40	40	NUM
ap-8720	80	14	frames	frame	NOUN
ap-8720	80	15	,	,	PUNCT
ap-8720	80	16	and	and	CCONJ
ap-8720	80	17	97.1	97.1	NUM
ap-8720	80	18	%	%	NOUN
ap-8720	80	19	for	for	ADP
ap-8720	80	20	80	80	NUM
ap-8720	80	21	frames	frame	NOUN
ap-8720	80	22	.	.	PUNCT
ap-8720	81	1	liu	liu	PROPN
ap-8720	81	2	et	et	PROPN
ap-8720	81	3	al	al	PROPN
ap-8720	81	4	.	.	PUNCT
ap-8720	82	1	[	[	X
ap-8720	82	2	17	17	NUM
ap-8720	82	3	]	]	PUNCT
ap-8720	82	4	,	,	PUNCT
ap-8720	82	5	showed	show	VERB
ap-8720	82	6	that	that	SCONJ
ap-8720	82	7	cnn	cnn	PROPN
ap-8720	82	8	models	model	NOUN
ap-8720	82	9	are	be	AUX
ap-8720	82	10	strongly	strongly	ADV
ap-8720	82	11	texture	texture	NOUN
ap-8720	82	12	-	-	PUNCT
ap-8720	82	13	based	base	VERB
ap-8720	82	14	rather	rather	ADV
ap-8720	82	15	than	than	ADP
ap-8720	82	16	shape	shape	NOUN
ap-8720	82	17	-	-	PUNCT
ap-8720	82	18	based	base	VERB
ap-8720	82	19	.	.	PUNCT
ap-8720	83	1	they	they	PRON
ap-8720	83	2	found	find	VERB
ap-8720	83	3	that	that	SCONJ
ap-8720	83	4	these	these	DET
ap-8720	83	5	models	model	NOUN
ap-8720	83	6	are	be	AUX
ap-8720	83	7	focus	focus	VERB
ap-8720	83	8	predominantly	predominantly	ADV
ap-8720	83	9	on	on	ADP
ap-8720	83	10	texture	texture	ADJ
ap-8720	83	11	regions	region	NOUN
ap-8720	83	12	,	,	PUNCT
ap-8720	83	13	such	such	ADJ
ap-8720	83	14	as	as	ADP
ap-8720	83	15	skin	skin	NOUN
ap-8720	83	16	and	and	CCONJ
ap-8720	83	17	hair	hair	NOUN
ap-8720	83	18	,	,	PUNCT
ap-8720	83	19	while	while	SCONJ
ap-8720	83	20	giving	give	VERB
ap-8720	83	21	less	less	ADJ
ap-8720	83	22	importance	importance	NOUN
ap-8720	83	23	to	to	ADP
ap-8720	83	24	other	other	ADJ
ap-8720	83	25	regions	region	NOUN
ap-8720	83	26	.	.	PUNCT
ap-8720	84	1	to	to	PART
ap-8720	84	2	investigate	investigate	VERB
ap-8720	84	3	this	this	DET
ap-8720	84	4	further	far	ADV
ap-8720	84	5	,	,	PUNCT
ap-8720	84	6	the	the	DET
ap-8720	84	7	researchers	researcher	NOUN
ap-8720	84	8	conducted	conduct	VERB
ap-8720	84	9	experiments	experiment	NOUN
ap-8720	84	10	specifically	specifically	ADV
ap-8720	84	11	targeting	target	VERB
ap-8720	84	12	skin	skin	NOUN
ap-8720	84	13	regions	region	NOUN
ap-8720	84	14	,	,	PUNCT
ap-8720	84	15	which	which	PRON
ap-8720	84	16	are	be	AUX
ap-8720	84	17	rich	rich	ADJ
ap-8720	84	18	in	in	ADP
ap-8720	84	19	texture	texture	ADJ
ap-8720	84	20	information	information	NOUN
ap-8720	84	21	,	,	PUNCT
ap-8720	84	22	but	but	CCONJ
ap-8720	84	23	relatively	relatively	ADV
ap-8720	84	24	poor	poor	ADJ
ap-8720	84	25	in	in	ADP
ap-8720	84	26	structural	structural	ADJ
ap-8720	84	27	information	information	NOUN
ap-8720	84	28	.	.	PUNCT
ap-8720	85	1	based	base	VERB
ap-8720	85	2	on	on	ADP
ap-8720	85	3	their	their	PRON
ap-8720	85	4	findings	finding	NOUN
ap-8720	85	5	,	,	PUNCT
ap-8720	85	6	they	they	PRON
ap-8720	85	7	concluded	conclude	VERB
ap-8720	85	8	that	that	SCONJ
ap-8720	85	9	the	the	DET
ap-8720	85	10	texture	texture	NOUN
ap-8720	85	11	of	of	ADP
ap-8720	85	12	fake	fake	ADJ
ap-8720	85	13	faces	face	NOUN
ap-8720	85	14	differs	differ	VERB
ap-8720	85	15	from	from	ADP
ap-8720	85	16	that	that	PRON
ap-8720	85	17	of	of	ADP
ap-8720	85	18	real	real	ADJ
ap-8720	85	19	faces	face	NOUN
ap-8720	85	20	.	.	PUNCT
ap-8720	86	1	based	base	VERB
ap-8720	86	2	on	on	ADP
ap-8720	86	3	these	these	DET
ap-8720	86	4	observations	observation	NOUN
ap-8720	86	5	,	,	PUNCT
ap-8720	86	6	liu	liu	PROPN
ap-8720	86	7	et	et	PROPN
ap-8720	86	8	al	al	PROPN
ap-8720	86	9	.	.	PUNCT
ap-8720	87	1	[	[	X
ap-8720	87	2	17	17	NUM
ap-8720	87	3	]	]	PUNCT
ap-8720	87	4	introduced	introduce	VERB
ap-8720	87	5	a	a	DET
ap-8720	87	6	novel	novel	ADJ
ap-8720	87	7	architecture	architecture	NOUN
ap-8720	87	8	called	call	VERB
ap-8720	87	9	gram	gram	NOUN
ap-8720	87	10	-	-	PUNCT
ap-8720	87	11	net	net	NOUN
ap-8720	87	12	,	,	PUNCT
ap-8720	87	13	which	which	PRON
ap-8720	87	14	improves	improve	VERB
ap-8720	87	15	the	the	DET
ap-8720	87	16	robustness	robustness	NOUN
ap-8720	87	17	and	and	CCONJ
ap-8720	87	18	generalisation	generalisation	NOUN
ap-8720	87	19	capabilities	capability	NOUN
ap-8720	87	20	of	of	ADP
ap-8720	87	21	cnns	cnn	NOUN
ap-8720	87	22	in	in	ADP
ap-8720	87	23	the	the	DET
ap-8720	87	24	detection	detection	NOUN
ap-8720	87	25	of	of	ADP
ap-8720	87	26	fake	fake	ADJ
ap-8720	87	27	faces	face	NOUN
ap-8720	87	28	.	.	PUNCT
ap-8720	88	1	the	the	DET
ap-8720	88	2	evaluation	evaluation	NOUN
ap-8720	88	3	of	of	ADP
ap-8720	88	4	their	their	PRON
ap-8720	88	5	approach	approach	NOUN
ap-8720	88	6	using	use	VERB
ap-8720	88	7	stylegan	stylegan	NOUN
ap-8720	88	8	[	[	X
ap-8720	88	9	18	18	NUM
ap-8720	88	10	]	]	PUNCT
ap-8720	88	11	and	and	CCONJ
ap-8720	88	12	celeba	celeba	PROPN
ap-8720	88	13	-	-	PUNCT
ap-8720	88	14	hq	hq	PROPN
ap-8720	89	1	[	[	X
ap-8720	89	2	19	19	NUM
ap-8720	89	3	]	]	ADJ
ap-8720	89	4	databases	database	NOUN
ap-8720	89	5	during	during	ADP
ap-8720	89	6	the	the	DET
ap-8720	89	7	test	test	NOUN
ap-8720	89	8	phase	phase	NOUN
ap-8720	89	9	yielded	yield	VERB
ap-8720	89	10	an	an	DET
ap-8720	89	11	estimated	estimate	VERB
ap-8720	89	12	accuracy	accuracy	NOUN
ap-8720	89	13	of	of	ADP
ap-8720	89	14	80.72	80.72	NUM
ap-8720	89	15	%	%	NOUN
ap-8720	89	16	in	in	ADP
ap-8720	89	17	fake	fake	ADJ
ap-8720	89	18	face	face	NOUN
ap-8720	89	19	detection	detection	NOUN
ap-8720	89	20	.	.	PUNCT
ap-8720	90	1	these	these	DET
ap-8720	90	2	results	result	NOUN
ap-8720	90	3	show	show	VERB
ap-8720	90	4	a	a	DET
ap-8720	90	5	promising	promising	ADJ
ap-8720	90	6	and	and	CCONJ
ap-8720	90	7	encouraging	encourage	VERB
ap-8720	90	8	direction	direction	NOUN
ap-8720	90	9	for	for	ADP
ap-8720	90	10	understanding	understand	VERB
ap-8720	90	11	false	false	ADJ
ap-8720	90	12	images	image	NOUN
ap-8720	90	13	from	from	ADP
ap-8720	90	14	gans	gan	NOUN
ap-8720	90	15	.	.	PUNCT
ap-8720	91	1	the	the	DET
ap-8720	91	2	detection	detection	NOUN
ap-8720	91	3	of	of	ADP
ap-8720	91	4	the	the	DET
ap-8720	91	5	source	source	NOUN
ap-8720	91	6	of	of	ADP
ap-8720	91	7	deepfake	deepfake	NOUN
ap-8720	91	8	videos	video	NOUN
ap-8720	91	9	was	be	AUX
ap-8720	91	10	performed	perform	VERB
ap-8720	91	11	by	by	ADP
ap-8720	91	12	analysing	analyse	VERB
ap-8720	91	13	the	the	DET
ap-8720	91	14	“	"	PUNCT
ap-8720	91	15	heartbeat	heartbeat	NOUN
ap-8720	91	16	”	"	PUNCT
ap-8720	91	17	of	of	ADP
ap-8720	91	18	deep	deep	ADJ
ap-8720	91	19	fakes	fake	NOUN
ap-8720	91	20	and	and	CCONJ
ap-8720	91	21	spatial	spatial	ADJ
ap-8720	91	22	and	and	CCONJ
ap-8720	91	23	temporal	temporal	ADJ
ap-8720	91	24	feature	feature	NOUN
ap-8720	91	25	analysis	analysis	NOUN
ap-8720	91	26	.	.	PUNCT
ap-8720	92	1	for	for	ADP
ap-8720	92	2	the	the	DET
ap-8720	92	3	detection	detection	NOUN
ap-8720	92	4	of	of	ADP
ap-8720	92	5	deepfake	deepfake	NOUN
ap-8720	92	6	images	image	NOUN
ap-8720	92	7	,	,	PUNCT
ap-8720	92	8	a	a	DET
ap-8720	92	9	hybrid	hybrid	ADJ
ap-8720	92	10	approach	approach	NOUN
ap-8720	92	11	was	be	AUX
ap-8720	92	12	introduced	introduce	VERB
ap-8720	92	13	,	,	PUNCT
ap-8720	92	14	which	which	PRON
ap-8720	92	15	uses	use	VERB
ap-8720	92	16	a	a	DET
ap-8720	92	17	pairwise	pairwise	NOUN
ap-8720	92	18	-	-	PUNCT
ap-8720	92	19	learning	learning	NOUN
ap-8720	92	20	;	;	PUNCT
ap-8720	92	21	the	the	DET
ap-8720	92	22	approach	approach	NOUN
ap-8720	92	23	first	first	ADV
ap-8720	92	24	uses	use	VERB
ap-8720	92	25	gans	gan	NOUN
ap-8720	92	26	to	to	PART
ap-8720	92	27	create	create	VERB
ap-8720	92	28	and	and	CCONJ
ap-8720	92	29	generate	generate	VERB
ap-8720	92	30	a	a	DET
ap-8720	92	31	fake	fake	ADJ
ap-8720	92	32	image	image	NOUN
ap-8720	92	33	.	.	PUNCT
ap-8720	93	1	then	then	ADV
ap-8720	93	2	,	,	PUNCT
ap-8720	93	3	on	on	ADP
ap-8720	93	4	the	the	DET
ap-8720	93	5	popular	popular	ADJ
ap-8720	93	6	fake	fake	ADJ
ap-8720	93	7	feature	feature	NOUN
ap-8720	93	8	network	network	NOUN
ap-8720	93	9	(	(	PUNCT
ap-8720	93	10	cffn	cffn	NOUN
ap-8720	93	11	)	)	PUNCT
ap-8720	93	12	generated	generate	VERB
ap-8720	93	13	by	by	ADP
ap-8720	93	14	gans	gan	NOUN
ap-8720	93	15	,	,	PUNCT
ap-8720	93	16	a	a	DET
ap-8720	93	17	pairwise	pairwise	NOUN
ap-8720	93	18	learning	learn	VERB
ap-8720	93	19	model	model	NOUN
ap-8720	93	20	is	be	AUX
ap-8720	93	21	used	use	VERB
ap-8720	93	22	to	to	PART
ap-8720	93	23	capture	capture	VERB
ap-8720	93	24	the	the	DET
ap-8720	93	25	discriminative	discriminative	NOUN
ap-8720	93	26	information	information	NOUN
ap-8720	93	27	between	between	ADP
ap-8720	93	28	the	the	DET
ap-8720	93	29	fake	fake	ADJ
ap-8720	93	30	and	and	CCONJ
ap-8720	93	31	real	real	ADJ
ap-8720	93	32	images	image	NOUN
ap-8720	93	33	.	.	PUNCT
ap-8720	94	1	a.	a.	PROPN
ap-8720	94	2	deshmukh	deshmukh	PROPN
ap-8720	94	3	and	and	CCONJ
ap-8720	94	4	s.	s.	PROPN
ap-8720	94	5	b.	b.	PROPN
ap-8720	94	6	wankhade	wankhade	PROPN
ap-8720	95	1	[	[	X
ap-8720	95	2	20	20	NUM
ap-8720	95	3	]	]	PUNCT
ap-8720	95	4	conducted	conduct	VERB
ap-8720	95	5	a	a	DET
ap-8720	95	6	study	study	NOUN
ap-8720	95	7	on	on	ADP
ap-8720	95	8	the	the	DET
ap-8720	95	9	tools	tool	NOUN
ap-8720	95	10	and	and	CCONJ
ap-8720	95	11	algorithms	algorithm	NOUN
ap-8720	95	12	used	use	VERB
ap-8720	95	13	to	to	PART
ap-8720	95	14	create	create	VERB
ap-8720	95	15	and	and	CCONJ
ap-8720	95	16	detect	detect	VERB
ap-8720	95	17	deepfakes	deepfake	NOUN
ap-8720	95	18	and	and	CCONJ
ap-8720	95	19	the	the	DET
ap-8720	95	20	strategies	strategy	NOUN
ap-8720	95	21	associated	associate	VERB
ap-8720	95	22	with	with	ADP
ap-8720	95	23	them	they	PRON
ap-8720	95	24	.	.	PUNCT
ap-8720	96	1	they	they	PRON
ap-8720	96	2	also	also	ADV
ap-8720	96	3	present	present	VERB
ap-8720	96	4	the	the	DET
ap-8720	96	5	ideologies	ideology	NOUN
ap-8720	96	6	of	of	ADP
ap-8720	96	7	deepfake	deepfake	NOUN
ap-8720	96	8	algorithms	algorithm	NOUN
ap-8720	96	9	and	and	CCONJ
ap-8720	96	10	how	how	SCONJ
ap-8720	96	11	deep	deep	ADJ
ap-8720	96	12	learning	learning	NOUN
ap-8720	96	13	has	have	AUX
ap-8720	96	14	been	be	AUX
ap-8720	96	15	used	use	VERB
ap-8720	96	16	to	to	PART
ap-8720	96	17	enable	enable	VERB
ap-8720	96	18	such	such	ADJ
ap-8720	96	19	technologies	technology	NOUN
ap-8720	96	20	.	.	PUNCT
ap-8720	97	1	according	accord	VERB
ap-8720	97	2	to	to	ADP
ap-8720	97	3	their	their	PRON
ap-8720	97	4	findings	finding	NOUN
ap-8720	97	5	,	,	PUNCT
ap-8720	97	6	the	the	DET
ap-8720	97	7	most	most	ADV
ap-8720	97	8	commonly	commonly	ADV
ap-8720	97	9	used	use	VERB
ap-8720	97	10	deepface	deepface	NOUN
ap-8720	97	11	tools	tool	NOUN
ap-8720	97	12	are	be	AUX
ap-8720	97	13	dfaker	dfaker	NOUN
ap-8720	97	14	,	,	PUNCT
ap-8720	97	15	faceswap	faceswap	NOUN
ap-8720	97	16	-	-	PUNCT
ap-8720	97	17	gan	gin	VERB
ap-8720	97	18	,	,	PUNCT
ap-8720	97	19	faceswap	faceswap	NOUN
ap-8720	97	20	,	,	PUNCT
ap-8720	97	21	deepfakerlab	deepfakerlab	PROPN
ap-8720	97	22	,	,	PUNCT
ap-8720	97	23	deepfakertk	deepfakertk	NOUN
ap-8720	97	24	.	.	PUNCT
ap-8720	98	1	deepfake	deepfake	NOUN
ap-8720	98	2	detection	detection	NOUN
ap-8720	98	3	methods	method	NOUN
ap-8720	98	4	are	be	AUX
ap-8720	98	5	divided	divide	VERB
ap-8720	98	6	into	into	ADP
ap-8720	98	7	two	two	NUM
ap-8720	98	8	dominant	dominant	ADJ
ap-8720	98	9	classes	class	NOUN
ap-8720	98	10	.	.	PUNCT
ap-8720	99	1	the	the	DET
ap-8720	99	2	first	first	ADJ
ap-8720	99	3	class	class	NOUN
ap-8720	99	4	comprises	comprise	VERB
ap-8720	99	5	false	false	ADJ
ap-8720	99	6	image	image	NOUN
ap-8720	99	7	detection	detection	NOUN
ap-8720	99	8	approaches	approach	NOUN
ap-8720	99	9	that	that	PRON
ap-8720	99	10	rely	rely	VERB
ap-8720	99	11	on	on	ADP
ap-8720	99	12	deep	deep	ADJ
ap-8720	99	13	learning	learning	NOUN
ap-8720	99	14	techniques	technique	NOUN
ap-8720	99	15	such	such	ADJ
ap-8720	99	16	as	as	ADP
ap-8720	99	17	cnn	cnn	PROPN
ap-8720	99	18	,	,	PUNCT
ap-8720	99	19	svm	svm	ADJ
ap-8720	99	20	,	,	PUNCT
ap-8720	99	21	random	random	ADJ
ap-8720	99	22	forest	forest	NOUN
ap-8720	99	23	,	,	PUNCT
ap-8720	99	24	multilayer	multilayer	PROPN
ap-8720	99	25	perceptron	perceptron	PROPN
ap-8720	99	26	,	,	PUNCT
ap-8720	99	27	and	and	CCONJ
ap-8720	99	28	gan	gin	VERB
ap-8720	100	1	[	[	PUNCT
ap-8720	100	2	21	21	NUM
ap-8720	100	3	]	]	PUNCT
ap-8720	100	4	.	.	PUNCT
ap-8720	101	1	these	these	DET
ap-8720	101	2	methods	method	NOUN
ap-8720	101	3	generate	generate	VERB
ap-8720	101	4	images	image	NOUN
ap-8720	101	5	that	that	PRON
ap-8720	101	6	are	be	AUX
ap-8720	101	7	noticeable	noticeable	ADJ
ap-8720	101	8	as	as	ADP
ap-8720	101	9	fake	fake	ADJ
ap-8720	101	10	.	.	PUNCT
ap-8720	102	1	the	the	DET
ap-8720	102	2	second	second	ADJ
ap-8720	102	3	class	class	NOUN
ap-8720	102	4	,	,	PUNCT
ap-8720	102	5	known	know	VERB
ap-8720	102	6	as	as	ADP
ap-8720	102	7	the	the	DET
ap-8720	102	8	false	false	ADJ
ap-8720	102	9	video	video	NOUN
ap-8720	102	10	detection	detection	NOUN
ap-8720	102	11	approach	approach	NOUN
ap-8720	102	12	,	,	PUNCT
ap-8720	102	13	can	can	AUX
ap-8720	102	14	also	also	ADV
ap-8720	102	15	be	be	AUX
ap-8720	102	16	divided	divide	VERB
ap-8720	102	17	into	into	ADP
ap-8720	102	18	two	two	NUM
ap-8720	102	19	groups	group	NOUN
ap-8720	102	20	.	.	PUNCT
ap-8720	103	1	the	the	DET
ap-8720	103	2	first	first	ADJ
ap-8720	103	3	group	group	NOUN
ap-8720	103	4	uses	use	VERB
ap-8720	103	5	chronological	chronological	ADJ
ap-8720	103	6	features	feature	NOUN
ap-8720	103	7	to	to	PART
ap-8720	103	8	identify	identify	VERB
ap-8720	103	9	deepfake	deepfake	NOUN
ap-8720	103	10	videos	video	NOUN
ap-8720	103	11	,	,	PUNCT
ap-8720	103	12	while	while	SCONJ
ap-8720	103	13	the	the	DET
ap-8720	103	14	second	second	ADJ
ap-8720	103	15	group	group	NOUN
ap-8720	103	16	focuses	focus	VERB
ap-8720	103	17	on	on	ADP
ap-8720	103	18	exploring	explore	VERB
ap-8720	103	19	visual	visual	ADJ
ap-8720	103	20	artfacts	artfact	NOUN
ap-8720	103	21	.	.	PUNCT
ap-8720	104	1	the	the	DET
ap-8720	104	2	researchers	researcher	NOUN
ap-8720	104	3	have	have	AUX
ap-8720	104	4	proposed	propose	VERB
ap-8720	104	5	a	a	DET
ap-8720	104	6	pipeline	pipeline	NOUN
ap-8720	104	7	technique	technique	NOUN
ap-8720	104	8	that	that	PRON
ap-8720	104	9	uses	use	VERB
ap-8720	104	10	cnn	cnn	PROPN
ap-8720	104	11	and	and	CCONJ
ap-8720	104	12	lstm	lstm	NOUN
ap-8720	104	13	to	to	PART
ap-8720	104	14	detect	detect	VERB
ap-8720	104	15	deepfake	deepfake	NOUN
ap-8720	104	16	videos	video	NOUN
ap-8720	104	17	from	from	ADP
ap-8720	104	18	blinking	blink	VERB
ap-8720	104	19	eyes	eye	NOUN
ap-8720	104	20	.	.	PUNCT
ap-8720	105	1	the	the	DET
ap-8720	105	2	authors	author	NOUN
ap-8720	105	3	verified	verify	VERB
ap-8720	105	4	the	the	DET
ap-8720	105	5	proposed	propose	VERB
ap-8720	105	6	techniques	technique	NOUN
ap-8720	105	7	on	on	ADP
ap-8720	105	8	the	the	DET
ap-8720	105	9	following	follow	VERB
ap-8720	105	10	datasets	dataset	NOUN
ap-8720	105	11	:	:	PUNCT
ap-8720	105	12	forensics++	forensics++	PROPN
ap-8720	105	13	dataset	dataset	PROPN
ap-8720	105	14	,	,	PUNCT
ap-8720	105	15	the	the	DET
ap-8720	105	16	uadfv	uadfv	NOUN
ap-8720	105	17	,	,	PUNCT
ap-8720	105	18	and	and	CCONJ
ap-8720	105	19	deepfaketimit	deepfaketimit	ADJ
ap-8720	105	20	datasets	dataset	NOUN
ap-8720	105	21	.	.	PUNCT
ap-8720	106	1	during	during	ADP
ap-8720	106	2	the	the	DET
ap-8720	106	3	tests	test	NOUN
ap-8720	106	4	conducted	conduct	VERB
ap-8720	106	5	,	,	PUNCT
ap-8720	106	6	the	the	DET
ap-8720	106	7	authors	author	NOUN
ap-8720	106	8	found	find	VERB
ap-8720	106	9	that	that	SCONJ
ap-8720	106	10	general	general	ADJ
ap-8720	106	11	facial	facial	ADJ
ap-8720	106	12	recognition	recognition	NOUN
ap-8720	106	13	systems	system	NOUN
ap-8720	106	14	,	,	PUNCT
ap-8720	106	15	such	such	ADJ
ap-8720	106	16	as	as	ADP
ap-8720	106	17	vgg	vgg	NOUN
ap-8720	106	18	and	and	CCONJ
ap-8720	106	19	facenet	facenet	NOUN
ap-8720	106	20	,	,	PUNCT
ap-8720	106	21	are	be	AUX
ap-8720	106	22	unable	unable	ADJ
ap-8720	106	23	to	to	PART
ap-8720	106	24	successfully	successfully	ADV
ap-8720	106	25	detect	detect	VERB
ap-8720	106	26	deepfakes	deepfake	NOUN
ap-8720	106	27	.	.	PUNCT
ap-8720	107	1	in	in	ADP
ap-8720	107	2	addition	addition	NOUN
ap-8720	107	3	,	,	PUNCT
ap-8720	107	4	the	the	DET
ap-8720	107	5	use	use	NOUN
ap-8720	107	6	of	of	ADP
ap-8720	107	7	svm	svm	NOUN
ap-8720	107	8	to	to	PART
ap-8720	107	9	assess	assess	VERB
ap-8720	107	10	image	image	NOUN
ap-8720	107	11	quality	quality	NOUN
ap-8720	107	12	and	and	CCONJ
ap-8720	107	13	lip	lip	NOUN
ap-8720	107	14	synchronisation	synchronisation	NOUN
ap-8720	107	15	resulted	result	VERB
ap-8720	107	16	in	in	ADP
ap-8720	107	17	a	a	DET
ap-8720	107	18	significantly	significantly	ADV
ap-8720	107	19	high	high	ADJ
ap-8720	107	20	error	error	NOUN
ap-8720	107	21	rate	rate	NOUN
ap-8720	107	22	.	.	PUNCT
ap-8720	108	1	the	the	DET
ap-8720	108	2	aim	aim	NOUN
ap-8720	108	3	of	of	ADP
ap-8720	108	4	the	the	DET
ap-8720	108	5	work	work	NOUN
ap-8720	108	6	by	by	ADP
ap-8720	108	7	jung	jung	PROPN
ap-8720	108	8	et	et	PROPN
ap-8720	108	9	al	al	PROPN
ap-8720	108	10	.	.	PUNCT
ap-8720	109	1	[	[	X
ap-8720	109	2	22	22	NUM
ap-8720	109	3	]	]	PUNCT
ap-8720	109	4	is	be	AUX
ap-8720	109	5	to	to	PART
ap-8720	109	6	detect	detect	VERB
ap-8720	109	7	deepfakes	deepfake	NOUN
ap-8720	109	8	generated	generate	VERB
ap-8720	109	9	by	by	ADP
ap-8720	109	10	generative	generative	ADJ
ap-8720	109	11	adversarial	adversarial	ADJ
ap-8720	109	12	networks	network	NOUN
ap-8720	109	13	(	(	PUNCT
ap-8720	109	14	gans	gan	NOUN
ap-8720	109	15	)	)	PUNCT
ap-8720	109	16	using	use	VERB
ap-8720	109	17	blink	blink	ADJ
ap-8720	109	18	analysis	analysis	NOUN
ap-8720	109	19	.	.	PUNCT
ap-8720	110	1	the	the	DET
ap-8720	110	2	authors	author	NOUN
ap-8720	110	3	propose	propose	VERB
ap-8720	110	4	a	a	DET
ap-8720	110	5	method	method	NOUN
ap-8720	110	6	called	call	VERB
ap-8720	110	7	deepvision	deepvision	NOUN
ap-8720	110	8	that	that	PRON
ap-8720	110	9	incorporates	incorporate	VERB
ap-8720	110	10	machine	machine	NOUN
ap-8720	110	11	learning	learning	NOUN
ap-8720	110	12	techniques	technique	NOUN
ap-8720	110	13	.	.	PUNCT
ap-8720	111	1	the	the	DET
ap-8720	111	2	deepvision	deepvision	NOUN
ap-8720	111	3	architecture	architecture	NOUN
ap-8720	111	4	has	have	VERB
ap-8720	111	5	a	a	DET
ap-8720	111	6	pre	pre	NOUN
ap-8720	111	7	-	-	NOUN
ap-8720	111	8	processor	processor	NOUN
ap-8720	111	9	that	that	PRON
ap-8720	111	10	receives	receive	VERB
ap-8720	111	11	input	input	NOUN
ap-8720	111	12	information	information	NOUN
ap-8720	111	13	.	.	PUNCT
ap-8720	112	1	this	this	DET
ap-8720	112	2	input	input	NOUN
ap-8720	112	3	data	datum	NOUN
ap-8720	112	4	includes	include	VERB
ap-8720	112	5	parameters	parameter	NOUN
ap-8720	112	6	such	such	ADJ
ap-8720	112	7	as	as	ADP
ap-8720	112	8	gender	gender	NOUN
ap-8720	112	9	,	,	PUNCT
ap-8720	112	10	age	age	NOUN
ap-8720	112	11	,	,	PUNCT
ap-8720	112	12	activity	activity	NOUN
ap-8720	112	13	,	,	PUNCT
ap-8720	112	14	and	and	CCONJ
ap-8720	112	15	time	time	NOUN
ap-8720	112	16	,	,	PUNCT
ap-8720	112	17	which	which	PRON
ap-8720	112	18	play	play	VERB
ap-8720	112	19	a	a	DET
ap-8720	112	20	crucial	crucial	ADJ
ap-8720	112	21	role	role	NOUN
ap-8720	112	22	in	in	ADP
ap-8720	112	23	assessing	assess	VERB
ap-8720	112	24	changes	change	NOUN
ap-8720	112	25	in	in	ADP
ap-8720	112	26	the	the	DET
ap-8720	112	27	human	human	ADJ
ap-8720	112	28	eye	eye	NOUN
ap-8720	112	29	blink	blink	VERB
ap-8720	112	30	.	.	PUNCT
ap-8720	113	1	to	to	PART
ap-8720	113	2	perform	perform	VERB
ap-8720	113	3	the	the	DET
ap-8720	113	4	detection	detection	NOUN
ap-8720	113	5	process	process	NOUN
ap-8720	113	6	,	,	PUNCT
ap-8720	113	7	deepvision	deepvision	NOUN
ap-8720	113	8	uses	use	VERB
ap-8720	113	9	the	the	DET
ap-8720	113	10	target	target	NOUN
ap-8720	113	11	detector	detector	NOUN
ap-8720	113	12	algorithm	algorithm	NOUN
ap-8720	113	13	for	for	ADP
ap-8720	113	14	face	face	NOUN
ap-8720	113	15	detection	detection	NOUN
ap-8720	113	16	and	and	CCONJ
ap-8720	113	17	then	then	ADV
ap-8720	113	18	uses	use	VERB
ap-8720	113	19	the	the	DET
ap-8720	113	20	eye	eye	NOUN
ap-8720	113	21	tracker	tracker	NOUN
ap-8720	113	22	algorithm	algorithm	NOUN
ap-8720	113	23	to	to	PART
ap-8720	113	24	track	track	VERB
ap-8720	113	25	and	and	CCONJ
ap-8720	113	26	measure	measure	VERB
ap-8720	113	27	the	the	DET
ap-8720	113	28	number	number	NOUN
ap-8720	113	29	of	of	ADP
ap-8720	113	30	blink	blink	ADJ
ap-8720	113	31	repetitions	repetition	NOUN
ap-8720	113	32	and	and	CCONJ
ap-8720	113	33	their	their	PRON
ap-8720	113	34	duration	duration	NOUN
ap-8720	113	35	.	.	PUNCT
ap-8720	114	1	these	these	DET
ap-8720	114	2	measurements	measurement	NOUN
ap-8720	114	3	are	be	AUX
ap-8720	114	4	then	then	ADV
ap-8720	114	5	compared	compare	VERB
ap-8720	114	6	to	to	ADP
ap-8720	114	7	a	a	DET
ap-8720	114	8	database	database	NOUN
ap-8720	114	9	of	of	ADP
ap-8720	114	10	natural	natural	ADJ
ap-8720	114	11	movement	movement	NOUN
ap-8720	114	12	to	to	PART
ap-8720	114	13	determine	determine	VERB
ap-8720	114	14	whether	whether	SCONJ
ap-8720	114	15	the	the	DET
ap-8720	114	16	blink	blink	NOUN
ap-8720	114	17	is	be	AUX
ap-8720	114	18	natural	natural	ADJ
ap-8720	114	19	.	.	PUNCT
ap-8720	115	1	deepvision	deepvision	NOUN
ap-8720	115	2	achieves	achieve	VERB
ap-8720	115	3	87.5	87.5	NUM
ap-8720	115	4	%	%	NOUN
ap-8720	115	5	accuracy	accuracy	NOUN
ap-8720	115	6	in	in	ADP
ap-8720	115	7	detecting	detect	VERB
ap-8720	115	8	deepfake	deepfake	NOUN
ap-8720	115	9	videos	video	NOUN
ap-8720	115	10	.	.	PUNCT
ap-8720	116	1	it	it	PRON
ap-8720	116	2	should	should	AUX
ap-8720	116	3	be	be	AUX
ap-8720	116	4	noted	note	VERB
ap-8720	116	5	,	,	PUNCT
ap-8720	116	6	however	however	ADV
ap-8720	116	7	,	,	PUNCT
ap-8720	116	8	this	this	DET
ap-8720	116	9	method	method	NOUN
ap-8720	116	10	may	may	AUX
ap-8720	116	11	be	be	AUX
ap-8720	116	12	less	less	ADV
ap-8720	116	13	effective	effective	ADJ
ap-8720	116	14	for	for	ADP
ap-8720	116	15	people	people	NOUN
ap-8720	116	16	with	with	ADP
ap-8720	116	17	mental	mental	ADJ
ap-8720	116	18	illness	illness	NOUN
ap-8720	116	19	,	,	PUNCT
ap-8720	116	20	as	as	SCONJ
ap-8720	116	21	abnormal	abnormal	ADJ
ap-8720	116	22	blinking	blinking	NOUN
ap-8720	116	23	is	be	AUX
ap-8720	116	24	often	often	ADV
ap-8720	116	25	observed	observe	VERB
ap-8720	116	26	in	in	ADP
ap-8720	116	27	such	such	ADJ
ap-8720	116	28	cases	case	NOUN
ap-8720	116	29	.	.	PUNCT
ap-8720	117	1	korshunov	korshunov	PROPN
ap-8720	117	2	and	and	CCONJ
ap-8720	117	3	marcel	marcel	PROPN
ap-8720	117	4	[	[	X
ap-8720	117	5	23	23	NUM
ap-8720	117	6	]	]	PUNCT
ap-8720	117	7	presented	present	VERB
ap-8720	117	8	two	two	NUM
ap-8720	117	9	sets	set	NOUN
ap-8720	117	10	of	of	ADP
ap-8720	117	11	deepfake	deepfake	NOUN
ap-8720	117	12	videos	video	NOUN
ap-8720	117	13	generated	generate	VERB
ap-8720	117	14	from	from	ADP
ap-8720	117	15	the	the	DET
ap-8720	117	16	vidtimit	vidtimit	NOUN
ap-8720	117	17	database	database	NOUN
ap-8720	117	18	.	.	PUNCT
ap-8720	118	1	first	first	ADV
ap-8720	118	2	,	,	PUNCT
ap-8720	118	3	they	they	PRON
ap-8720	118	4	generated	generate	VERB
ap-8720	118	5	videos	video	NOUN
ap-8720	118	6	of	of	ADP
ap-8720	118	7	different	different	ADJ
ap-8720	118	8	resolution	resolution	NOUN
ap-8720	118	9	qualities	quality	NOUN
ap-8720	118	10	:	:	PUNCT
ap-8720	118	11	(	(	PUNCT
ap-8720	118	12	64	64	NUM
ap-8720	118	13	×	×	NOUN
ap-8720	118	14	64	64	NUM
ap-8720	118	15	)	)	PUNCT
ap-8720	118	16	and	and	CCONJ
ap-8720	118	17	(	(	PUNCT
ap-8720	118	18	128	128	NUM
ap-8720	118	19	×	×	NOUN
ap-8720	118	20	128	128	NUM
ap-8720	118	21	)	)	PUNCT
ap-8720	118	22	.	.	PUNCT
ap-8720	119	1	then	then	ADV
ap-8720	119	2	,	,	PUNCT
ap-8720	119	3	they	they	PRON
ap-8720	119	4	evaluated	evaluate	VERB
ap-8720	119	5	two	two	NUM
ap-8720	119	6	neural	neural	ADJ
ap-8720	119	7	network	network	NOUN
ap-8720	119	8	based	base	VERB
ap-8720	119	9	systems	system	NOUN
ap-8720	119	10	,	,	PUNCT
ap-8720	119	11	namely	namely	ADV
ap-8720	119	12	vgg	vgg	VERB
ap-8720	120	1	[	[	X
ap-8720	120	2	24	24	NUM
ap-8720	120	3	]	]	PUNCT
ap-8720	120	4	and	and	CCONJ
ap-8720	120	5	facenet	facenet	NOUN
ap-8720	120	6	6	6	NUM
ap-8720	120	7	[	[	X
ap-8720	120	8	25	25	NUM
ap-8720	120	9	]	]	PUNCT
ap-8720	120	10	,	,	PUNCT
ap-8720	120	11	to	to	PART
ap-8720	120	12	distinguish	distinguish	VERB
ap-8720	120	13	between	between	ADP
ap-8720	120	14	deepfake	deepfake	NOUN
ap-8720	120	15	and	and	CCONJ
ap-8720	120	16	real	real	ADJ
ap-8720	120	17	videos	video	NOUN
ap-8720	120	18	.	.	PUNCT
ap-8720	121	1	however	however	ADV
ap-8720	121	2	,	,	PUNCT
ap-8720	121	3	both	both	DET
ap-8720	121	4	systems	system	NOUN
ap-8720	121	5	showed	show	VERB
ap-8720	121	6	a	a	DET
ap-8720	121	7	high	high	ADJ
ap-8720	121	8	error	error	NOUN
ap-8720	121	9	rate	rate	NOUN
ap-8720	121	10	of	of	ADP
ap-8720	121	11	95	95	NUM
ap-8720	121	12	%	%	NOUN
ap-8720	121	13	,	,	PUNCT
ap-8720	121	14	indicating	indicate	VERB
ap-8720	121	15	their	their	PRON
ap-8720	121	16	inability	inability	NOUN
ap-8720	121	17	to	to	PART
ap-8720	121	18	effectively	effectively	ADV
ap-8720	121	19	distinguish	distinguish	VERB
ap-8720	121	20	between	between	ADP
ap-8720	121	21	the	the	DET
ap-8720	121	22	two	two	NUM
ap-8720	121	23	.	.	PUNCT
ap-8720	122	1	to	to	PART
ap-8720	122	2	overcome	overcome	VERB
ap-8720	122	3	this	this	DET
ap-8720	122	4	limitation	limitation	NOUN
ap-8720	122	5	,	,	PUNCT
ap-8720	122	6	the	the	DET
ap-8720	122	7	authors	author	NOUN
ap-8720	122	8	investigated	investigate	VERB
ap-8720	122	9	an	an	DET
ap-8720	122	10	audiovisual	audiovisual	ADJ
ap-8720	122	11	approach	approach	NOUN
ap-8720	122	12	to	to	ADP
ap-8720	122	13	detecting	detect	VERB
ap-8720	122	14	inconsistencies	inconsistency	NOUN
ap-8720	122	15	between	between	ADP
ap-8720	122	16	lip	lip	NOUN
ap-8720	122	17	movements	movement	NOUN
ap-8720	122	18	and	and	CCONJ
ap-8720	122	19	speech	speech	NOUN
ap-8720	122	20	in	in	ADP
ap-8720	122	21	the	the	DET
ap-8720	122	22	audio	audio	ADJ
ap-8720	122	23	track	track	NOUN
ap-8720	122	24	.	.	PUNCT
ap-8720	123	1	they	they	PRON
ap-8720	123	2	found	find	VERB
ap-8720	123	3	that	that	SCONJ
ap-8720	123	4	lip	lip	NOUN
ap-8720	123	5	-	-	PUNCT
ap-8720	123	6	synchronisation	synchronisation	NOUN
ap-8720	123	7	-	-	PUNCT
ap-8720	123	8	based	base	VERB
ap-8720	123	9	methods	method	NOUN
ap-8720	123	10	failed	fail	VERB
ap-8720	123	11	to	to	PART
ap-8720	123	12	detect	detect	VERB
ap-8720	123	13	inconsistencies	inconsistency	NOUN
ap-8720	123	14	between	between	ADP
ap-8720	123	15	lip	lip	NOUN
ap-8720	123	16	movements	movement	NOUN
ap-8720	123	17	and	and	CCONJ
ap-8720	123	18	speech	speech	NOUN
ap-8720	123	19	,	,	PUNCT
ap-8720	123	20	while	while	SCONJ
ap-8720	123	21	pure	pure	ADJ
ap-8720	123	22	image	image	NOUN
ap-8720	123	23	-	-	PUNCT
ap-8720	123	24	based	base	VERB
ap-8720	123	25	approaches	approach	NOUN
ap-8720	123	26	proved	prove	VERB
ap-8720	123	27	effective	effective	ADJ
ap-8720	123	28	in	in	ADP
ap-8720	123	29	detecting	detect	VERB
ap-8720	123	30	video	video	NOUN
ap-8720	123	31	deepfakes	deepfake	NOUN
ap-8720	123	32	.	.	PUNCT
ap-8720	124	1	by	by	ADP
ap-8720	124	2	using	use	VERB
ap-8720	124	3	iqm	iqm	ADJ
ap-8720	124	4	techniques	technique	NOUN
ap-8720	124	5	together	together	ADV
ap-8720	124	6	with	with	ADP
ap-8720	124	7	an	an	DET
ap-8720	124	8	svm	svm	ADJ
ap-8720	124	9	classifier	classifier	NOUN
ap-8720	124	10	,	,	PUNCT
ap-8720	124	11	they	they	PRON
ap-8720	124	12	achieved	achieve	VERB
ap-8720	124	13	a	a	DET
ap-8720	124	14	significantly	significantly	ADV
ap-8720	124	15	lower	low	ADJ
ap-8720	124	16	error	error	NOUN
ap-8720	124	17	rate	rate	NOUN
ap-8720	124	18	of	of	ADP
ap-8720	124	19	8.97	8.97	NUM
ap-8720	124	20	%	%	NOUN
ap-8720	124	21	in	in	ADP
ap-8720	124	22	detecting	detect	VERB
ap-8720	124	23	deepfakes	deepfake	NOUN
ap-8720	124	24	.	.	PUNCT
ap-8720	125	1	li	li	PROPN
ap-8720	125	2	and	and	CCONJ
ap-8720	125	3	lyu	lyu	X
ap-8720	126	1	[	[	X
ap-8720	126	2	26	26	NUM
ap-8720	126	3	]	]	PUNCT
ap-8720	126	4	aimed	aim	VERB
ap-8720	126	5	to	to	PART
ap-8720	126	6	describe	describe	VERB
ap-8720	126	7	a	a	DET
ap-8720	126	8	new	new	ADJ
ap-8720	126	9	method	method	NOUN
ap-8720	126	10	based	base	VERB
ap-8720	126	11	on	on	ADP
ap-8720	126	12	deep	deep	ADJ
ap-8720	126	13	learning	learning	NOUN
ap-8720	126	14	that	that	PRON
ap-8720	126	15	can	can	AUX
ap-8720	126	16	effectively	effectively	ADV
ap-8720	126	17	distinguish	distinguish	VERB
ap-8720	126	18	deepfake	deepfake	NOUN
ap-8720	126	19	videos	video	NOUN
ap-8720	126	20	from	from	ADP
ap-8720	126	21	real	real	ADJ
ap-8720	126	22	ones	one	NOUN
ap-8720	126	23	.	.	PUNCT
ap-8720	127	1	due	due	ADP
ap-8720	127	2	to	to	ADP
ap-8720	127	3	computational	computational	ADJ
ap-8720	127	4	resources	resource	NOUN
ap-8720	127	5	and	and	CCONJ
ap-8720	127	6	production	production	NOUN
ap-8720	127	7	time	time	NOUN
ap-8720	127	8	limitations	limitation	NOUN
ap-8720	127	9	,	,	PUNCT
ap-8720	127	10	the	the	DET
ap-8720	127	11	deepfake	deepfake	NOUN
ap-8720	127	12	algorithm	algorithm	NOUN
ap-8720	127	13	can	can	AUX
ap-8720	127	14	not	not	PART
ap-8720	127	15	distinguish	distinguish	VERB
ap-8720	127	16	between	between	ADP
ap-8720	127	17	real	real	ADJ
ap-8720	127	18	and	and	CCONJ
ap-8720	127	19	fake	fake	ADJ
ap-8720	127	20	videos	video	NOUN
ap-8720	127	21	.	.	PUNCT
ap-8720	128	1	the	the	DET
ap-8720	128	2	images	image	NOUN
ap-8720	128	3	undergo	undergo	VERB
ap-8720	128	4	an	an	DET
ap-8720	128	5	affine	affine	ADJ
ap-8720	128	6	deformation	deformation	NOUN
ap-8720	128	7	corresponding	correspond	VERB
ap-8720	128	8	to	to	ADP
ap-8720	128	9	the	the	DET
ap-8720	128	10	facial	facial	ADJ
ap-8720	128	11	configuration	configuration	NOUN
ap-8720	128	12	of	of	ADP
ap-8720	128	13	the	the	DET
ap-8720	128	14	source	source	NOUN
ap-8720	128	15	.	.	PUNCT
ap-8720	129	1	therefore	therefore	ADV
ap-8720	129	2	,	,	PUNCT
ap-8720	129	3	the	the	DET
ap-8720	129	4	authors	author	NOUN
ap-8720	129	5	proposed	propose	VERB
ap-8720	129	6	a	a	DET
ap-8720	129	7	convolutional	convolutional	ADJ
ap-8720	129	8	neural	neural	ADJ
ap-8720	129	9	network	network	NOUN
ap-8720	129	10	(	(	PUNCT
ap-8720	129	11	cnn	cnn	PROPN
ap-8720	129	12	)	)	PUNCT
ap-8720	129	13	model	model	NOUN
ap-8720	129	14	to	to	PART
ap-8720	129	15	detect	detect	VERB
ap-8720	129	16	the	the	DET
ap-8720	129	17	presence	presence	NOUN
ap-8720	129	18	of	of	ADP
ap-8720	129	19	artifacts	artifact	NOUN
ap-8720	129	20	from	from	ADP
ap-8720	129	21	the	the	DET
ap-8720	129	22	face	face	NOUN
ap-8720	129	23	regions	region	NOUN
ap-8720	129	24	and	and	CCONJ
ap-8720	129	25	surrounding	surround	VERB
ap-8720	129	26	areas	area	NOUN
ap-8720	129	27	.	.	PUNCT
ap-8720	130	1	the	the	DET
ap-8720	130	2	cnn	cnn	PROPN
ap-8720	130	3	training	training	NOUN
ap-8720	130	4	is	be	AUX
ap-8720	130	5	based	base	VERB
ap-8720	130	6	on	on	ADP
ap-8720	130	7	images	image	NOUN
ap-8720	130	8	collected	collect	VERB
ap-8720	130	9	from	from	ADP
ap-8720	130	10	the	the	DET
ap-8720	130	11	internet	internet	NOUN
ap-8720	130	12	(	(	PUNCT
ap-8720	130	13	24442	24442	NUM
ap-8720	130	14	jpeg	jpeg	NOUN
ap-8720	130	15	face	face	VERB
ap-8720	130	16	307	307	NUM
ap-8720	130	17	ch	ch	NOUN
ap-8720	130	18	.	.	PUNCT
ap-8720	131	1	e.	e.	PROPN
ap-8720	131	2	bencheriet	bencheriet	PROPN
ap-8720	131	3	,	,	PUNCT
ap-8720	131	4	h.	h.	PROPN
ap-8720	131	5	abdelmoumène	abdelmoumène	PROPN
ap-8720	131	6	,	,	PUNCT
ap-8720	131	7	a.	a.	PROPN
ap-8720	131	8	sebbagh	sebbagh	PROPN
ap-8720	131	9	et	et	PROPN
ap-8720	131	10	al	al	PROPN
ap-8720	131	11	.	.	PROPN
ap-8720	131	12	acta	acta	PROPN
ap-8720	131	13	polytechnica	polytechnica	PROPN
ap-8720	131	14	images	image	NOUN
ap-8720	131	15	)	)	PUNCT
ap-8720	131	16	.	.	PUNCT
ap-8720	132	1	to	to	PART
ap-8720	132	2	increase	increase	VERB
ap-8720	132	3	the	the	DET
ap-8720	132	4	diversity	diversity	NOUN
ap-8720	132	5	of	of	ADP
ap-8720	132	6	the	the	DET
ap-8720	132	7	training	training	NOUN
ap-8720	132	8	,	,	PUNCT
ap-8720	132	9	the	the	DET
ap-8720	132	10	authors	author	NOUN
ap-8720	132	11	modified	modify	VERB
ap-8720	132	12	the	the	DET
ap-8720	132	13	colour	colour	NOUN
ap-8720	132	14	information	information	NOUN
ap-8720	132	15	,	,	PUNCT
ap-8720	132	16	brightness	brightness	NOUN
ap-8720	132	17	,	,	PUNCT
ap-8720	132	18	contrast	contrast	NOUN
ap-8720	132	19	,	,	PUNCT
ap-8720	132	20	and	and	CCONJ
ap-8720	132	21	sharpness	sharpness	NOUN
ap-8720	132	22	for	for	ADP
ap-8720	132	23	all	all	DET
ap-8720	132	24	training	training	NOUN
ap-8720	132	25	examples	example	NOUN
ap-8720	132	26	;	;	PUNCT
ap-8720	132	27	they	they	PRON
ap-8720	132	28	validated	validate	VERB
ap-8720	132	29	their	their	PRON
ap-8720	132	30	work	work	NOUN
ap-8720	132	31	on	on	ADP
ap-8720	132	32	the	the	DET
ap-8720	132	33	deepfake	deepfake	NOUN
ap-8720	132	34	uadfv	uadfv	PROPN
ap-8720	132	35	video	video	NOUN
ap-8720	132	36	dataset	dataset	VERB
ap-8720	133	1	[	[	X
ap-8720	133	2	12	12	NUM
ap-8720	133	3	]	]	PUNCT
ap-8720	133	4	.	.	PUNCT
ap-8720	134	1	this	this	DET
ap-8720	134	2	dataset	dataset	NOUN
ap-8720	134	3	contains	contain	VERB
ap-8720	134	4	98	98	NUM
ap-8720	134	5	videos	video	NOUN
ap-8720	134	6	(	(	PUNCT
ap-8720	134	7	32752	32752	NUM
ap-8720	134	8	frames	frame	NOUN
ap-8720	134	9	)	)	PUNCT
ap-8720	134	10	,	,	PUNCT
ap-8720	134	11	including	include	VERB
ap-8720	134	12	49	49	NUM
ap-8720	134	13	real	real	ADJ
ap-8720	134	14	and	and	CCONJ
ap-8720	134	15	49	49	NUM
ap-8720	134	16	fake	fake	ADJ
ap-8720	134	17	videos	video	NOUN
ap-8720	134	18	.	.	PUNCT
ap-8720	135	1	the	the	DET
ap-8720	135	2	tests	test	NOUN
ap-8720	135	3	performed	perform	VERB
ap-8720	135	4	on	on	ADP
ap-8720	135	5	the	the	DET
ap-8720	135	6	deepfake	deepfake	NOUN
ap-8720	135	7	videos	video	NOUN
ap-8720	135	8	demonstrate	demonstrate	VERB
ap-8720	135	9	the	the	DET
ap-8720	135	10	effectiveness	effectiveness	NOUN
ap-8720	135	11	of	of	ADP
ap-8720	135	12	the	the	DET
ap-8720	135	13	proposed	propose	VERB
ap-8720	135	14	method	method	NOUN
ap-8720	135	15	in	in	ADP
ap-8720	135	16	practice	practice	NOUN
ap-8720	135	17	with	with	ADP
ap-8720	135	18	an	an	DET
ap-8720	135	19	accuracy	accuracy	NOUN
ap-8720	135	20	rate	rate	NOUN
ap-8720	135	21	of	of	ADP
ap-8720	135	22	up	up	ADP
ap-8720	135	23	to	to	PART
ap-8720	135	24	97	97	NUM
ap-8720	135	25	%	%	NOUN
ap-8720	135	26	.	.	PUNCT
ap-8720	136	1	yang	yang	PROPN
ap-8720	136	2	et	et	PROPN
ap-8720	136	3	al	al	PROPN
ap-8720	136	4	.	.	PUNCT
ap-8720	137	1	[	[	X
ap-8720	137	2	27	27	NUM
ap-8720	137	3	]	]	PUNCT
ap-8720	137	4	introduced	introduce	VERB
ap-8720	137	5	a	a	DET
ap-8720	137	6	pre	pre	ADJ
ap-8720	137	7	-	-	ADJ
ap-8720	137	8	processing	processing	ADJ
ap-8720	137	9	module	module	NOUN
ap-8720	137	10	called	call	VERB
ap-8720	137	11	amten	amten	NOUN
ap-8720	137	12	,	,	PUNCT
ap-8720	137	13	which	which	PRON
ap-8720	137	14	uses	use	VERB
ap-8720	137	15	the	the	DET
ap-8720	137	16	convolutional	convolutional	ADJ
ap-8720	137	17	layer	layer	NOUN
ap-8720	137	18	as	as	ADP
ap-8720	137	19	a	a	DET
ap-8720	137	20	predictor	predictor	NOUN
ap-8720	137	21	to	to	PART
ap-8720	137	22	detect	detect	VERB
ap-8720	137	23	image	image	NOUN
ap-8720	137	24	manipulation	manipulation	NOUN
ap-8720	137	25	.	.	PUNCT
ap-8720	138	1	this	this	DET
ap-8720	138	2	module	module	NOUN
ap-8720	138	3	can	can	AUX
ap-8720	138	4	be	be	AUX
ap-8720	138	5	integrated	integrate	VERB
ap-8720	138	6	into	into	ADP
ap-8720	138	7	cnn	cnn	PROPN
ap-8720	138	8	-	-	PUNCT
ap-8720	138	9	based	base	VERB
ap-8720	138	10	models	model	NOUN
ap-8720	138	11	to	to	PART
ap-8720	138	12	improve	improve	VERB
ap-8720	138	13	their	their	PRON
ap-8720	138	14	ability	ability	NOUN
ap-8720	138	15	to	to	PART
ap-8720	138	16	detect	detect	VERB
ap-8720	138	17	image	image	NOUN
ap-8720	138	18	manipulation	manipulation	NOUN
ap-8720	138	19	.	.	PUNCT
ap-8720	139	1	by	by	ADP
ap-8720	139	2	applying	apply	VERB
ap-8720	139	3	some	some	DET
ap-8720	139	4	post	post	ADJ
ap-8720	139	5	-	-	ADJ
ap-8720	139	6	processing	processing	ADJ
ap-8720	139	7	operations	operation	NOUN
ap-8720	139	8	,	,	PUNCT
ap-8720	139	9	(	(	PUNCT
ap-8720	139	10	including	include	VERB
ap-8720	139	11	lossy	lossy	ADJ
ap-8720	139	12	compression	compression	NOUN
ap-8720	139	13	)	)	PUNCT
ap-8720	139	14	,	,	PUNCT
ap-8720	139	15	the	the	DET
ap-8720	139	16	authors	author	NOUN
ap-8720	139	17	simulated	simulate	VERB
ap-8720	139	18	a	a	DET
ap-8720	139	19	real	real	ADJ
ap-8720	139	20	analysis	analysis	NOUN
ap-8720	139	21	of	of	ADP
ap-8720	139	22	face	face	NOUN
ap-8720	139	23	images	image	NOUN
ap-8720	139	24	in	in	ADP
ap-8720	139	25	various	various	ADJ
ap-8720	139	26	complex	complex	ADJ
ap-8720	139	27	scenarios	scenario	NOUN
ap-8720	139	28	.	.	PUNCT
ap-8720	140	1	to	to	PART
ap-8720	140	2	demonstrate	demonstrate	VERB
ap-8720	140	3	the	the	DET
ap-8720	140	4	effectiveness	effectiveness	NOUN
ap-8720	140	5	of	of	ADP
ap-8720	140	6	the	the	DET
ap-8720	140	7	amtennet	amtennet	NOUN
ap-8720	140	8	project	project	NOUN
ap-8720	140	9	,	,	PUNCT
ap-8720	140	10	a	a	DET
ap-8720	140	11	series	series	NOUN
ap-8720	140	12	of	of	ADP
ap-8720	140	13	experiments	experiment	NOUN
ap-8720	140	14	were	be	AUX
ap-8720	140	15	conducted	conduct	VERB
ap-8720	140	16	to	to	PART
ap-8720	140	17	achieve	achieve	VERB
ap-8720	140	18	a	a	DET
ap-8720	140	19	higher	high	ADJ
ap-8720	140	20	detection	detection	NOUN
ap-8720	140	21	rate	rate	NOUN
ap-8720	140	22	.	.	PUNCT
ap-8720	141	1	yang	yang	PROPN
ap-8720	141	2	et	et	PROPN
ap-8720	141	3	al	al	PROPN
ap-8720	141	4	.	.	PUNCT
ap-8720	142	1	[	[	X
ap-8720	142	2	27	27	NUM
ap-8720	142	3	]	]	PUNCT
ap-8720	142	4	used	use	VERB
ap-8720	142	5	a	a	DET
ap-8720	142	6	large	large	ADJ
ap-8720	142	7	hff	hff	NOUN
ap-8720	142	8	database	database	NOUN
ap-8720	142	9	of	of	ADP
ap-8720	142	10	116	116	NUM
ap-8720	142	11	000	000	NUM
ap-8720	142	12	face	face	NOUN
ap-8720	142	13	images	image	NOUN
ap-8720	142	14	for	for	ADP
ap-8720	142	15	training	training	NOUN
ap-8720	142	16	including	include	VERB
ap-8720	142	17	real	real	ADJ
ap-8720	142	18	images	image	NOUN
ap-8720	142	19	of	of	ADP
ap-8720	142	20	different	different	ADJ
ap-8720	142	21	resolutions	resolution	NOUN
ap-8720	142	22	and	and	CCONJ
ap-8720	142	23	five	five	NUM
ap-8720	142	24	types	type	NOUN
ap-8720	142	25	of	of	ADP
ap-8720	142	26	false	false	ADJ
ap-8720	142	27	images	image	NOUN
ap-8720	142	28	.	.	PUNCT
ap-8720	143	1	the	the	DET
ap-8720	143	2	dataset	dataset	NOUN
ap-8720	143	3	was	be	AUX
ap-8720	143	4	divided	divide	VERB
ap-8720	143	5	into	into	ADP
ap-8720	143	6	three	three	NUM
ap-8720	143	7	subsets	subset	NOUN
ap-8720	143	8	:	:	PUNCT
ap-8720	143	9	75	75	NUM
ap-8720	143	10	%	%	NOUN
ap-8720	143	11	for	for	ADP
ap-8720	143	12	training	training	NOUN
ap-8720	143	13	,	,	PUNCT
ap-8720	143	14	5	5	NUM
ap-8720	143	15	%	%	NOUN
ap-8720	143	16	for	for	ADP
ap-8720	143	17	validation	validation	NOUN
ap-8720	143	18	,	,	PUNCT
ap-8720	143	19	and	and	CCONJ
ap-8720	143	20	20	20	NUM
ap-8720	143	21	%	%	NOUN
ap-8720	143	22	for	for	ADP
ap-8720	143	23	testing	testing	NOUN
ap-8720	143	24	.	.	PUNCT
ap-8720	144	1	as	as	ADP
ap-8720	144	2	a	a	DET
ap-8720	144	3	result	result	NOUN
ap-8720	144	4	,	,	PUNCT
ap-8720	144	5	amtennet	amtennet	NOUN
ap-8720	144	6	achieved	achieve	VERB
ap-8720	144	7	an	an	DET
ap-8720	144	8	accuracy	accuracy	NOUN
ap-8720	144	9	of	of	ADP
ap-8720	144	10	95.17	95.17	NUM
ap-8720	144	11	%	%	NOUN
ap-8720	144	12	.	.	PUNCT
ap-8720	145	1	in	in	ADP
ap-8720	145	2	their	their	PRON
ap-8720	145	3	work	work	NOUN
ap-8720	145	4	,	,	PUNCT
ap-8720	145	5	chintha	chintha	PROPN
ap-8720	145	6	et	et	PROPN
ap-8720	145	7	al	al	PROPN
ap-8720	145	8	.	.	PUNCT
ap-8720	146	1	[	[	X
ap-8720	146	2	28	28	NUM
ap-8720	146	3	]	]	PUNCT
ap-8720	146	4	proposed	propose	VERB
ap-8720	146	5	an	an	DET
ap-8720	146	6	xceptemporal	xceptemporal	ADJ
ap-8720	146	7	convolutional	convolutional	ADJ
ap-8720	146	8	recurrent	recurrent	ADJ
ap-8720	146	9	neural	neural	ADJ
ap-8720	146	10	network	network	NOUN
ap-8720	146	11	for	for	ADP
ap-8720	146	12	deepfake	deepfake	NOUN
ap-8720	146	13	detection	detection	NOUN
ap-8720	146	14	.	.	PUNCT
ap-8720	147	1	they	they	PRON
ap-8720	147	2	used	use	VERB
ap-8720	147	3	a	a	DET
ap-8720	147	4	face	face	NOUN
ap-8720	147	5	detector	detector	NOUN
ap-8720	147	6	called	call	VERB
ap-8720	147	7	dlib	dlib	NOUN
ap-8720	147	8	to	to	PART
ap-8720	147	9	determine	determine	VERB
ap-8720	147	10	the	the	DET
ap-8720	147	11	main	main	ADJ
ap-8720	147	12	face	face	NOUN
ap-8720	147	13	in	in	ADP
ap-8720	147	14	each	each	DET
ap-8720	147	15	frame	frame	NOUN
ap-8720	147	16	of	of	ADP
ap-8720	147	17	the	the	DET
ap-8720	147	18	video	video	NOUN
ap-8720	147	19	.	.	PUNCT
ap-8720	148	1	the	the	DET
ap-8720	148	2	faces	face	NOUN
ap-8720	148	3	were	be	AUX
ap-8720	148	4	encoded	encode	VERB
ap-8720	148	5	using	use	VERB
ap-8720	148	6	the	the	DET
ap-8720	148	7	xceptionnet	xceptionnet	NOUN
ap-8720	148	8	architecture	architecture	NOUN
ap-8720	148	9	[	[	X
ap-8720	148	10	29	29	NUM
ap-8720	148	11	]	]	PUNCT
ap-8720	148	12	.	.	PUNCT
ap-8720	149	1	the	the	DET
ap-8720	149	2	spatial	spatial	ADJ
ap-8720	149	3	features	feature	NOUN
ap-8720	149	4	extracted	extract	VERB
ap-8720	149	5	from	from	ADP
ap-8720	149	6	xceptionnet	xceptionnet	PROPN
ap-8720	149	7	were	be	AUX
ap-8720	149	8	fed	feed	VERB
ap-8720	149	9	into	into	ADP
ap-8720	149	10	a	a	DET
ap-8720	149	11	first	first	ADJ
ap-8720	149	12	bidirectional	bidirectional	ADJ
ap-8720	149	13	lstm	lstm	NOUN
ap-8720	149	14	layer	layer	NOUN
ap-8720	149	15	and	and	CCONJ
ap-8720	149	16	the	the	DET
ap-8720	149	17	outputs	output	NOUN
ap-8720	149	18	from	from	ADP
ap-8720	149	19	this	this	DET
ap-8720	149	20	layer	layer	NOUN
ap-8720	149	21	were	be	AUX
ap-8720	149	22	fed	feed	VERB
ap-8720	149	23	to	to	ADP
ap-8720	149	24	a	a	DET
ap-8720	149	25	second	second	ADJ
ap-8720	149	26	bidirectional	bidirectional	ADJ
ap-8720	149	27	lstm	lstm	NOUN
ap-8720	149	28	layer	layer	NOUN
ap-8720	149	29	to	to	PART
ap-8720	149	30	produce	produce	VERB
ap-8720	149	31	secondary	secondary	ADJ
ap-8720	149	32	features	feature	NOUN
ap-8720	149	33	.	.	PUNCT
ap-8720	150	1	the	the	DET
ap-8720	150	2	feature	feature	NOUN
ap-8720	150	3	vector	vector	NOUN
ap-8720	150	4	obtained	obtain	VERB
ap-8720	150	5	from	from	ADP
ap-8720	150	6	the	the	DET
ap-8720	150	7	last	last	ADJ
ap-8720	150	8	lstm	lstm	ADJ
ap-8720	150	9	unit	unit	NOUN
ap-8720	150	10	in	in	ADP
ap-8720	150	11	the	the	DET
ap-8720	150	12	second	second	ADJ
ap-8720	150	13	bidirectional	bidirectional	ADJ
ap-8720	150	14	layer	layer	NOUN
ap-8720	150	15	was	be	AUX
ap-8720	150	16	then	then	ADV
ap-8720	150	17	fed	feed	VERB
ap-8720	150	18	through	through	ADP
ap-8720	150	19	a	a	DET
ap-8720	150	20	fully	fully	ADV
ap-8720	150	21	connected	connect	VERB
ap-8720	150	22	layer	layer	NOUN
ap-8720	150	23	and	and	CCONJ
ap-8720	150	24	a	a	DET
ap-8720	150	25	classification	classification	NOUN
ap-8720	150	26	layer	layer	NOUN
ap-8720	150	27	.	.	PUNCT
ap-8720	151	1	a	a	DET
ap-8720	151	2	dropout	dropout	NOUN
ap-8720	151	3	was	be	AUX
ap-8720	151	4	added	add	VERB
ap-8720	151	5	to	to	ADP
ap-8720	151	6	the	the	DET
ap-8720	151	7	fully	fully	ADV
ap-8720	151	8	connected	connected	ADJ
ap-8720	151	9	layer	layer	NOUN
ap-8720	151	10	to	to	PART
ap-8720	151	11	improve	improve	VERB
ap-8720	151	12	regularisation	regularisation	NOUN
ap-8720	151	13	.	.	PUNCT
ap-8720	152	1	the	the	DET
ap-8720	152	2	model	model	NOUN
ap-8720	152	3	was	be	AUX
ap-8720	152	4	trained	train	VERB
ap-8720	152	5	using	use	VERB
ap-8720	152	6	the	the	DET
ap-8720	152	7	traditional	traditional	ADJ
ap-8720	152	8	cross	cross	NOUN
ap-8720	152	9	-	-	NOUN
ap-8720	152	10	entropy	entropy	NOUN
ap-8720	152	11	and	and	CCONJ
ap-8720	152	12	kl	kl	PROPN
ap-8720	152	13	divergence	divergence	NOUN
ap-8720	152	14	loss	loss	NOUN
ap-8720	152	15	functions	function	NOUN
ap-8720	152	16	.	.	PUNCT
ap-8720	153	1	in	in	ADP
ap-8720	153	2	addition	addition	NOUN
ap-8720	153	3	,	,	PUNCT
ap-8720	153	4	the	the	DET
ap-8720	153	5	researchers	researcher	NOUN
ap-8720	153	6	introduced	introduce	VERB
ap-8720	153	7	a	a	DET
ap-8720	153	8	complementary	complementary	ADJ
ap-8720	153	9	architecture	architecture	NOUN
ap-8720	153	10	for	for	ADP
ap-8720	153	11	audio	audio	ADJ
ap-8720	153	12	analysis	analysis	NOUN
ap-8720	153	13	by	by	ADP
ap-8720	153	14	using	use	VERB
ap-8720	153	15	multiple	multiple	ADJ
ap-8720	153	16	convolution	convolution	NOUN
ap-8720	153	17	modules	module	NOUN
ap-8720	153	18	to	to	PART
ap-8720	153	19	obtain	obtain	VERB
ap-8720	153	20	audio	audio	ADJ
ap-8720	153	21	feature	feature	NOUN
ap-8720	153	22	representations	representation	NOUN
ap-8720	153	23	.	.	PUNCT
ap-8720	154	1	these	these	DET
ap-8720	154	2	audio	audio	ADJ
ap-8720	154	3	embeddings	embedding	NOUN
ap-8720	154	4	were	be	AUX
ap-8720	154	5	also	also	ADV
ap-8720	154	6	fed	feed	VERB
ap-8720	154	7	through	through	ADP
ap-8720	154	8	a	a	DET
ap-8720	154	9	bidirectional	bidirectional	ADJ
ap-8720	154	10	recursive	recursive	ADJ
ap-8720	154	11	layer	layer	NOUN
ap-8720	154	12	.	.	PUNCT
ap-8720	155	1	training	train	VERB
ap-8720	155	2	the	the	DET
ap-8720	155	3	model	model	NOUN
ap-8720	155	4	on	on	ADP
ap-8720	155	5	the	the	DET
ap-8720	155	6	faceforensics++	faceforensics++	PROPN
ap-8720	155	7	[	[	X
ap-8720	155	8	9	9	NUM
ap-8720	155	9	]	]	PUNCT
ap-8720	155	10	and	and	CCONJ
ap-8720	155	11	celeb	celeb	PROPN
ap-8720	155	12	-	-	PUNCT
ap-8720	155	13	df	df	PROPN
ap-8720	156	1	[	[	X
ap-8720	156	2	10	10	NUM
ap-8720	156	3	]	]	PUNCT
ap-8720	156	4	databases	database	NOUN
ap-8720	156	5	resulted	result	VERB
ap-8720	156	6	in	in	ADP
ap-8720	156	7	visual	visual	ADJ
ap-8720	156	8	detection	detection	NOUN
ap-8720	156	9	accuracies	accuracy	NOUN
ap-8720	156	10	ranging	range	VERB
ap-8720	156	11	from	from	ADP
ap-8720	156	12	84.8	84.8	NUM
ap-8720	156	13	%	%	NOUN
ap-8720	156	14	to	to	ADP
ap-8720	156	15	100	100	NUM
ap-8720	156	16	%	%	NOUN
ap-8720	156	17	.	.	PUNCT
ap-8720	157	1	for	for	ADP
ap-8720	157	2	audio	audio	ADJ
ap-8720	157	3	detection	detection	NOUN
ap-8720	157	4	,	,	PUNCT
ap-8720	157	5	the	the	DET
ap-8720	157	6	authors	author	NOUN
ap-8720	157	7	demonstrated	demonstrate	VERB
ap-8720	157	8	the	the	DET
ap-8720	157	9	robustness	robustness	NOUN
ap-8720	157	10	of	of	ADP
ap-8720	157	11	their	their	PRON
ap-8720	157	12	method	method	NOUN
ap-8720	157	13	,	,	PUNCT
ap-8720	157	14	achieving	achieve	VERB
ap-8720	157	15	a	a	DET
ap-8720	157	16	t	t	PROPN
ap-8720	157	17	-	-	PUNCT
ap-8720	157	18	dcf	dcf	PROPN
ap-8720	157	19	(	(	PUNCT
ap-8720	157	20	tandem	tandem	NOUN
ap-8720	157	21	detection	detection	NOUN
ap-8720	157	22	cost	cost	NOUN
ap-8720	157	23	function	function	NOUN
ap-8720	157	24	)	)	PUNCT
ap-8720	157	25	of	of	ADP
ap-8720	157	26	0.1424	0.1424	NUM
ap-8720	157	27	.	.	PUNCT
ap-8720	158	1	this	this	PRON
ap-8720	158	2	indicates	indicate	VERB
ap-8720	158	3	the	the	DET
ap-8720	158	4	generalisability	generalisability	NOUN
ap-8720	158	5	of	of	ADP
ap-8720	158	6	their	their	PRON
ap-8720	158	7	approach	approach	NOUN
ap-8720	158	8	to	to	ADP
ap-8720	158	9	different	different	ADJ
ap-8720	158	10	types	type	NOUN
ap-8720	158	11	of	of	ADP
ap-8720	158	12	attacks	attack	NOUN
ap-8720	158	13	.	.	PUNCT
ap-8720	159	1	lee	lee	PROPN
ap-8720	159	2	et	et	PROPN
ap-8720	159	3	al	al	PROPN
ap-8720	159	4	.	.	PUNCT
ap-8720	160	1	[	[	X
ap-8720	160	2	30	30	NUM
ap-8720	160	3	]	]	PUNCT
ap-8720	160	4	aimed	aim	VERB
ap-8720	160	5	to	to	PART
ap-8720	160	6	detect	detect	VERB
ap-8720	160	7	a	a	DET
ap-8720	160	8	new	new	ADJ
ap-8720	160	9	manually	manually	ADV
ap-8720	160	10	produced	produce	VERB
ap-8720	160	11	facial	facial	ADJ
ap-8720	160	12	manipulation	manipulation	NOUN
ap-8720	160	13	by	by	ADP
ap-8720	160	14	introducing	introduce	VERB
ap-8720	160	15	a	a	DET
ap-8720	160	16	new	new	ADJ
ap-8720	160	17	dataset	dataset	NOUN
ap-8720	160	18	containing	contain	VERB
ap-8720	160	19	photoshopped	photoshoppe	VERB
ap-8720	160	20	face	face	NOUN
ap-8720	160	21	images	image	NOUN
ap-8720	160	22	created	create	VERB
ap-8720	160	23	with	with	ADP
ap-8720	160	24	tools	tool	NOUN
ap-8720	160	25	such	such	ADJ
ap-8720	160	26	as	as	ADP
ap-8720	160	27	adobe	adobe	PROPN
ap-8720	160	28	photoshop	photoshop	PROPN
ap-8720	160	29	to	to	PART
ap-8720	160	30	detect	detect	VERB
ap-8720	160	31	both	both	DET
ap-8720	160	32	human	human	NOUN
ap-8720	160	33	-	-	PUNCT
ap-8720	160	34	manipulated	manipulate	VERB
ap-8720	160	35	and	and	CCONJ
ap-8720	160	36	machine	machine	NOUN
ap-8720	160	37	-	-	PUNCT
ap-8720	160	38	generated	generate	VERB
ap-8720	160	39	images	image	NOUN
ap-8720	160	40	.	.	PUNCT
ap-8720	161	1	this	this	DET
ap-8720	161	2	dataset	dataset	NOUN
ap-8720	161	3	contains	contain	VERB
ap-8720	161	4	1527	1527	NUM
ap-8720	161	5	images	image	NOUN
ap-8720	161	6	developed	develop	VERB
ap-8720	161	7	with	with	ADP
ap-8720	161	8	several	several	ADJ
ap-8720	161	9	levels	level	NOUN
ap-8720	161	10	of	of	ADP
ap-8720	161	11	editing	edit	VERB
ap-8720	161	12	complexity	complexity	NOUN
ap-8720	161	13	and	and	CCONJ
ap-8720	161	14	621	621	NUM
ap-8720	161	15	original	original	ADJ
ap-8720	161	16	images	image	NOUN
ap-8720	161	17	used	use	VERB
ap-8720	161	18	to	to	PART
ap-8720	161	19	create	create	VERB
ap-8720	161	20	the	the	DET
ap-8720	161	21	fake	fake	ADJ
ap-8720	161	22	faces	face	NOUN
ap-8720	161	23	.	.	PUNCT
ap-8720	162	1	to	to	PART
ap-8720	162	2	address	address	VERB
ap-8720	162	3	this	this	DET
ap-8720	162	4	challenge	challenge	NOUN
ap-8720	162	5	,	,	PUNCT
ap-8720	162	6	the	the	DET
ap-8720	162	7	researchers	researcher	NOUN
ap-8720	162	8	developed	develop	VERB
ap-8720	162	9	a	a	DET
ap-8720	162	10	cnn	cnn	PROPN
ap-8720	162	11	-	-	PUNCT
ap-8720	162	12	based	base	VERB
ap-8720	162	13	model	model	NOUN
ap-8720	162	14	called	call	VERB
ap-8720	162	15	shallow	shallow	NOUN
ap-8720	162	16	-	-	PUNCT
ap-8720	162	17	fakeface	fakeface	NOUN
ap-8720	162	18	(	(	PUNCT
ap-8720	162	19	sffn	sffn	NOUN
ap-8720	162	20	)	)	PUNCT
ap-8720	162	21	.	.	PUNCT
ap-8720	163	1	this	this	DET
ap-8720	163	2	model	model	NOUN
ap-8720	163	3	shows	show	VERB
ap-8720	163	4	promising	promise	VERB
ap-8720	163	5	results	result	NOUN
ap-8720	163	6	in	in	ADP
ap-8720	163	7	detecting	detect	VERB
ap-8720	163	8	human	human	NOUN
ap-8720	163	9	-	-	PUNCT
ap-8720	163	10	created	create	VERB
ap-8720	163	11	fake	fake	ADJ
ap-8720	163	12	images	image	NOUN
ap-8720	163	13	,	,	PUNCT
ap-8720	163	14	achieving	achieve	VERB
ap-8720	163	15	a	a	DET
ap-8720	163	16	72.52	72.52	NUM
ap-8720	163	17	%	%	NOUN
ap-8720	163	18	ocr	ocr	NOUN
ap-8720	163	19	using	use	VERB
ap-8720	163	20	less	less	ADJ
ap-8720	163	21	than	than	ADP
ap-8720	163	22	2500	2500	NUM
ap-8720	163	23	fake	fake	ADJ
ap-8720	163	24	images	image	NOUN
ap-8720	163	25	for	for	ADP
ap-8720	163	26	training	training	NOUN
ap-8720	163	27	.	.	PUNCT
ap-8720	164	1	the	the	DET
ap-8720	164	2	effectiveness	effectiveness	NOUN
ap-8720	164	3	of	of	ADP
ap-8720	164	4	the	the	DET
ap-8720	164	5	sffn	sffn	ADJ
ap-8720	164	6	model	model	NOUN
ap-8720	164	7	was	be	AUX
ap-8720	164	8	alos	alo	NOUN
ap-8720	164	9	evaluated	evaluate	VERB
ap-8720	164	10	on	on	ADP
ap-8720	164	11	other	other	ADJ
ap-8720	164	12	gangenerated	gangenerate	VERB
ap-8720	164	13	fake	fake	ADJ
ap-8720	164	14	faces	face	NOUN
ap-8720	164	15	,	,	PUNCT
ap-8720	164	16	demonstrating	demonstrate	VERB
ap-8720	164	17	a	a	DET
ap-8720	164	18	recognition	recognition	NOUN
ap-8720	164	19	rate	rate	NOUN
ap-8720	164	20	of	of	ADP
ap-8720	164	21	93.99	93.99	NUM
ap-8720	164	22	%	%	NOUN
ap-8720	164	23	for	for	ADP
ap-8720	164	24	low	low	ADJ
ap-8720	164	25	-	-	PUNCT
ap-8720	164	26	resolution	resolution	NOUN
ap-8720	164	27	images	image	NOUN
ap-8720	164	28	.	.	PUNCT
ap-8720	165	1	these	these	DET
ap-8720	165	2	results	result	NOUN
ap-8720	165	3	highlight	highlight	VERB
ap-8720	165	4	the	the	DET
ap-8720	165	5	model	model	NOUN
ap-8720	165	6	’s	’s	PART
ap-8720	165	7	ability	ability	NOUN
ap-8720	165	8	to	to	PART
ap-8720	165	9	generalise	generalise	VERB
ap-8720	165	10	well	well	ADV
ap-8720	165	11	against	against	ADP
ap-8720	165	12	potential	potential	ADJ
ap-8720	165	13	attackers	attacker	NOUN
ap-8720	165	14	.	.	PUNCT
ap-8720	166	1	in	in	ADP
ap-8720	166	2	this	this	DET
ap-8720	166	3	study	study	NOUN
ap-8720	166	4	,	,	PUNCT
ap-8720	166	5	we	we	PRON
ap-8720	166	6	propose	propose	VERB
ap-8720	166	7	a	a	DET
ap-8720	166	8	new	new	ADJ
ap-8720	166	9	multi	multi	ADJ
ap-8720	166	10	-	-	ADJ
ap-8720	166	11	discriminator	discriminator	ADJ
ap-8720	166	12	architecture	architecture	NOUN
ap-8720	166	13	for	for	ADP
ap-8720	166	14	distinguishing	distinguish	VERB
ap-8720	166	15	fake	fake	ADJ
ap-8720	166	16	faces	face	NOUN
ap-8720	166	17	from	from	ADP
ap-8720	166	18	real	real	ADJ
ap-8720	166	19	ones	one	NOUN
ap-8720	166	20	.	.	PUNCT
ap-8720	167	1	the	the	DET
ap-8720	167	2	proposed	propose	VERB
ap-8720	167	3	architecture	architecture	NOUN
ap-8720	167	4	is	be	AUX
ap-8720	167	5	a	a	DET
ap-8720	167	6	competition	competition	NOUN
ap-8720	167	7	between	between	ADP
ap-8720	167	8	three	three	NUM
ap-8720	167	9	deep	deep	ADJ
ap-8720	167	10	networks	network	NOUN
ap-8720	167	11	(	(	PUNCT
ap-8720	167	12	discriminators	discriminator	NOUN
ap-8720	167	13	)	)	PUNCT
ap-8720	167	14	,	,	PUNCT
ap-8720	167	15	each	each	PRON
ap-8720	167	16	having	having	AUX
ap-8720	167	17	been	be	AUX
ap-8720	167	18	trained	train	VERB
ap-8720	167	19	differently	differently	ADV
ap-8720	167	20	.	.	PUNCT
ap-8720	168	1	the	the	DET
ap-8720	168	2	final	final	ADJ
ap-8720	168	3	decision	decision	NOUN
ap-8720	168	4	is	be	AUX
ap-8720	168	5	made	make	VERB
ap-8720	168	6	by	by	ADP
ap-8720	168	7	voting	vote	VERB
ap-8720	168	8	on	on	ADP
ap-8720	168	9	the	the	DET
ap-8720	168	10	decisions	decision	NOUN
ap-8720	168	11	of	of	ADP
ap-8720	168	12	the	the	DET
ap-8720	168	13	three	three	NUM
ap-8720	168	14	discriminators	discriminator	NOUN
ap-8720	168	15	.	.	PUNCT
ap-8720	169	1	in	in	ADP
ap-8720	169	2	summary	summary	NOUN
ap-8720	169	3	,	,	PUNCT
ap-8720	169	4	the	the	DET
ap-8720	169	5	paper	paper	NOUN
ap-8720	169	6	presents	present	VERB
ap-8720	169	7	the	the	DET
ap-8720	169	8	following	follow	VERB
ap-8720	169	9	contributions	contribution	NOUN
ap-8720	169	10	:	:	PUNCT
ap-8720	169	11	•	•	NUM
ap-8720	169	12	proposes	propose	VERB
ap-8720	169	13	a	a	DET
ap-8720	169	14	new	new	ADJ
ap-8720	169	15	architecture	architecture	NOUN
ap-8720	169	16	of	of	ADP
ap-8720	169	17	dcgan	dcgan	NOUN
ap-8720	169	18	called	call	VERB
ap-8720	169	19	ndcgan	ndcgan	PROPN
ap-8720	169	20	where	where	SCONJ
ap-8720	169	21	the	the	DET
ap-8720	169	22	discriminator	discriminator	NOUN
ap-8720	169	23	1	1	NUM
ap-8720	169	24	trains	train	NOUN
ap-8720	169	25	with	with	ADP
ap-8720	169	26	its	its	PRON
ap-8720	169	27	opponent	opponent	NOUN
ap-8720	169	28	(	(	PUNCT
ap-8720	169	29	the	the	DET
ap-8720	169	30	generator	generator	NOUN
ap-8720	169	31	)	)	PUNCT
ap-8720	169	32	to	to	PART
ap-8720	169	33	get	get	VERB
ap-8720	169	34	an	an	DET
ap-8720	169	35	advantage	advantage	NOUN
ap-8720	169	36	and	and	CCONJ
ap-8720	169	37	recognise	recognise	VERB
ap-8720	169	38	the	the	DET
ap-8720	169	39	fake	fake	ADJ
ap-8720	169	40	faces	face	NOUN
ap-8720	169	41	generated	generate	VERB
ap-8720	169	42	by	by	ADP
ap-8720	169	43	the	the	DET
ap-8720	169	44	latter	latter	ADJ
ap-8720	169	45	with	with	ADP
ap-8720	169	46	a	a	DET
ap-8720	169	47	considerable	considerable	ADJ
ap-8720	169	48	rate	rate	NOUN
ap-8720	169	49	.	.	PUNCT
ap-8720	170	1	•	•	NOUN
ap-8720	170	2	presents	present	VERB
ap-8720	170	3	a	a	DET
ap-8720	170	4	multi	multi	ADJ
ap-8720	170	5	discriminator	discriminator	NOUN
ap-8720	170	6	deep	deep	PROPN
ap-8720	170	7	cnn	cnn	PROPN
ap-8720	170	8	(	(	PUNCT
ap-8720	170	9	mddcnn	mddcnn	PROPN
ap-8720	170	10	)	)	PUNCT
ap-8720	170	11	architecture	architecture	NOUN
ap-8720	170	12	based	base	VERB
ap-8720	170	13	on	on	ADP
ap-8720	170	14	three	three	NUM
ap-8720	170	15	competing	compete	VERB
ap-8720	170	16	discriminators	discriminator	NOUN
ap-8720	170	17	.	.	PUNCT
ap-8720	171	1	the	the	DET
ap-8720	171	2	main	main	ADJ
ap-8720	171	3	advantage	advantage	NOUN
ap-8720	171	4	is	be	AUX
ap-8720	171	5	that	that	SCONJ
ap-8720	171	6	the	the	DET
ap-8720	171	7	three	three	NUM
ap-8720	171	8	networks	network	NOUN
ap-8720	171	9	are	be	AUX
ap-8720	171	10	involved	involve	VERB
ap-8720	171	11	in	in	ADP
ap-8720	171	12	the	the	DET
ap-8720	171	13	final	final	ADJ
ap-8720	171	14	decision	decision	NOUN
ap-8720	171	15	by	by	ADP
ap-8720	171	16	voting	vote	VERB
ap-8720	171	17	on	on	ADP
ap-8720	171	18	the	the	DET
ap-8720	171	19	predictions	prediction	NOUN
ap-8720	171	20	made	make	VERB
ap-8720	171	21	by	by	ADP
ap-8720	171	22	each	each	PRON
ap-8720	171	23	.	.	PUNCT
ap-8720	172	1	this	this	DET
ap-8720	172	2	significantly	significantly	ADV
ap-8720	172	3	minimises	minimise	NOUN
ap-8720	172	4	prediction	prediction	NOUN
ap-8720	172	5	errors	error	NOUN
ap-8720	172	6	.	.	PUNCT
ap-8720	173	1	3	3	X
ap-8720	173	2	.	.	NUM
ap-8720	173	3	proposed	propose	VERB
ap-8720	173	4	system	system	NOUN
ap-8720	173	5	architecture	architecture	NOUN
ap-8720	173	6	the	the	DET
ap-8720	173	7	multi	multi	ADJ
ap-8720	173	8	-	-	ADJ
ap-8720	173	9	discriminator	discriminator	ADJ
ap-8720	173	10	architecture	architecture	NOUN
ap-8720	173	11	proposed	propose	VERB
ap-8720	173	12	in	in	ADP
ap-8720	173	13	this	this	DET
ap-8720	173	14	work	work	NOUN
ap-8720	173	15	is	be	AUX
ap-8720	173	16	based	base	VERB
ap-8720	173	17	on	on	ADP
ap-8720	173	18	the	the	DET
ap-8720	173	19	deep	deep	ADJ
ap-8720	173	20	convolutional	convolutional	ADJ
ap-8720	173	21	generative	generative	ADJ
ap-8720	173	22	adversarial	adversarial	ADJ
ap-8720	173	23	network	network	NOUN
ap-8720	173	24	(	(	PUNCT
ap-8720	173	25	dcgan	dcgan	NOUN
ap-8720	173	26	)	)	PUNCT
ap-8720	173	27	,	,	PUNCT
ap-8720	173	28	which	which	PRON
ap-8720	173	29	is	be	AUX
ap-8720	173	30	a	a	DET
ap-8720	173	31	direct	direct	ADJ
ap-8720	173	32	extension	extension	NOUN
ap-8720	173	33	of	of	ADP
ap-8720	173	34	the	the	DET
ap-8720	173	35	gan	gan	PROPN
ap-8720	173	36	.	.	PUNCT
ap-8720	174	1	using	use	VERB
ap-8720	174	2	the	the	DET
ap-8720	174	3	same	same	ADJ
ap-8720	174	4	distribution	distribution	NOUN
ap-8720	174	5	of	of	ADP
ap-8720	174	6	training	training	NOUN
ap-8720	174	7	data	datum	NOUN
ap-8720	174	8	,	,	PUNCT
ap-8720	174	9	gans	gan	NOUN
ap-8720	174	10	are	be	AUX
ap-8720	174	11	used	use	VERB
ap-8720	174	12	to	to	PART
ap-8720	174	13	train	train	VERB
ap-8720	174	14	deep	deep	ADJ
ap-8720	174	15	learning	learning	NOUN
ap-8720	174	16	models	model	NOUN
ap-8720	174	17	to	to	PART
ap-8720	174	18	produce	produce	VERB
ap-8720	174	19	new	new	ADJ
ap-8720	174	20	data	datum	NOUN
ap-8720	174	21	.	.	PUNCT
ap-8720	175	1	created	create	VERB
ap-8720	175	2	in	in	ADP
ap-8720	175	3	2014	2014	NUM
ap-8720	175	4	by	by	ADP
ap-8720	175	5	ian	ian	PROPN
ap-8720	175	6	goodfellow	goodfellow	PROPN
ap-8720	176	1	[	[	X
ap-8720	176	2	31	31	NUM
ap-8720	176	3	]	]	PUNCT
ap-8720	176	4	and	and	CCONJ
ap-8720	176	5	published	publish	VERB
ap-8720	176	6	in	in	ADP
ap-8720	176	7	the	the	DET
ap-8720	176	8	paper	paper	NOUN
ap-8720	176	9	generative	generative	ADJ
ap-8720	176	10	adversarial	adversarial	ADJ
ap-8720	176	11	nets	net	NOUN
ap-8720	176	12	.	.	PUNCT
ap-8720	177	1	these	these	DET
ap-8720	177	2	networks	network	NOUN
ap-8720	177	3	consist	consist	VERB
ap-8720	177	4	of	of	ADP
ap-8720	177	5	two	two	NUM
ap-8720	177	6	different	different	ADJ
ap-8720	177	7	models	model	NOUN
ap-8720	177	8	:	:	PUNCT
ap-8720	177	9	the	the	DET
ap-8720	177	10	generator	generator	NOUN
ap-8720	177	11	and	and	CCONJ
ap-8720	177	12	the	the	DET
ap-8720	177	13	discriminator	discriminator	NOUN
ap-8720	177	14	.	.	PUNCT
ap-8720	178	1	the	the	DET
ap-8720	178	2	generator	generator	NOUN
ap-8720	178	3	creates	create	VERB
ap-8720	178	4	artificial	artificial	ADJ
ap-8720	178	5	or	or	CCONJ
ap-8720	178	6	synthetic	synthetic	ADJ
ap-8720	178	7	visuals	visual	NOUN
ap-8720	178	8	that	that	PRON
ap-8720	178	9	resemble	resemble	VERB
ap-8720	178	10	training	training	NOUN
ap-8720	178	11	images	image	NOUN
ap-8720	178	12	.	.	PUNCT
ap-8720	179	1	the	the	DET
ap-8720	179	2	discriminator	discriminator	NOUN
ap-8720	179	3	examines	examine	VERB
ap-8720	179	4	an	an	DET
ap-8720	179	5	image	image	NOUN
ap-8720	179	6	and	and	CCONJ
ap-8720	179	7	the	the	DET
ap-8720	179	8	output	output	NOUN
ap-8720	179	9	to	to	PART
ap-8720	179	10	determine	determine	VERB
ap-8720	179	11	whether	whether	SCONJ
ap-8720	179	12	it	it	PRON
ap-8720	179	13	is	be	AUX
ap-8720	179	14	authentic	authentic	ADJ
ap-8720	179	15	or	or	CCONJ
ap-8720	179	16	fraudulent	fraudulent	ADJ
ap-8720	179	17	.	.	PUNCT
ap-8720	180	1	during	during	ADP
ap-8720	180	2	the	the	DET
ap-8720	180	3	training	training	NOUN
ap-8720	180	4	process	process	NOUN
ap-8720	180	5	,	,	PUNCT
ap-8720	180	6	the	the	DET
ap-8720	180	7	discriminator	discriminator	NOUN
ap-8720	180	8	strives	strive	VERB
ap-8720	180	9	to	to	PART
ap-8720	180	10	get	get	VERB
ap-8720	180	11	better	well	ADJ
ap-8720	180	12	at	at	ADP
ap-8720	180	13	detecting	detect	VERB
ap-8720	180	14	and	and	CCONJ
ap-8720	180	15	categorising	categorise	VERB
ap-8720	180	16	whether	whether	SCONJ
ap-8720	180	17	an	an	DET
ap-8720	180	18	image	image	NOUN
ap-8720	180	19	is	be	AUX
ap-8720	180	20	real	real	ADJ
ap-8720	180	21	or	or	CCONJ
ap-8720	180	22	fake	fake	ADJ
ap-8720	180	23	by	by	ADP
ap-8720	180	24	learning	learn	VERB
ap-8720	180	25	to	to	PART
ap-8720	180	26	create	create	VERB
ap-8720	180	27	better	well	ADV
ap-8720	180	28	fake	fake	ADJ
ap-8720	180	29	images	image	NOUN
ap-8720	180	30	that	that	PRON
ap-8720	180	31	trick	trick	VERB
ap-8720	180	32	the	the	DET
ap-8720	180	33	discriminator	discriminator	NOUN
ap-8720	180	34	into	into	ADP
ap-8720	180	35	categorising	categorise	VERB
ap-8720	180	36	them	they	PRON
ap-8720	180	37	as	as	ADP
ap-8720	180	38	real	real	ADJ
ap-8720	180	39	images	image	NOUN
ap-8720	180	40	.	.	PUNCT
ap-8720	181	1	3.1	3.1	NUM
ap-8720	181	2	.	.	PUNCT
ap-8720	182	1	deep	deep	ADJ
ap-8720	182	2	convolutional	convolutional	ADJ
ap-8720	182	3	generative	generative	ADJ
ap-8720	182	4	adversarial	adversarial	ADJ
ap-8720	182	5	network	network	NOUN
ap-8720	182	6	(	(	PUNCT
ap-8720	182	7	dcgan	dcgan	NOUN
ap-8720	182	8	)	)	PUNCT
ap-8720	182	9	as	as	ADP
ap-8720	182	10	an	an	DET
ap-8720	182	11	extension	extension	NOUN
ap-8720	182	12	,	,	PUNCT
ap-8720	182	13	dcgan	dcgan	NOUN
ap-8720	182	14	incorporates	incorporate	VERB
ap-8720	182	15	convolutional	convolutional	ADJ
ap-8720	182	16	and	and	CCONJ
ap-8720	182	17	convolutional	convolutional	ADJ
ap-8720	182	18	-	-	PUNCT
ap-8720	182	19	transposed	transpose	VERB
ap-8720	182	20	layers	layer	NOUN
ap-8720	182	21	explicitly	explicitly	ADV
ap-8720	182	22	in	in	ADP
ap-8720	182	23	both	both	CCONJ
ap-8720	182	24	the	the	DET
ap-8720	182	25	discriminator	discriminator	NOUN
ap-8720	182	26	and	and	CCONJ
ap-8720	182	27	the	the	DET
ap-8720	182	28	generator	generator	NOUN
ap-8720	182	29	components	component	NOUN
ap-8720	182	30	.	.	PUNCT
ap-8720	183	1	convolutional	convolutional	ADJ
ap-8720	183	2	layers	layer	NOUN
ap-8720	183	3	preserve	preserve	VERB
ap-8720	183	4	the	the	DET
ap-8720	183	5	spatial	spatial	ADJ
ap-8720	183	6	structure	structure	NOUN
ap-8720	183	7	of	of	ADP
ap-8720	183	8	an	an	DET
ap-8720	183	9	308	308	NUM
ap-8720	183	10	vol	vol	NOUN
ap-8720	183	11	.	.	PUNCT
ap-8720	184	1	63	63	NUM
ap-8720	185	1	no	no	NOUN
ap-8720	185	2	.	.	PUNCT
ap-8720	186	1	5/2023	5/2023	NUM
ap-8720	186	2	fake	fake	ADJ
ap-8720	186	3	face	face	NOUN
ap-8720	186	4	detection	detection	NOUN
ap-8720	186	5	based	base	VERB
ap-8720	186	6	on	on	ADP
ap-8720	186	7	a	a	DET
ap-8720	186	8	multi	multi	ADJ
ap-8720	186	9	discriminator	discriminator	NOUN
ap-8720	186	10	deep	deep	ADJ
ap-8720	186	11	cnn	cnn	PROPN
ap-8720	186	12	architecture	architecture	NOUN
ap-8720	186	13	figure	figure	NOUN
ap-8720	186	14	3	3	NUM
ap-8720	186	15	.	.	PUNCT
ap-8720	187	1	deep	deep	ADJ
ap-8720	187	2	convolutional	convolutional	ADJ
ap-8720	187	3	gan	gan	NOUN
ap-8720	187	4	network	network	NOUN
ap-8720	187	5	.	.	PUNCT
ap-8720	188	1	image	image	NOUN
ap-8720	188	2	,	,	PUNCT
ap-8720	188	3	which	which	PRON
ap-8720	188	4	means	mean	VERB
ap-8720	188	5	that	that	SCONJ
ap-8720	188	6	the	the	DET
ap-8720	188	7	most	most	ADV
ap-8720	188	8	accurate	accurate	ADJ
ap-8720	188	9	and	and	CCONJ
ap-8720	188	10	detailed	detailed	ADJ
ap-8720	188	11	features	feature	NOUN
ap-8720	188	12	are	be	AUX
ap-8720	188	13	extracted	extract	VERB
ap-8720	188	14	from	from	ADP
ap-8720	188	15	an	an	DET
ap-8720	188	16	image	image	NOUN
ap-8720	188	17	.	.	PUNCT
ap-8720	189	1	the	the	DET
ap-8720	189	2	generator	generator	NOUN
ap-8720	189	3	and	and	CCONJ
ap-8720	189	4	discriminator	discriminator	NOUN
ap-8720	189	5	will	will	AUX
ap-8720	189	6	,	,	PUNCT
ap-8720	189	7	therefore	therefore	ADV
ap-8720	189	8	,	,	PUNCT
ap-8720	189	9	have	have	VERB
ap-8720	189	10	more	more	ADV
ap-8720	189	11	advanced	advanced	ADJ
ap-8720	189	12	spatial	spatial	ADJ
ap-8720	189	13	reasoning	reasoning	NOUN
ap-8720	189	14	capabilities	capability	NOUN
ap-8720	189	15	about	about	ADP
ap-8720	189	16	the	the	DET
ap-8720	189	17	output	output	NOUN
ap-8720	189	18	they	they	PRON
ap-8720	189	19	will	will	AUX
ap-8720	189	20	generate	generate	VERB
ap-8720	189	21	and	and	CCONJ
ap-8720	189	22	how	how	SCONJ
ap-8720	189	23	to	to	PART
ap-8720	189	24	distinguish	distinguish	VERB
ap-8720	189	25	between	between	ADP
ap-8720	189	26	the	the	DET
ap-8720	189	27	features	feature	NOUN
ap-8720	189	28	of	of	ADP
ap-8720	189	29	the	the	DET
ap-8720	189	30	real	real	ADJ
ap-8720	189	31	images	image	NOUN
ap-8720	189	32	and	and	CCONJ
ap-8720	189	33	the	the	DET
ap-8720	189	34	fake	fake	ADJ
ap-8720	189	35	images	image	NOUN
ap-8720	189	36	.	.	PUNCT
ap-8720	190	1	improving	improve	VERB
ap-8720	190	2	the	the	DET
ap-8720	190	3	extracted	extract	VERB
ap-8720	190	4	features	feature	NOUN
ap-8720	190	5	’	'	PUNCT
ap-8720	190	6	quality	quality	NOUN
ap-8720	190	7	is	be	AUX
ap-8720	190	8	usually	usually	ADV
ap-8720	190	9	the	the	DET
ap-8720	190	10	reason	reason	NOUN
ap-8720	190	11	why	why	SCONJ
ap-8720	190	12	dcgans	dcgan	NOUN
ap-8720	190	13	are	be	AUX
ap-8720	190	14	used	use	VERB
ap-8720	190	15	for	for	ADP
ap-8720	190	16	image	image	NOUN
ap-8720	190	17	processing	processing	NOUN
ap-8720	190	18	[	[	X
ap-8720	190	19	32	32	NUM
ap-8720	190	20	]	]	PUNCT
ap-8720	190	21	.	.	PUNCT
ap-8720	191	1	like	like	ADP
ap-8720	191	2	all	all	DET
ap-8720	191	3	gans	gan	NOUN
ap-8720	191	4	,	,	PUNCT
ap-8720	191	5	the	the	DET
ap-8720	191	6	dcgan	dcgan	NOUN
ap-8720	191	7	consists	consist	VERB
ap-8720	191	8	of	of	ADP
ap-8720	191	9	a	a	DET
ap-8720	191	10	generator	generator	NOUN
ap-8720	191	11	and	and	CCONJ
ap-8720	191	12	a	a	DET
ap-8720	191	13	discriminator	discriminator	NOUN
ap-8720	191	14	(	(	PUNCT
ap-8720	191	15	figure	figure	NOUN
ap-8720	191	16	3	3	NUM
ap-8720	191	17	)	)	PUNCT
ap-8720	191	18	.	.	PUNCT
ap-8720	192	1	3.2	3.2	NUM
ap-8720	192	2	.	.	PUNCT
ap-8720	193	1	multi	multi	PROPN
ap-8720	193	2	discriminator	discriminator	PROPN
ap-8720	193	3	deep	deep	PROPN
ap-8720	193	4	cnn	cnn	PROPN
ap-8720	193	5	architecture	architecture	NOUN
ap-8720	193	6	(	(	PUNCT
ap-8720	193	7	mdd	mdd	PROPN
ap-8720	193	8	-	-	PUNCT
ap-8720	193	9	cnn	cnn	PROPN
ap-8720	193	10	)	)	PUNCT
ap-8720	193	11	our	our	PRON
ap-8720	193	12	proposed	propose	VERB
ap-8720	193	13	architecture	architecture	NOUN
ap-8720	193	14	is	be	AUX
ap-8720	193	15	based	base	VERB
ap-8720	193	16	on	on	ADP
ap-8720	193	17	three	three	NUM
ap-8720	193	18	discriminators	discriminator	NOUN
ap-8720	193	19	.	.	PUNCT
ap-8720	194	1	the	the	DET
ap-8720	194	2	architecture	architecture	NOUN
ap-8720	194	3	is	be	AUX
ap-8720	194	4	described	describe	VERB
ap-8720	194	5	in	in	ADP
ap-8720	194	6	figure	figure	NOUN
ap-8720	194	7	4	4	NUM
ap-8720	194	8	.	.	PUNCT
ap-8720	194	9	where	where	SCONJ
ap-8720	194	10	the	the	DET
ap-8720	194	11	main	main	ADJ
ap-8720	194	12	modules	module	NOUN
ap-8720	194	13	are	be	AUX
ap-8720	194	14	:	:	PUNCT
ap-8720	194	15	•	•	ADP
ap-8720	194	16	a	a	DET
ap-8720	194	17	new	new	ADJ
ap-8720	194	18	dcgan	dcgan	NOUN
ap-8720	194	19	architecture	architecture	NOUN
ap-8720	194	20	called	call	VERB
ap-8720	194	21	ndcgan	ndcgan	NOUN
ap-8720	194	22	,	,	PUNCT
ap-8720	194	23	where	where	SCONJ
ap-8720	194	24	the	the	DET
ap-8720	194	25	discriminator	discriminator	NOUN
ap-8720	194	26	1	1	NUM
ap-8720	194	27	trains	train	NOUN
ap-8720	194	28	with	with	ADP
ap-8720	194	29	its	its	PRON
ap-8720	194	30	opponent	opponent	NOUN
ap-8720	194	31	(	(	PUNCT
ap-8720	194	32	the	the	DET
ap-8720	194	33	generator	generator	NOUN
ap-8720	194	34	)	)	PUNCT
ap-8720	194	35	to	to	PART
ap-8720	194	36	overtake	overtake	VERB
ap-8720	194	37	it	it	PRON
ap-8720	194	38	and	and	CCONJ
ap-8720	194	39	recognise	recognise	VERB
ap-8720	194	40	the	the	DET
ap-8720	194	41	fake	fake	ADJ
ap-8720	194	42	faces	face	NOUN
ap-8720	194	43	generated	generate	VERB
ap-8720	194	44	by	by	ADP
ap-8720	194	45	the	the	DET
ap-8720	194	46	latter	latter	ADJ
ap-8720	194	47	with	with	ADP
ap-8720	194	48	a	a	DET
ap-8720	194	49	significant	significant	ADJ
ap-8720	194	50	rate	rate	NOUN
ap-8720	194	51	.	.	PUNCT
ap-8720	195	1	•	•	NUM
ap-8720	195	2	the	the	DET
ap-8720	195	3	second	second	ADJ
ap-8720	195	4	module	module	NOUN
ap-8720	195	5	of	of	ADP
ap-8720	195	6	our	our	PRON
ap-8720	195	7	architecture	architecture	NOUN
ap-8720	195	8	is	be	AUX
ap-8720	195	9	the	the	DET
ap-8720	195	10	discriminator	discriminator	NOUN
ap-8720	195	11	2	2	NUM
ap-8720	195	12	:	:	PUNCT
ap-8720	195	13	this	this	PRON
ap-8720	195	14	is	be	AUX
ap-8720	195	15	the	the	DET
ap-8720	195	16	ndcgan	ndcgan	ADJ
ap-8720	195	17	discriminator	discriminator	NOUN
ap-8720	195	18	(	(	PUNCT
ap-8720	195	19	discriminator	discriminator	NOUN
ap-8720	195	20	1	1	NUM
ap-8720	195	21	)	)	PUNCT
ap-8720	195	22	trained	train	VERB
ap-8720	195	23	by	by	ADP
ap-8720	195	24	transfer	transfer	NOUN
ap-8720	195	25	learning	learning	NOUN
ap-8720	195	26	.	.	PUNCT
ap-8720	196	1	•	•	NUM
ap-8720	196	2	the	the	DET
ap-8720	196	3	third	third	ADJ
ap-8720	196	4	module	module	NOUN
ap-8720	196	5	is	be	AUX
ap-8720	196	6	the	the	DET
ap-8720	196	7	discriminator	discriminator	NOUN
ap-8720	196	8	3	3	NUM
ap-8720	196	9	:	:	PUNCT
ap-8720	196	10	which	which	PRON
ap-8720	196	11	is	be	AUX
ap-8720	196	12	the	the	DET
ap-8720	196	13	discriminator	discriminator	NOUN
ap-8720	196	14	of	of	ADP
ap-8720	196	15	the	the	DET
ap-8720	196	16	ndcgan	ndcgan	NOUN
ap-8720	196	17	trained	train	VERB
ap-8720	196	18	from	from	ADP
ap-8720	196	19	scratch	scratch	NOUN
ap-8720	196	20	on	on	ADP
ap-8720	196	21	faces	face	NOUN
ap-8720	196	22	and	and	CCONJ
ap-8720	196	23	fake	fake	ADJ
ap-8720	196	24	faces	face	NOUN
ap-8720	196	25	,	,	PUNCT
ap-8720	196	26	without	without	ADP
ap-8720	196	27	its	its	PRON
ap-8720	196	28	generator	generator	NOUN
ap-8720	196	29	.	.	PUNCT
ap-8720	197	1	the	the	DET
ap-8720	197	2	final	final	ADJ
ap-8720	197	3	decision	decision	NOUN
ap-8720	197	4	of	of	ADP
ap-8720	197	5	whether	whether	SCONJ
ap-8720	197	6	the	the	DET
ap-8720	197	7	face	face	NOUN
ap-8720	197	8	belongs	belong	VERB
ap-8720	197	9	to	to	ADP
ap-8720	197	10	the	the	DET
ap-8720	197	11	real	real	ADJ
ap-8720	197	12	face	face	NOUN
ap-8720	197	13	class	class	NOUN
ap-8720	197	14	or	or	CCONJ
ap-8720	197	15	the	the	DET
ap-8720	197	16	fake	fake	ADJ
ap-8720	197	17	face	face	NOUN
ap-8720	197	18	class	class	NOUN
ap-8720	197	19	is	be	AUX
ap-8720	197	20	the	the	DET
ap-8720	197	21	voting	voting	NOUN
ap-8720	197	22	result	result	NOUN
ap-8720	197	23	of	of	ADP
ap-8720	197	24	the	the	DET
ap-8720	197	25	three	three	NUM
ap-8720	197	26	discriminators	discriminator	NOUN
ap-8720	197	27	.	.	PUNCT
ap-8720	198	1	3.2.1	3.2.1	X
ap-8720	198	2	.	.	PUNCT
ap-8720	198	3	new	new	ADJ
ap-8720	198	4	dcgan	dcgan	NOUN
ap-8720	198	5	architecture	architecture	NOUN
ap-8720	198	6	in	in	ADP
ap-8720	198	7	the	the	DET
ap-8720	198	8	following	following	NOUN
ap-8720	198	9	,	,	PUNCT
ap-8720	198	10	we	we	PRON
ap-8720	198	11	describe	describe	VERB
ap-8720	198	12	the	the	DET
ap-8720	198	13	dcgan	dcgan	NOUN
ap-8720	198	14	architecture	architecture	NOUN
ap-8720	198	15	that	that	PRON
ap-8720	198	16	consists	consist	VERB
ap-8720	198	17	of	of	ADP
ap-8720	198	18	a	a	DET
ap-8720	198	19	generator	generator	NOUN
ap-8720	198	20	network	network	NOUN
ap-8720	198	21	and	and	CCONJ
ap-8720	198	22	a	a	DET
ap-8720	198	23	discriminator	discriminator	NOUN
ap-8720	198	24	network	network	NOUN
ap-8720	198	25	,	,	PUNCT
ap-8720	198	26	which	which	PRON
ap-8720	198	27	are	be	AUX
ap-8720	198	28	the	the	DET
ap-8720	198	29	main	main	ADJ
ap-8720	198	30	parts	part	NOUN
ap-8720	198	31	of	of	ADP
ap-8720	198	32	the	the	DET
ap-8720	198	33	mdd	mdd	PROPN
ap-8720	198	34	-	-	PUNCT
ap-8720	198	35	cnn	cnn	PROPN
ap-8720	198	36	architecture	architecture	NOUN
ap-8720	198	37	.	.	PUNCT
ap-8720	199	1	a.	a.	NOUN
ap-8720	199	2	architecture	architecture	NOUN
ap-8720	199	3	of	of	ADP
ap-8720	199	4	generator	generator	NOUN
ap-8720	199	5	network	network	NOUN
ap-8720	199	6	the	the	DET
ap-8720	199	7	generator	generator	NOUN
ap-8720	199	8	consists	consist	VERB
ap-8720	199	9	of	of	ADP
ap-8720	199	10	transposed	transpose	VERB
ap-8720	199	11	convolutional	convolutional	ADJ
ap-8720	199	12	layers	layer	NOUN
ap-8720	199	13	that	that	PRON
ap-8720	199	14	allow	allow	VERB
ap-8720	199	15	the	the	DET
ap-8720	199	16	noise	noise	NOUN
ap-8720	199	17	vector	vector	NOUN
ap-8720	199	18	to	to	PART
ap-8720	199	19	be	be	AUX
ap-8720	199	20	unsampled	unsample	VERB
ap-8720	199	21	and	and	CCONJ
ap-8720	199	22	transformed	transform	VERB
ap-8720	199	23	into	into	ADP
ap-8720	199	24	an	an	DET
ap-8720	199	25	image	image	NOUN
ap-8720	199	26	.	.	PUNCT
ap-8720	200	1	in	in	ADP
ap-8720	200	2	a	a	DET
ap-8720	200	3	classical	classical	ADJ
ap-8720	200	4	cnn	cnn	NOUN
ap-8720	200	5	network	network	NOUN
ap-8720	200	6	,	,	PUNCT
ap-8720	200	7	the	the	DET
ap-8720	200	8	convolutional	convolutional	ADJ
ap-8720	200	9	layers	layer	NOUN
ap-8720	200	10	will	will	AUX
ap-8720	200	11	try	try	VERB
ap-8720	200	12	to	to	PART
ap-8720	200	13	extract	extract	VERB
ap-8720	200	14	smaller	small	ADJ
ap-8720	200	15	and	and	CCONJ
ap-8720	200	16	smaller	small	ADJ
ap-8720	200	17	features	feature	NOUN
ap-8720	200	18	which	which	PRON
ap-8720	200	19	are	be	AUX
ap-8720	200	20	then	then	ADV
ap-8720	200	21	be	be	AUX
ap-8720	200	22	classified	classify	VERB
ap-8720	200	23	.	.	PUNCT
ap-8720	201	1	in	in	ADP
ap-8720	201	2	a	a	DET
ap-8720	201	3	generator	generator	NOUN
ap-8720	201	4	,	,	PUNCT
ap-8720	201	5	the	the	DET
ap-8720	201	6	transposed	transpose	VERB
ap-8720	201	7	convolutional	convolutional	ADJ
ap-8720	201	8	layers	layer	NOUN
ap-8720	201	9	are	be	AUX
ap-8720	201	10	designed	design	VERB
ap-8720	201	11	to	to	PART
ap-8720	201	12	reverse	reverse	VERB
ap-8720	201	13	the	the	DET
ap-8720	201	14	operations	operation	NOUN
ap-8720	201	15	of	of	ADP
ap-8720	201	16	the	the	DET
ap-8720	201	17	convolutional	convolutional	ADJ
ap-8720	201	18	layer	layer	NOUN
ap-8720	201	19	.	.	PUNCT
ap-8720	202	1	this	this	DET
ap-8720	202	2	upsampling	upsampling	NOUN
ap-8720	202	3	process	process	NOUN
ap-8720	202	4	enlarges	enlarge	VERB
ap-8720	202	5	the	the	DET
ap-8720	202	6	image	image	NOUN
ap-8720	202	7	and	and	CCONJ
ap-8720	202	8	adds	add	VERB
ap-8720	202	9	details	detail	NOUN
ap-8720	202	10	to	to	ADP
ap-8720	202	11	the	the	DET
ap-8720	202	12	final	final	ADJ
ap-8720	202	13	result	result	NOUN
ap-8720	202	14	.	.	PUNCT
ap-8720	203	1	this	this	PRON
ap-8720	203	2	is	be	AUX
ap-8720	203	3	the	the	DET
ap-8720	203	4	part	part	NOUN
ap-8720	203	5	of	of	ADP
ap-8720	203	6	the	the	DET
ap-8720	203	7	generator	generator	NOUN
ap-8720	203	8	that	that	SCONJ
ap-8720	203	9	“	"	PUNCT
ap-8720	203	10	draws	draw	VERB
ap-8720	203	11	”	"	PUNCT
ap-8720	203	12	the	the	DET
ap-8720	203	13	actual	actual	ADJ
ap-8720	203	14	image	image	NOUN
ap-8720	203	15	.	.	PUNCT
ap-8720	204	1	the	the	DET
ap-8720	204	2	transposed	transpose	VERB
ap-8720	204	3	convolution	convolution	NOUN
ap-8720	204	4	layers	layer	NOUN
ap-8720	204	5	of	of	ADP
ap-8720	204	6	the	the	DET
ap-8720	204	7	generator	generator	NOUN
ap-8720	204	8	will	will	AUX
ap-8720	204	9	decompress	decompress	VERB
ap-8720	204	10	the	the	DET
ap-8720	204	11	noise	noise	NOUN
ap-8720	204	12	so	so	SCONJ
ap-8720	204	13	that	that	SCONJ
ap-8720	204	14	it	it	PRON
ap-8720	204	15	becomes	become	VERB
ap-8720	204	16	a	a	DET
ap-8720	204	17	128	128	NUM
ap-8720	204	18	×	×	NOUN
ap-8720	204	19	128	128	NUM
ap-8720	204	20	image	image	NOUN
ap-8720	204	21	with	with	ADP
ap-8720	204	22	all	all	DET
ap-8720	204	23	the	the	DET
ap-8720	204	24	details	detail	NOUN
ap-8720	204	25	in	in	ADP
ap-8720	204	26	the	the	DET
ap-8720	204	27	right	right	ADJ
ap-8720	204	28	place	place	NOUN
ap-8720	204	29	in	in	ADP
ap-8720	204	30	the	the	DET
ap-8720	204	31	image	image	NOUN
ap-8720	204	32	.	.	PUNCT
ap-8720	205	1	the	the	DET
ap-8720	205	2	details	detail	NOUN
ap-8720	205	3	of	of	ADP
ap-8720	205	4	each	each	DET
ap-8720	205	5	layer	layer	NOUN
ap-8720	205	6	are	be	AUX
ap-8720	205	7	shown	show	VERB
ap-8720	205	8	in	in	ADP
ap-8720	205	9	figure	figure	NOUN
ap-8720	205	10	5	5	NUM
ap-8720	205	11	and	and	CCONJ
ap-8720	205	12	table	table	NOUN
ap-8720	205	13	1	1	NUM
ap-8720	205	14	.	.	PUNCT
ap-8720	206	1	note	note	VERB
ap-8720	206	2	that	that	SCONJ
ap-8720	206	3	the	the	DET
ap-8720	206	4	activation	activation	NOUN
ap-8720	206	5	of	of	ADP
ap-8720	206	6	all	all	DET
ap-8720	206	7	layers	layer	NOUN
ap-8720	206	8	is	be	AUX
ap-8720	206	9	leakyrelu	leakyrelu	ADJ
ap-8720	206	10	(	(	PUNCT
ap-8720	206	11	given	give	VERB
ap-8720	206	12	by	by	ADP
ap-8720	206	13	equation	equation	NOUN
ap-8720	206	14	(	(	PUNCT
ap-8720	206	15	1	1	NUM
ap-8720	206	16	)	)	PUNCT
ap-8720	206	17	)	)	PUNCT
ap-8720	206	18	,	,	PUNCT
ap-8720	206	19	except	except	SCONJ
ap-8720	206	20	for	for	ADP
ap-8720	206	21	the	the	DET
ap-8720	206	22	output	output	NOUN
ap-8720	206	23	layer	layer	NOUN
ap-8720	206	24	which	which	PRON
ap-8720	206	25	uses	use	VERB
ap-8720	206	26	tanh	tanh	PROPN
ap-8720	206	27	(	(	PUNCT
ap-8720	206	28	equation	equation	NOUN
ap-8720	206	29	(	(	PUNCT
ap-8720	206	30	2	2	NUM
ap-8720	206	31	)	)	PUNCT
ap-8720	206	32	)	)	PUNCT
ap-8720	206	33	.	.	PUNCT
ap-8720	207	1	∅(x	∅(x	NOUN
ap-8720	207	2	)	)	PUNCT
ap-8720	208	1	=	=	PRON
ap-8720	208	2	{	{	PUNCT
ap-8720	208	3	αx	αx	ADV
ap-8720	208	4	if	if	SCONJ
ap-8720	208	5	x	x	SYM
ap-8720	208	6	≤	≤	ADV
ap-8720	208	7	0	0	NUM
ap-8720	208	8	x	x	SYM
ap-8720	208	9	if	if	SCONJ
ap-8720	208	10	x	x	PROPN
ap-8720	208	11	>	>	X
ap-8720	208	12	0	0	NUM
ap-8720	208	13	,	,	PUNCT
ap-8720	208	14	(	(	PUNCT
ap-8720	208	15	1	1	X
ap-8720	208	16	)	)	PUNCT
ap-8720	208	17	tanh(x	tanh(x	PROPN
ap-8720	208	18	)	)	PUNCT
ap-8720	208	19	=	=	SYM
ap-8720	209	1	2	2	NUM
ap-8720	209	2	1	1	NUM
ap-8720	209	3	+	+	NUM
ap-8720	209	4	e−2x	e−2x	NOUN
ap-8720	209	5	−	−	NOUN
ap-8720	209	6	1	1	NUM
ap-8720	209	7	.	.	PUNCT
ap-8720	210	1	(	(	PUNCT
ap-8720	210	2	2	2	X
ap-8720	210	3	)	)	PUNCT
ap-8720	210	4	b.	b.	PROPN
ap-8720	210	5	discriminator	discriminator	NOUN
ap-8720	210	6	network	network	NOUN
ap-8720	210	7	architecture	architecture	NOUN
ap-8720	210	8	the	the	DET
ap-8720	210	9	discriminator	discriminator	NOUN
ap-8720	210	10	is	be	AUX
ap-8720	210	11	composed	compose	VERB
ap-8720	210	12	of	of	ADP
ap-8720	210	13	strided	strided	ADJ
ap-8720	210	14	convolutional	convolutional	ADJ
ap-8720	210	15	layers	layer	NOUN
ap-8720	210	16	and	and	CCONJ
ap-8720	210	17	leakyrelu	leakyrelu	ADJ
ap-8720	210	18	activations	activation	NOUN
ap-8720	210	19	(	(	PUNCT
ap-8720	210	20	table	table	NOUN
ap-8720	210	21	2	2	NUM
ap-8720	210	22	,	,	PUNCT
ap-8720	210	23	figure	figure	VERB
ap-8720	210	24	6	6	NUM
ap-8720	210	25	)	)	PUNCT
ap-8720	210	26	.	.	PUNCT
ap-8720	211	1	the	the	DET
ap-8720	211	2	input	input	NOUN
ap-8720	211	3	is	be	AUX
ap-8720	211	4	a	a	DET
ap-8720	211	5	128	128	NUM
ap-8720	211	6	×	×	NOUN
ap-8720	211	7	128	128	NUM
ap-8720	211	8	×	×	NOUN
ap-8720	211	9	3	3	NUM
ap-8720	211	10	image	image	NOUN
ap-8720	211	11	,	,	PUNCT
ap-8720	211	12	and	and	CCONJ
ap-8720	211	13	the	the	DET
ap-8720	211	14	output	output	NOUN
ap-8720	211	15	is	be	AUX
ap-8720	211	16	a	a	DET
ap-8720	211	17	probability	probability	NOUN
ap-8720	211	18	between	between	ADP
ap-8720	211	19	0	0	NUM
ap-8720	211	20	and	and	CCONJ
ap-8720	211	21	1	1	NUM
ap-8720	211	22	.	.	X
ap-8720	211	23	3.2.2	3.2.2	NUM
ap-8720	211	24	.	.	PUNCT
ap-8720	212	1	stability	stability	NOUN
ap-8720	212	2	conditions	condition	NOUN
ap-8720	212	3	to	to	PART
ap-8720	212	4	have	have	AUX
ap-8720	212	5	a	a	DET
ap-8720	212	6	stable	stable	ADJ
ap-8720	212	7	network	network	NOUN
ap-8720	212	8	,	,	PUNCT
ap-8720	212	9	the	the	DET
ap-8720	212	10	following	follow	VERB
ap-8720	212	11	is	be	AUX
ap-8720	212	12	necessary	necessary	ADJ
ap-8720	212	13	:	:	PUNCT
ap-8720	212	14	•	•	ADP
ap-8720	212	15	replacing	replace	VERB
ap-8720	212	16	pooling	pool	VERB
ap-8720	212	17	by	by	ADP
ap-8720	212	18	stride	stride	ADJ
ap-8720	212	19	convolution	convolution	NOUN
ap-8720	212	20	.	.	PUNCT
ap-8720	213	1	this	this	PRON
ap-8720	213	2	allows	allow	VERB
ap-8720	213	3	the	the	DET
ap-8720	213	4	network	network	NOUN
ap-8720	213	5	to	to	PART
ap-8720	213	6	learn	learn	VERB
ap-8720	213	7	its	its	PRON
ap-8720	213	8	own	own	ADJ
ap-8720	213	9	spatial	spatial	ADJ
ap-8720	213	10	downsampling	downsampling	NOUN
ap-8720	213	11	(	(	PUNCT
ap-8720	213	12	by	by	ADP
ap-8720	213	13	changing	change	VERB
ap-8720	213	14	the	the	DET
ap-8720	213	15	size	size	NOUN
ap-8720	213	16	of	of	ADP
ap-8720	213	17	the	the	DET
ap-8720	213	18	input	input	NOUN
ap-8720	213	19	)	)	PUNCT
ap-8720	213	20	.	.	PUNCT
ap-8720	214	1	cnns	cnns	PROPN
ap-8720	214	2	use	use	VERB
ap-8720	214	3	pooling	pool	VERB
ap-8720	214	4	layers	layer	NOUN
ap-8720	214	5	to	to	PART
ap-8720	214	6	reduce	reduce	VERB
ap-8720	214	7	dimensions	dimension	NOUN
ap-8720	214	8	.	.	PUNCT
ap-8720	215	1	for	for	ADP
ap-8720	215	2	example	example	NOUN
ap-8720	215	3	,	,	PUNCT
ap-8720	215	4	a	a	DET
ap-8720	215	5	2	2	NUM
ap-8720	215	6	×	×	NOUN
ap-8720	215	7	2	2	NUM
ap-8720	215	8	max	max	NOUN
ap-8720	215	9	pooling	pooling	NOUN
ap-8720	215	10	layer	layer	NOUN
ap-8720	215	11	would	would	AUX
ap-8720	215	12	take	take	VERB
ap-8720	215	13	a	a	DET
ap-8720	215	14	2	2	NUM
ap-8720	215	15	×	×	NOUN
ap-8720	215	16	2	2	NUM
ap-8720	215	17	array	array	NOUN
ap-8720	215	18	of	of	ADP
ap-8720	215	19	pixels	pixel	NOUN
ap-8720	215	20	and	and	CCONJ
ap-8720	215	21	map	map	VERB
ap-8720	215	22	it	it	PRON
ap-8720	215	23	to	to	ADP
ap-8720	215	24	a	a	DET
ap-8720	215	25	number	number	NOUN
ap-8720	215	26	that	that	PRON
ap-8720	215	27	is	be	AUX
ap-8720	215	28	the	the	DET
ap-8720	215	29	maximum	maximum	NOUN
ap-8720	215	30	among	among	ADP
ap-8720	215	31	them	they	PRON
ap-8720	215	32	.	.	PUNCT
ap-8720	216	1	stride	stride	ADJ
ap-8720	216	2	convolution	convolution	NOUN
ap-8720	216	3	can	can	AUX
ap-8720	216	4	reduce	reduce	VERB
ap-8720	216	5	the	the	DET
ap-8720	216	6	dimension	dimension	NOUN
ap-8720	216	7	by	by	ADP
ap-8720	216	8	skipping	skip	VERB
ap-8720	216	9	several	several	ADJ
ap-8720	216	10	pixels	pixel	NOUN
ap-8720	216	11	between	between	ADP
ap-8720	216	12	convolutions	convolution	NOUN
ap-8720	216	13	instead	instead	ADV
ap-8720	216	14	of	of	ADP
ap-8720	216	15	dragging	drag	VERB
ap-8720	216	16	the	the	DET
ap-8720	216	17	kernel	kernel	NOUN
ap-8720	216	18	one	one	NUM
ap-8720	216	19	by	by	ADP
ap-8720	216	20	one	one	NUM
ap-8720	216	21	.	.	PUNCT
ap-8720	217	1	it	it	PRON
ap-8720	217	2	can	can	AUX
ap-8720	217	3	also	also	ADV
ap-8720	217	4	increase	increase	VERB
ap-8720	217	5	the	the	DET
ap-8720	217	6	dimension	dimension	NOUN
ap-8720	217	7	by	by	ADP
ap-8720	217	8	adding	add	VERB
ap-8720	217	9	empty	empty	ADJ
ap-8720	217	10	pixels	pixel	NOUN
ap-8720	217	11	between	between	ADP
ap-8720	217	12	the	the	DET
ap-8720	217	13	actual	actual	ADJ
ap-8720	217	14	pixels	pixel	NOUN
ap-8720	217	15	.	.	PUNCT
ap-8720	218	1	•	•	NUM
ap-8720	218	2	no	no	DET
ap-8720	218	3	fully	fully	ADV
ap-8720	218	4	connected	connect	VERB
ap-8720	218	5	layer	layer	NOUN
ap-8720	218	6	at	at	ADP
ap-8720	218	7	the	the	DET
ap-8720	218	8	end	end	NOUN
ap-8720	218	9	of	of	ADP
ap-8720	218	10	the	the	DET
ap-8720	218	11	cnn	cnn	PROPN
ap-8720	218	12	.	.	PUNCT
ap-8720	219	1	the	the	DET
ap-8720	219	2	generator	generator	NOUN
ap-8720	219	3	is	be	AUX
ap-8720	219	4	not	not	PART
ap-8720	219	5	a	a	DET
ap-8720	219	6	classifier	classifier	NOUN
ap-8720	219	7	,	,	PUNCT
ap-8720	219	8	so	so	SCONJ
ap-8720	219	9	this	this	DET
ap-8720	219	10	part	part	NOUN
ap-8720	219	11	is	be	AUX
ap-8720	219	12	not	not	PART
ap-8720	219	13	needed	need	VERB
ap-8720	219	14	.	.	PUNCT
ap-8720	220	1	•	•	NUM
ap-8720	220	2	use	use	NOUN
ap-8720	220	3	leakyrelu	leakyrelu	NOUN
ap-8720	220	4	(	(	PUNCT
ap-8720	220	5	figure	figure	NOUN
ap-8720	220	6	7	7	NUM
ap-8720	220	7	)	)	PUNCT
ap-8720	220	8	in	in	ADP
ap-8720	220	9	the	the	DET
ap-8720	220	10	generator	generator	NOUN
ap-8720	220	11	except	except	SCONJ
ap-8720	220	12	for	for	ADP
ap-8720	220	13	the	the	DET
ap-8720	220	14	output	output	NOUN
ap-8720	220	15	,	,	PUNCT
ap-8720	220	16	which	which	PRON
ap-8720	220	17	uses	use	VERB
ap-8720	220	18	tanh	tanh	PROPN
ap-8720	220	19	.	.	PUNCT
ap-8720	221	1	the	the	DET
ap-8720	221	2	symmetry	symmetry	NOUN
ap-8720	221	3	of	of	ADP
ap-8720	221	4	the	the	DET
ap-8720	221	5	tanh	tanh	PROPN
ap-8720	221	6	function	function	NOUN
ap-8720	221	7	allows	allow	VERB
ap-8720	221	8	the	the	DET
ap-8720	221	9	model	model	NOUN
ap-8720	221	10	to	to	PART
ap-8720	221	11	learn	learn	VERB
ap-8720	221	12	faster	fast	ADV
ap-8720	221	13	and	and	CCONJ
ap-8720	221	14	leakyrelu	leakyrelu	VERB
ap-8720	221	15	for	for	ADP
ap-8720	221	16	the	the	DET
ap-8720	221	17	discriminator	discriminator	NOUN
ap-8720	221	18	.	.	PUNCT
ap-8720	222	1	the	the	DET
ap-8720	222	2	output	output	NOUN
ap-8720	222	3	of	of	ADP
ap-8720	222	4	the	the	DET
ap-8720	222	5	leaky	leaky	ADJ
ap-8720	222	6	relu	relu	NOUN
ap-8720	222	7	activation	activation	NOUN
ap-8720	222	8	function	function	NOUN
ap-8720	222	9	is	be	AUX
ap-8720	222	10	positive	positive	ADJ
ap-8720	222	11	if	if	SCONJ
ap-8720	222	12	the	the	DET
ap-8720	222	13	input	input	NOUN
ap-8720	222	14	is	be	AUX
ap-8720	222	15	positive	positive	ADJ
ap-8720	222	16	,	,	PUNCT
ap-8720	222	17	and	and	CCONJ
ap-8720	222	18	a	a	DET
ap-8720	222	19	controlled	control	VERB
ap-8720	222	20	negative	negative	ADJ
ap-8720	222	21	value	value	NOUN
ap-8720	222	22	if	if	SCONJ
ap-8720	222	23	the	the	DET
ap-8720	222	24	input	input	NOUN
ap-8720	222	25	is	be	AUX
ap-8720	222	26	negative	negative	ADJ
ap-8720	222	27	.	.	PUNCT
ap-8720	223	1	the	the	DET
ap-8720	223	2	negative	negative	ADJ
ap-8720	223	3	value	value	NOUN
ap-8720	223	4	309	309	NUM
ap-8720	223	5	ch	ch	NOUN
ap-8720	223	6	.	.	PUNCT
ap-8720	223	7	e.	e.	PROPN
ap-8720	223	8	bencheriet	bencheriet	PROPN
ap-8720	223	9	,	,	PUNCT
ap-8720	223	10	h.	h.	PROPN
ap-8720	223	11	abdelmoumène	abdelmoumène	PROPN
ap-8720	223	12	,	,	PUNCT
ap-8720	223	13	a.	a.	PROPN
ap-8720	223	14	sebbagh	sebbagh	PROPN
ap-8720	223	15	et	et	PROPN
ap-8720	223	16	al	al	PROPN
ap-8720	223	17	.	.	PROPN
ap-8720	223	18	acta	acta	PROPN
ap-8720	223	19	polytechnica	polytechnica	PROPN
ap-8720	223	20	figure	figure	NOUN
ap-8720	223	21	4	4	NUM
ap-8720	223	22	.	.	PUNCT
ap-8720	223	23	architecture	architecture	NOUN
ap-8720	223	24	of	of	ADP
ap-8720	223	25	the	the	DET
ap-8720	223	26	proposed	propose	VERB
ap-8720	223	27	system	system	NOUN
ap-8720	223	28	.	.	PUNCT
ap-8720	224	1	figure	figure	NOUN
ap-8720	224	2	5	5	NUM
ap-8720	224	3	.	.	PUNCT
ap-8720	225	1	architecture	architecture	NOUN
ap-8720	225	2	of	of	ADP
ap-8720	225	3	the	the	DET
ap-8720	225	4	proposed	propose	VERB
ap-8720	225	5	ndcgan	ndcgan	PROPN
ap-8720	225	6	generator	generator	PROPN
ap-8720	225	7	.	.	PUNCT
ap-8720	226	1	310	310	NUM
ap-8720	226	2	vol	vol	NOUN
ap-8720	226	3	.	.	PROPN
ap-8720	226	4	63	63	NUM
ap-8720	226	5	no	no	NOUN
ap-8720	226	6	.	.	PUNCT
ap-8720	227	1	5/2023	5/2023	NUM
ap-8720	227	2	fake	fake	ADJ
ap-8720	227	3	face	face	NOUN
ap-8720	227	4	detection	detection	NOUN
ap-8720	227	5	based	base	VERB
ap-8720	227	6	on	on	ADP
ap-8720	227	7	a	a	DET
ap-8720	227	8	multi	multi	ADJ
ap-8720	227	9	discriminator	discriminator	NOUN
ap-8720	227	10	deep	deep	ADJ
ap-8720	227	11	cnn	cnn	PROPN
ap-8720	227	12	architecture	architecture	NOUN
ap-8720	227	13	layer	layer	NOUN
ap-8720	227	14	type	type	NOUN
ap-8720	227	15	dimension	dimension	NOUN
ap-8720	227	16	kernel	kernel	PROPN
ap-8720	227	17	size	size	NOUN
ap-8720	227	18	stride	stride	NOUN
ap-8720	227	19	activation	activation	NOUN
ap-8720	227	20	input	input	NOUN
ap-8720	227	21	32	32	NUM
ap-8720	227	22	–	–	PUNCT
ap-8720	227	23	–	–	PUNCT
ap-8720	227	24	–	–	PUNCT
ap-8720	227	25	dense	dense	ADJ
ap-8720	227	26	–	–	PUNCT
ap-8720	227	27	–	–	PUNCT
ap-8720	227	28	–	–	PUNCT
ap-8720	227	29	leakyrelu	leakyrelu	NOUN
ap-8720	227	30	conv2d	conv2d	NOUN
ap-8720	227	31	(	(	PUNCT
ap-8720	227	32	16	16	NUM
ap-8720	227	33	,	,	PUNCT
ap-8720	227	34	16	16	NUM
ap-8720	227	35	,	,	PUNCT
ap-8720	227	36	256	256	NUM
ap-8720	227	37	)	)	PUNCT
ap-8720	227	38	5	5	NUM
ap-8720	227	39	1	1	NUM
ap-8720	227	40	leakyrelu	leakyrelu	NOUN
ap-8720	227	41	conv2dtranspose	conv2dtranspose	NOUN
ap-8720	227	42	(	(	PUNCT
ap-8720	227	43	32	32	NUM
ap-8720	227	44	,	,	PUNCT
ap-8720	227	45	32	32	NUM
ap-8720	227	46	,	,	PUNCT
ap-8720	227	47	256	256	NUM
ap-8720	227	48	)	)	PUNCT
ap-8720	227	49	4	4	NUM
ap-8720	227	50	2	2	NUM
ap-8720	227	51	leakyrelu	leakyrelu	NOUN
ap-8720	227	52	conv2dtranspose	conv2dtranspose	VERB
ap-8720	227	53	(	(	PUNCT
ap-8720	227	54	64	64	NUM
ap-8720	227	55	,	,	PUNCT
ap-8720	227	56	64	64	NUM
ap-8720	227	57	,	,	PUNCT
ap-8720	227	58	256	256	NUM
ap-8720	227	59	)	)	PUNCT
ap-8720	227	60	4	4	NUM
ap-8720	227	61	2	2	NUM
ap-8720	227	62	leakyrelu	leakyrelu	NOUN
ap-8720	227	63	conv2dtranspose	conv2dtranspose	VERB
ap-8720	227	64	(	(	PUNCT
ap-8720	227	65	128	128	NUM
ap-8720	227	66	,	,	PUNCT
ap-8720	227	67	128	128	NUM
ap-8720	227	68	,	,	PUNCT
ap-8720	227	69	256	256	NUM
ap-8720	227	70	)	)	PUNCT
ap-8720	227	71	4	4	NUM
ap-8720	227	72	2	2	NUM
ap-8720	227	73	leakyrelu	leakyrelu	NOUN
ap-8720	227	74	conv2d	conv2d	VERB
ap-8720	227	75	(	(	PUNCT
ap-8720	227	76	128	128	NUM
ap-8720	227	77	,	,	PUNCT
ap-8720	227	78	128	128	NUM
ap-8720	227	79	,	,	PUNCT
ap-8720	227	80	512	512	NUM
ap-8720	227	81	)	)	PUNCT
ap-8720	227	82	5	5	NUM
ap-8720	227	83	1	1	NUM
ap-8720	227	84	leakyrelu	leakyrelu	NOUN
ap-8720	227	85	conv2d	conv2d	VERB
ap-8720	227	86	(	(	PUNCT
ap-8720	227	87	128	128	NUM
ap-8720	227	88	,	,	PUNCT
ap-8720	227	89	128	128	NUM
ap-8720	227	90	,	,	PUNCT
ap-8720	227	91	512	512	NUM
ap-8720	227	92	)	)	PUNCT
ap-8720	227	93	5	5	NUM
ap-8720	227	94	1	1	NUM
ap-8720	227	95	leakyrelu	leakyrelu	NOUN
ap-8720	227	96	conv2d	conv2d	VERB
ap-8720	227	97	(	(	PUNCT
ap-8720	227	98	128	128	NUM
ap-8720	227	99	,	,	PUNCT
ap-8720	227	100	128	128	NUM
ap-8720	227	101	,	,	PUNCT
ap-8720	227	102	3	3	X
ap-8720	227	103	)	)	PUNCT
ap-8720	227	104	7	7	NUM
ap-8720	227	105	1	1	NUM
ap-8720	227	106	tanh	tanh	NOUN
ap-8720	227	107	table	table	NOUN
ap-8720	227	108	1	1	NUM
ap-8720	227	109	.	.	PUNCT
ap-8720	227	110	architecture	architecture	NOUN
ap-8720	227	111	of	of	ADP
ap-8720	227	112	the	the	DET
ap-8720	227	113	generator	generator	NOUN
ap-8720	227	114	.	.	PUNCT
ap-8720	228	1	layer	layer	NOUN
ap-8720	228	2	type	type	NOUN
ap-8720	228	3	dimension	dimension	NOUN
ap-8720	228	4	kernel	kernel	PROPN
ap-8720	228	5	size	size	NOUN
ap-8720	228	6	stride	stride	NOUN
ap-8720	228	7	activation	activation	NOUN
ap-8720	228	8	input	input	NOUN
ap-8720	228	9	(	(	PUNCT
ap-8720	228	10	128,128,3	128,128,3	NUM
ap-8720	228	11	)	)	PUNCT
ap-8720	228	12	–	–	PUNCT
ap-8720	228	13	–	–	PUNCT
ap-8720	228	14	–	–	PUNCT
ap-8720	228	15	conv2d	conv2d	VERB
ap-8720	228	16	(	(	PUNCT
ap-8720	228	17	126	126	NUM
ap-8720	228	18	,	,	PUNCT
ap-8720	228	19	126	126	NUM
ap-8720	228	20	,	,	PUNCT
ap-8720	228	21	256	256	NUM
ap-8720	228	22	)	)	PUNCT
ap-8720	228	23	3	3	NUM
ap-8720	228	24	1	1	NUM
ap-8720	228	25	leakyrelu	leakyrelu	NOUN
ap-8720	228	26	conv2d	conv2d	NOUN
ap-8720	228	27	(	(	PUNCT
ap-8720	228	28	62	62	NUM
ap-8720	228	29	,	,	PUNCT
ap-8720	228	30	62	62	NUM
ap-8720	228	31	,	,	PUNCT
ap-8720	228	32	256	256	NUM
ap-8720	228	33	)	)	PUNCT
ap-8720	228	34	4	4	NUM
ap-8720	228	35	2	2	NUM
ap-8720	228	36	leakyrelu	leakyrelu	NOUN
ap-8720	228	37	conv2d	conv2d	VERB
ap-8720	228	38	(	(	PUNCT
ap-8720	228	39	30	30	NUM
ap-8720	228	40	,	,	PUNCT
ap-8720	228	41	30	30	NUM
ap-8720	228	42	,	,	PUNCT
ap-8720	228	43	256	256	NUM
ap-8720	228	44	)	)	PUNCT
ap-8720	228	45	4	4	NUM
ap-8720	228	46	2	2	NUM
ap-8720	228	47	leakyrelu	leakyrelu	NOUN
ap-8720	228	48	conv2d	conv2d	VERB
ap-8720	228	49	(	(	PUNCT
ap-8720	228	50	14	14	NUM
ap-8720	228	51	,	,	PUNCT
ap-8720	228	52	14	14	NUM
ap-8720	228	53	,	,	PUNCT
ap-8720	228	54	256	256	NUM
ap-8720	228	55	)	)	PUNCT
ap-8720	228	56	4	4	NUM
ap-8720	228	57	2	2	NUM
ap-8720	228	58	leakyrelu	leakyrelu	NOUN
ap-8720	228	59	conv2d	conv2d	VERB
ap-8720	228	60	(	(	PUNCT
ap-8720	228	61	6	6	NUM
ap-8720	228	62	,	,	PUNCT
ap-8720	228	63	6	6	NUM
ap-8720	228	64	,	,	PUNCT
ap-8720	228	65	256	256	NUM
ap-8720	228	66	)	)	PUNCT
ap-8720	228	67	4	4	NUM
ap-8720	228	68	2	2	NUM
ap-8720	228	69	leakyrelu	leakyrelu	NOUN
ap-8720	228	70	dense	dense	ADJ
ap-8720	228	71	1	1	NUM
ap-8720	228	72	–	–	PUNCT
ap-8720	228	73	–	–	PUNCT
ap-8720	228	74	sigmoid	sigmoid	NOUN
ap-8720	228	75	table	table	NOUN
ap-8720	228	76	2	2	NUM
ap-8720	228	77	.	.	PUNCT
ap-8720	228	78	architecture	architecture	NOUN
ap-8720	228	79	of	of	ADP
ap-8720	228	80	discriminator	discriminator	NOUN
ap-8720	228	81	1	1	NUM
ap-8720	228	82	.	.	PUNCT
ap-8720	228	83	figure	figure	VERB
ap-8720	228	84	6	6	NUM
ap-8720	228	85	.	.	PUNCT
ap-8720	229	1	architecture	architecture	NOUN
ap-8720	229	2	of	of	ADP
ap-8720	229	3	the	the	DET
ap-8720	229	4	proposed	propose	VERB
ap-8720	229	5	ndcgan	ndcgan	PROPN
ap-8720	229	6	discriminator	discriminator	NOUN
ap-8720	229	7	.	.	PUNCT
ap-8720	230	1	311	311	NUM
ap-8720	230	2	ch	ch	NOUN
ap-8720	230	3	.	.	PUNCT
ap-8720	230	4	e.	e.	PROPN
ap-8720	230	5	bencheriet	bencheriet	PROPN
ap-8720	230	6	,	,	PUNCT
ap-8720	230	7	h.	h.	PROPN
ap-8720	230	8	abdelmoumène	abdelmoumène	PROPN
ap-8720	230	9	,	,	PUNCT
ap-8720	230	10	a.	a.	PROPN
ap-8720	230	11	sebbagh	sebbagh	PROPN
ap-8720	230	12	et	et	PROPN
ap-8720	230	13	al	al	PROPN
ap-8720	230	14	.	.	PROPN
ap-8720	230	15	acta	acta	PROPN
ap-8720	230	16	polytechnica	polytechnica	PROPN
ap-8720	230	17	(	(	PUNCT
ap-8720	230	18	a	a	NOUN
ap-8720	230	19	)	)	PUNCT
ap-8720	230	20	.	.	PUNCT
ap-8720	231	1	relu	relu	NOUN
ap-8720	231	2	.	.	PUNCT
ap-8720	232	1	(	(	PUNCT
ap-8720	232	2	b	b	NOUN
ap-8720	232	3	)	)	PUNCT
ap-8720	232	4	.	.	PUNCT
ap-8720	233	1	leakyrelu	leakyrelu	PROPN
ap-8720	233	2	.	.	PUNCT
ap-8720	234	1	figure	figure	VERB
ap-8720	234	2	7	7	NUM
ap-8720	234	3	.	.	PUNCT
ap-8720	234	4	activation	activation	NOUN
ap-8720	234	5	function	function	NOUN
ap-8720	234	6	.	.	PUNCT
ap-8720	235	1	is	be	AUX
ap-8720	235	2	controlled	control	VERB
ap-8720	235	3	by	by	ADP
ap-8720	235	4	a	a	DET
ap-8720	235	5	parameter	parameter	NOUN
ap-8720	235	6	called	call	VERB
ap-8720	235	7	alpha	alpha	NOUN
ap-8720	235	8	,	,	PUNCT
ap-8720	235	9	which	which	PRON
ap-8720	235	10	introduces	introduce	VERB
ap-8720	235	11	network	network	NOUN
ap-8720	235	12	tolerance	tolerance	NOUN
ap-8720	235	13	by	by	ADP
ap-8720	235	14	allowing	allow	VERB
ap-8720	235	15	certain	certain	ADJ
ap-8720	235	16	negative	negative	ADJ
ap-8720	235	17	values	value	NOUN
ap-8720	235	18	to	to	PART
ap-8720	235	19	pass	pass	VERB
ap-8720	235	20	.	.	PUNCT
ap-8720	236	1	4	4	X
ap-8720	236	2	.	.	X
ap-8720	236	3	experimental	experimental	ADJ
ap-8720	236	4	results	result	NOUN
ap-8720	236	5	in	in	ADP
ap-8720	236	6	this	this	DET
ap-8720	236	7	section	section	NOUN
ap-8720	236	8	,	,	PUNCT
ap-8720	236	9	we	we	PRON
ap-8720	236	10	provide	provide	VERB
ap-8720	236	11	an	an	DET
ap-8720	236	12	overview	overview	NOUN
ap-8720	236	13	of	of	ADP
ap-8720	236	14	the	the	DET
ap-8720	236	15	network	network	NOUN
ap-8720	236	16	training	training	NOUN
ap-8720	236	17	and	and	CCONJ
ap-8720	236	18	testing	testing	NOUN
ap-8720	236	19	process	process	NOUN
ap-8720	236	20	.	.	PUNCT
ap-8720	237	1	for	for	ADP
ap-8720	237	2	our	our	PRON
ap-8720	237	3	purposes	purpose	NOUN
ap-8720	237	4	,	,	PUNCT
ap-8720	237	5	we	we	PRON
ap-8720	237	6	chose	choose	VERB
ap-8720	237	7	to	to	PART
ap-8720	237	8	use	use	VERB
ap-8720	237	9	kaggle	kaggle	VERB
ap-8720	237	10	[	[	X
ap-8720	237	11	33	33	NUM
ap-8720	237	12	]	]	PUNCT
ap-8720	237	13	rather	rather	ADV
ap-8720	237	14	than	than	ADP
ap-8720	237	15	google	google	PROPN
ap-8720	237	16	colab	colab	PROPN
ap-8720	238	1	[	[	X
ap-8720	238	2	34	34	NUM
ap-8720	238	3	]	]	PUNCT
ap-8720	238	4	.	.	PUNCT
ap-8720	239	1	we	we	PRON
ap-8720	239	2	made	make	VERB
ap-8720	239	3	this	this	DET
ap-8720	239	4	choice	choice	NOUN
ap-8720	239	5	because	because	SCONJ
ap-8720	239	6	of	of	ADP
ap-8720	239	7	kaggle	kaggle	PROPN
ap-8720	239	8	’s	’s	PART
ap-8720	239	9	extensive	extensive	ADJ
ap-8720	239	10	collection	collection	NOUN
ap-8720	239	11	of	of	ADP
ap-8720	239	12	importable	importable	ADJ
ap-8720	239	13	datasets	dataset	NOUN
ap-8720	239	14	and	and	CCONJ
ap-8720	239	15	the	the	DET
ap-8720	239	16	availability	availability	NOUN
ap-8720	239	17	of	of	ADP
ap-8720	239	18	up	up	ADP
ap-8720	239	19	to	to	PART
ap-8720	239	20	forty	forty	NUM
ap-8720	239	21	gpu	gpu	PROPN
ap-8720	239	22	/	/	SYM
ap-8720	239	23	tpu	tpu	NOUN
ap-8720	239	24	hours	hour	NOUN
ap-8720	239	25	per	per	ADP
ap-8720	239	26	week	week	NOUN
ap-8720	239	27	,	,	PUNCT
ap-8720	239	28	all	all	PRON
ap-8720	239	29	provided	provide	VERB
ap-8720	239	30	free	free	ADJ
ap-8720	239	31	of	of	ADP
ap-8720	239	32	charge	charge	NOUN
ap-8720	239	33	.	.	PUNCT
ap-8720	240	1	in	in	ADP
ap-8720	240	2	contrast	contrast	NOUN
ap-8720	240	3	,	,	PUNCT
ap-8720	240	4	google	google	PROPN
ap-8720	240	5	colab	colab	PROPN
ap-8720	240	6	does	do	AUX
ap-8720	240	7	not	not	PART
ap-8720	240	8	offer	offer	VERB
ap-8720	240	9	the	the	DET
ap-8720	240	10	same	same	ADJ
ap-8720	240	11	level	level	NOUN
ap-8720	240	12	of	of	ADP
ap-8720	240	13	resource	resource	NOUN
ap-8720	240	14	availability	availability	NOUN
ap-8720	240	15	at	at	ADP
ap-8720	240	16	no	no	DET
ap-8720	240	17	additional	additional	ADJ
ap-8720	240	18	cost	cost	NOUN
ap-8720	240	19	.	.	PUNCT
ap-8720	241	1	to	to	PART
ap-8720	241	2	train	train	VERB
ap-8720	241	3	our	our	PRON
ap-8720	241	4	model	model	NOUN
ap-8720	241	5	,	,	PUNCT
ap-8720	241	6	we	we	PRON
ap-8720	241	7	used	use	VERB
ap-8720	241	8	the	the	DET
ap-8720	241	9	tesla	tesla	PROPN
ap-8720	241	10	p100	p100	PROPN
ap-8720	241	11	gpus	gpu	NOUN
ap-8720	241	12	available	available	ADJ
ap-8720	241	13	on	on	ADP
ap-8720	241	14	kaggle	kaggle	NOUN
ap-8720	241	15	as	as	ADP
ap-8720	241	16	our	our	PRON
ap-8720	241	17	training	training	NOUN
ap-8720	241	18	environment	environment	NOUN
ap-8720	241	19	,	,	PUNCT
ap-8720	241	20	along	along	ADP
ap-8720	241	21	with	with	ADP
ap-8720	241	22	tensorflow	tensorflow	NOUN
ap-8720	241	23	;	;	PUNCT
ap-8720	241	24	these	these	DET
ap-8720	241	25	gpus	gpu	NOUN
ap-8720	241	26	are	be	AUX
ap-8720	241	27	useful	useful	ADJ
ap-8720	241	28	for	for	ADP
ap-8720	241	29	training	train	VERB
ap-8720	241	30	deep	deep	ADJ
ap-8720	241	31	learning	learning	NOUN
ap-8720	241	32	models	model	NOUN
ap-8720	241	33	and	and	CCONJ
ap-8720	241	34	can	can	AUX
ap-8720	241	35	speed	speed	VERB
ap-8720	241	36	up	up	ADP
ap-8720	241	37	training	training	NOUN
ap-8720	241	38	13	13	NUM
ap-8720	241	39	times	time	NOUN
ap-8720	241	40	faster	fast	ADV
ap-8720	241	41	than	than	ADP
ap-8720	241	42	traditional	traditional	ADJ
ap-8720	241	43	training	training	NOUN
ap-8720	241	44	on	on	ADP
ap-8720	241	45	a	a	DET
ap-8720	241	46	cpu	cpu	NOUN
ap-8720	241	47	.	.	PUNCT
ap-8720	242	1	the	the	DET
ap-8720	242	2	specifications	specification	NOUN
ap-8720	242	3	of	of	ADP
ap-8720	242	4	cpu	cpu	NOUN
ap-8720	242	5	runtime	runtime	NOUN
ap-8720	242	6	provided	provide	VERB
ap-8720	242	7	by	by	ADP
ap-8720	242	8	kaggle	kaggle	PROPN
ap-8720	242	9	are	be	AUX
ap-8720	242	10	intel	intel	PROPN
ap-8720	242	11	xeon	xeon	PROPN
ap-8720	242	12	processor	processor	NOUN
ap-8720	242	13	with	with	ADP
ap-8720	242	14	two	two	NUM
ap-8720	242	15	cores	core	NOUN
ap-8720	242	16	@2.30	@2.30	NOUN
ap-8720	242	17	ghz	ghz	NOUN
ap-8720	242	18	and	and	CCONJ
ap-8720	242	19	16	16	NUM
ap-8720	242	20	gb	gb	NOUN
ap-8720	242	21	ram	ram	NOUN
ap-8720	242	22	,	,	PUNCT
ap-8720	242	23	and	and	CCONJ
ap-8720	242	24	17	17	NUM
ap-8720	242	25	gpu	gpu	NOUN
ap-8720	242	26	ram	ram	NOUN
ap-8720	242	27	.	.	PUNCT
ap-8720	243	1	python	python	PROPN
ap-8720	243	2	was	be	AUX
ap-8720	243	3	used	use	VERB
ap-8720	243	4	as	as	ADP
ap-8720	243	5	the	the	DET
ap-8720	243	6	programming	programming	NOUN
ap-8720	243	7	language	language	NOUN
ap-8720	243	8	and	and	CCONJ
ap-8720	243	9	pycharm	pycharm	PROPN
ap-8720	243	10	was	be	AUX
ap-8720	243	11	used	use	VERB
ap-8720	243	12	as	as	ADP
ap-8720	243	13	the	the	DET
ap-8720	243	14	ide	ide	NOUN
ap-8720	243	15	for	for	ADP
ap-8720	243	16	coding	coding	NOUN
ap-8720	243	17	and	and	CCONJ
ap-8720	243	18	development	development	NOUN
ap-8720	243	19	.	.	PUNCT
ap-8720	244	1	after	after	ADP
ap-8720	244	2	network	network	NOUN
ap-8720	244	3	training	training	NOUN
ap-8720	244	4	and	and	CCONJ
ap-8720	244	5	validation	validation	NOUN
ap-8720	244	6	,	,	PUNCT
ap-8720	244	7	the	the	DET
ap-8720	244	8	trained	train	VERB
ap-8720	244	9	model	model	NOUN
ap-8720	244	10	is	be	AUX
ap-8720	244	11	imported	import	VERB
ap-8720	244	12	to	to	PART
ap-8720	244	13	run	run	VERB
ap-8720	244	14	all	all	DET
ap-8720	244	15	the	the	DET
ap-8720	244	16	tests	test	NOUN
ap-8720	244	17	on	on	ADP
ap-8720	244	18	a	a	DET
ap-8720	244	19	laptop	laptop	NOUN
ap-8720	244	20	with	with	ADP
ap-8720	244	21	an	an	DET
ap-8720	244	22	intel	intel	NOUN
ap-8720	244	23	core	core	NOUN
ap-8720	244	24	i7	i7	NOUN
ap-8720	244	25	and	and	CCONJ
ap-8720	244	26	16	16	NUM
ap-8720	244	27	go	go	NOUN
ap-8720	244	28	ram	ram	NOUN
ap-8720	244	29	.	.	PUNCT
ap-8720	245	1	4.1	4.1	NUM
ap-8720	245	2	.	.	PUNCT
ap-8720	245	3	dataset	dataset	VERB
ap-8720	245	4	in	in	ADP
ap-8720	245	5	the	the	DET
ap-8720	245	6	training	training	NOUN
ap-8720	245	7	phase	phase	NOUN
ap-8720	245	8	,	,	PUNCT
ap-8720	245	9	we	we	PRON
ap-8720	245	10	used	use	VERB
ap-8720	245	11	a	a	DET
ap-8720	245	12	database	database	NOUN
ap-8720	245	13	of	of	ADP
ap-8720	245	14	70,000	70,000	NUM
ap-8720	245	15	real	real	ADJ
ap-8720	245	16	faces	face	NOUN
ap-8720	245	17	from	from	ADP
ap-8720	245	18	the	the	DET
ap-8720	245	19	flickr	flickr	PROPN
ap-8720	245	20	-	-	PUNCT
ap-8720	245	21	faces	face	NOUN
ap-8720	245	22	-	-	PUNCT
ap-8720	245	23	hq	hq	NOUN
ap-8720	245	24	dataset	dataset	NOUN
ap-8720	245	25	[	[	X
ap-8720	245	26	18	18	NUM
ap-8720	245	27	]	]	PUNCT
ap-8720	245	28	(	(	PUNCT
ap-8720	245	29	figure	figure	NOUN
ap-8720	245	30	8)	8)	NUM
ap-8720	245	31	,	,	PUNCT
ap-8720	245	32	collected	collect	VERB
ap-8720	245	33	by	by	ADP
ap-8720	245	34	nvidia	nvidia	PROPN
ap-8720	245	35	,	,	PUNCT
ap-8720	245	36	and	and	CCONJ
ap-8720	245	37	70	70	NUM
ap-8720	245	38	000	000	NUM
ap-8720	245	39	fake	fake	ADJ
ap-8720	245	40	faces	face	NOUN
ap-8720	245	41	(	(	PUNCT
ap-8720	245	42	figure	figure	NOUN
ap-8720	245	43	9	9	NUM
ap-8720	245	44	)	)	PUNCT
ap-8720	245	45	,	,	PUNCT
ap-8720	245	46	generated	generate	VERB
ap-8720	245	47	using	use	VERB
ap-8720	245	48	nvidia	nvidia	PROPN
ap-8720	245	49	’s	’s	PART
ap-8720	245	50	stylegan	stylegan	NOUN
ap-8720	245	51	[	[	X
ap-8720	245	52	19	19	NUM
ap-8720	245	53	]	]	PUNCT
ap-8720	245	54	.	.	PUNCT
ap-8720	246	1	in	in	ADP
ap-8720	246	2	this	this	DET
ap-8720	246	3	dataset	dataset	NOUN
ap-8720	246	4	,	,	PUNCT
ap-8720	246	5	all	all	DET
ap-8720	246	6	images	image	NOUN
ap-8720	246	7	have	have	VERB
ap-8720	246	8	a	a	DET
ap-8720	246	9	dimension	dimension	NOUN
ap-8720	246	10	of	of	ADP
ap-8720	246	11	256	256	NUM
ap-8720	246	12	px	px	NOUN
ap-8720	246	13	and	and	CCONJ
ap-8720	246	14	are	be	AUX
ap-8720	246	15	divided	divide	VERB
ap-8720	246	16	into	into	ADP
ap-8720	246	17	training	training	NOUN
ap-8720	246	18	,	,	PUNCT
ap-8720	246	19	validation	validation	NOUN
ap-8720	246	20	,	,	PUNCT
ap-8720	246	21	and	and	CCONJ
ap-8720	246	22	test	test	NOUN
ap-8720	246	23	sets	set	NOUN
ap-8720	246	24	.	.	PUNCT
ap-8720	247	1	4.2	4.2	NUM
ap-8720	247	2	.	.	PUNCT
ap-8720	247	3	evaluation	evaluation	NOUN
ap-8720	247	4	metrics	metric	NOUN
ap-8720	247	5	we	we	PRON
ap-8720	247	6	have	have	AUX
ap-8720	247	7	implemented	implement	VERB
ap-8720	247	8	the	the	DET
ap-8720	247	9	discriminators	discriminator	NOUN
ap-8720	247	10	of	of	ADP
ap-8720	247	11	our	our	PRON
ap-8720	247	12	proposed	propose	VERB
ap-8720	247	13	system	system	NOUN
ap-8720	247	14	to	to	PART
ap-8720	247	15	compare	compare	VERB
ap-8720	247	16	the	the	DET
ap-8720	247	17	overall	overall	ADJ
ap-8720	247	18	accuracy	accuracy	NOUN
ap-8720	247	19	and	and	CCONJ
ap-8720	247	20	loss	loss	NOUN
ap-8720	247	21	.	.	PUNCT
ap-8720	248	1	additionally	additionally	ADV
ap-8720	248	2	,	,	PUNCT
ap-8720	248	3	to	to	PART
ap-8720	248	4	evaluate	evaluate	VERB
ap-8720	248	5	the	the	DET
ap-8720	248	6	effectiveness	effectiveness	NOUN
ap-8720	248	7	of	of	ADP
ap-8720	248	8	our	our	PRON
ap-8720	248	9	proposed	propose	VERB
ap-8720	248	10	architecture	architecture	NOUN
ap-8720	248	11	(	(	PUNCT
ap-8720	248	12	mdd	mdd	PROPN
ap-8720	248	13	-	-	PUNCT
ap-8720	248	14	cnn	cnn	PROPN
ap-8720	248	15	)	)	PUNCT
ap-8720	248	16	,	,	PUNCT
ap-8720	248	17	the	the	DET
ap-8720	248	18	following	follow	VERB
ap-8720	248	19	conventional	conventional	ADJ
ap-8720	248	20	evaluation	evaluation	NOUN
ap-8720	248	21	metrics	metric	NOUN
ap-8720	248	22	were	be	AUX
ap-8720	248	23	used	use	VERB
ap-8720	248	24	:	:	PUNCT
ap-8720	248	25	true	true	ADJ
ap-8720	248	26	positives	positive	NOUN
ap-8720	248	27	(	(	PUNCT
ap-8720	248	28	tp	tp	NOUN
ap-8720	248	29	)	)	PUNCT
ap-8720	248	30	,	,	PUNCT
ap-8720	248	31	which	which	PRON
ap-8720	248	32	refer	refer	VERB
ap-8720	248	33	to	to	ADP
ap-8720	248	34	examples	example	NOUN
ap-8720	248	35	correctly	correctly	ADV
ap-8720	248	36	labelled	label	VERB
ap-8720	248	37	as	as	ADP
ap-8720	248	38	positive	positive	ADJ
ap-8720	248	39	;	;	PUNCT
ap-8720	248	40	true	true	ADJ
ap-8720	248	41	negatives	negative	NOUN
ap-8720	248	42	(	(	PUNCT
ap-8720	248	43	tn	tn	NOUN
ap-8720	248	44	)	)	PUNCT
ap-8720	248	45	,	,	PUNCT
ap-8720	248	46	which	which	PRON
ap-8720	248	47	refer	refer	VERB
ap-8720	248	48	to	to	ADP
ap-8720	248	49	negative	negative	ADJ
ap-8720	248	50	examples	example	NOUN
ap-8720	248	51	correctly	correctly	ADV
ap-8720	248	52	labelled	label	VERB
ap-8720	248	53	as	as	ADP
ap-8720	248	54	negative	negative	ADJ
ap-8720	248	55	false	false	ADJ
ap-8720	248	56	positives	positive	NOUN
ap-8720	248	57	(	(	PUNCT
ap-8720	248	58	fp	fp	NOUN
ap-8720	248	59	)	)	PUNCT
ap-8720	248	60	,	,	PUNCT
ap-8720	248	61	which	which	PRON
ap-8720	248	62	refer	refer	VERB
ap-8720	248	63	to	to	ADP
ap-8720	248	64	negative	negative	ADJ
ap-8720	248	65	examples	example	NOUN
ap-8720	248	66	mislabelled	mislabelle	VERB
ap-8720	248	67	as	as	ADP
ap-8720	248	68	positive	positive	ADJ
ap-8720	248	69	;	;	PUNCT
ap-8720	248	70	figure	figure	VERB
ap-8720	248	71	8	8	NUM
ap-8720	248	72	.	.	PUNCT
ap-8720	249	1	flickr	flickr	PROPN
ap-8720	249	2	-	-	PUNCT
ap-8720	249	3	face	face	NOUN
ap-8720	249	4	-	-	PUNCT
ap-8720	249	5	hq	hq	NOUN
ap-8720	249	6	(	(	PUNCT
ap-8720	249	7	ffhq	ffhq	NOUN
ap-8720	249	8	)	)	PUNCT
ap-8720	249	9	dataset	dataset	NOUN
ap-8720	249	10	.	.	PUNCT
ap-8720	250	1	figure	figure	NOUN
ap-8720	250	2	9	9	NUM
ap-8720	250	3	.	.	NOUN
ap-8720	250	4	example	example	NOUN
ap-8720	250	5	of	of	ADP
ap-8720	250	6	fake	fake	ADJ
ap-8720	250	7	face	face	NOUN
ap-8720	250	8	dataset	dataset	NOUN
ap-8720	250	9	.	.	PUNCT
ap-8720	251	1	false	false	ADJ
ap-8720	251	2	negatives	negative	NOUN
ap-8720	251	3	(	(	PUNCT
ap-8720	251	4	fn	fn	NOUN
ap-8720	251	5	)	)	PUNCT
ap-8720	251	6	,	,	PUNCT
ap-8720	251	7	which	which	PRON
ap-8720	251	8	refer	refer	VERB
ap-8720	251	9	to	to	ADP
ap-8720	251	10	positive	positive	ADJ
ap-8720	251	11	examples	example	NOUN
ap-8720	251	12	mislabelled	mislabelle	VERB
ap-8720	251	13	as	as	ADP
ap-8720	251	14	negative	negative	ADJ
ap-8720	251	15	.	.	PUNCT
ap-8720	252	1	accuracy	accuracy	NOUN
ap-8720	252	2	(	(	PUNCT
ap-8720	252	3	precision	precision	NOUN
ap-8720	252	4	)	)	PUNCT
ap-8720	252	5	given	give	VERB
ap-8720	252	6	by	by	ADP
ap-8720	252	7	equation	equation	NOUN
ap-8720	252	8	(	(	PUNCT
ap-8720	252	9	3	3	NUM
ap-8720	252	10	):	):	PUNCT
ap-8720	252	11	is	be	AUX
ap-8720	252	12	the	the	DET
ap-8720	252	13	proportion	proportion	NOUN
ap-8720	252	14	of	of	ADP
ap-8720	252	15	observations	observation	NOUN
ap-8720	252	16	correctly	correctly	ADV
ap-8720	252	17	predicted	predict	VERB
ap-8720	252	18	to	to	ADP
ap-8720	252	19	the	the	DET
ap-8720	252	20	total	total	ADJ
ap-8720	252	21	number	number	NOUN
ap-8720	252	22	of	of	ADP
ap-8720	252	23	observations	observation	NOUN
ap-8720	252	24	.	.	PUNCT
ap-8720	253	1	accuracy	accuracy	NOUN
ap-8720	254	1	=	=	PRON
ap-8720	254	2	tp	tp	X
ap-8720	254	3	+	+	CCONJ
ap-8720	254	4	tn	tn	NOUN
ap-8720	254	5	tp	tp	NOUN
ap-8720	254	6	+	+	CCONJ
ap-8720	254	7	fp	fp	PROPN
ap-8720	254	8	+	+	NUM
ap-8720	254	9	tn	tn	PROPN
ap-8720	255	1	+	+	X
ap-8720	255	2	fn	fn	NOUN
ap-8720	255	3	.	.	PUNCT
ap-8720	256	1	(	(	PUNCT
ap-8720	256	2	3	3	X
ap-8720	256	3	)	)	PUNCT
ap-8720	256	4	loss	loss	NOUN
ap-8720	256	5	of	of	ADP
ap-8720	256	6	discriminator	discriminator	NOUN
ap-8720	256	7	:	:	PUNCT
ap-8720	256	8	is	be	AUX
ap-8720	256	9	the	the	DET
ap-8720	256	10	difference	difference	NOUN
ap-8720	256	11	between	between	ADP
ap-8720	256	12	the	the	DET
ap-8720	256	13	predicted	predict	VERB
ap-8720	256	14	output	output	NOUN
ap-8720	256	15	and	and	CCONJ
ap-8720	256	16	the	the	DET
ap-8720	256	17	actual	actual	ADJ
ap-8720	256	18	output	output	NOUN
ap-8720	256	19	.	.	PUNCT
ap-8720	257	1	it	it	PRON
ap-8720	257	2	measures	measure	VERB
ap-8720	257	3	the	the	DET
ap-8720	257	4	mistakes	mistake	NOUN
ap-8720	257	5	made	make	VERB
ap-8720	257	6	by	by	ADP
ap-8720	257	7	the	the	DET
ap-8720	257	8	network	network	NOUN
ap-8720	257	9	in	in	ADP
ap-8720	257	10	predicting	predict	VERB
ap-8720	257	11	the	the	DET
ap-8720	257	12	output	output	NOUN
ap-8720	257	13	.	.	PUNCT
ap-8720	258	1	truepositiverate	truepositiverate	ADJ
ap-8720	258	2	/	/	SYM
ap-8720	258	3	recall	recall	NOUN
ap-8720	258	4	/	/	SYM
ap-8720	258	5	sensitivity	sensitivity	NOUN
ap-8720	258	6	:	:	PUNCT
ap-8720	258	7	is	be	AUX
ap-8720	258	8	the	the	DET
ap-8720	258	9	proportion	proportion	NOUN
ap-8720	258	10	of	of	ADP
ap-8720	258	11	actual	actual	ADJ
ap-8720	258	12	positive	positive	ADJ
ap-8720	258	13	results	result	NOUN
ap-8720	258	14	that	that	PRON
ap-8720	258	15	were	be	AUX
ap-8720	258	16	correctly	correctly	ADV
ap-8720	258	17	identified	identify	VERB
ap-8720	258	18	given	give	VERB
ap-8720	258	19	by	by	ADP
ap-8720	258	20	formula	formula	NOUN
ap-8720	258	21	(	(	PUNCT
ap-8720	258	22	4	4	NUM
ap-8720	258	23	):	):	PUNCT
ap-8720	258	24	tpr	tpr	NOUN
ap-8720	258	25	=	=	NOUN
ap-8720	258	26	recall	recall	NOUN
ap-8720	258	27	=	=	PROPN
ap-8720	258	28	tp	tp	NOUN
ap-8720	258	29	tp	tp	NOUN
ap-8720	258	30	+	+	CCONJ
ap-8720	258	31	fn	fn	INTJ
ap-8720	258	32	.	.	PUNCT
ap-8720	259	1	(	(	PUNCT
ap-8720	259	2	4	4	X
ap-8720	259	3	)	)	PUNCT
ap-8720	259	4	receiver	receiver	NOUN
ap-8720	259	5	operating	operate	VERB
ap-8720	259	6	characteristic	characteristic	ADJ
ap-8720	259	7	curve	curve	NOUN
ap-8720	259	8	(	(	PUNCT
ap-8720	259	9	roc	roc	PROPN
ap-8720	259	10	curve	curve	NOUN
ap-8720	259	11	)	)	PUNCT
ap-8720	259	12	is	be	AUX
ap-8720	259	13	a	a	DET
ap-8720	259	14	graph	graph	NOUN
ap-8720	259	15	that	that	PRON
ap-8720	259	16	shows	show	VERB
ap-8720	259	17	the	the	DET
ap-8720	259	18	performance	performance	NOUN
ap-8720	259	19	of	of	ADP
ap-8720	259	20	a	a	DET
ap-8720	259	21	classification	classification	NOUN
ap-8720	259	22	model	model	NOUN
ap-8720	259	23	at	at	ADP
ap-8720	259	24	all	all	DET
ap-8720	259	25	classification	classification	NOUN
ap-8720	259	26	thresholds	threshold	NOUN
ap-8720	259	27	.	.	PUNCT
ap-8720	260	1	312	312	NUM
ap-8720	260	2	vol	vol	NOUN
ap-8720	260	3	.	.	PUNCT
ap-8720	260	4	63	63	NUM
ap-8720	260	5	no	no	NOUN
ap-8720	260	6	.	.	PUNCT
ap-8720	261	1	5/2023	5/2023	NUM
ap-8720	261	2	fake	fake	ADJ
ap-8720	261	3	face	face	NOUN
ap-8720	261	4	detection	detection	NOUN
ap-8720	261	5	based	base	VERB
ap-8720	261	6	on	on	ADP
ap-8720	261	7	a	a	DET
ap-8720	261	8	multi	multi	ADJ
ap-8720	261	9	discriminator	discriminator	NOUN
ap-8720	261	10	deep	deep	PROPN
ap-8720	261	11	cnn	cnn	PROPN
ap-8720	261	12	architecture	architecture	NOUN
ap-8720	261	13	the	the	DET
ap-8720	261	14	curve	curve	NOUN
ap-8720	261	15	represents	represent	VERB
ap-8720	261	16	two	two	NUM
ap-8720	261	17	parameters	parameter	NOUN
ap-8720	261	18	:	:	PUNCT
ap-8720	261	19	true	true	ADJ
ap-8720	261	20	positive	positive	ADJ
ap-8720	261	21	rate	rate	NOUN
ap-8720	261	22	(	(	PUNCT
ap-8720	261	23	tpr	tpr	NOUN
ap-8720	261	24	)	)	PUNCT
ap-8720	261	25	(	(	PUNCT
ap-8720	261	26	equation	equation	NOUN
ap-8720	261	27	(	(	PUNCT
ap-8720	261	28	4	4	NUM
ap-8720	261	29	)	)	PUNCT
ap-8720	261	30	)	)	PUNCT
ap-8720	261	31	and	and	CCONJ
ap-8720	261	32	false	false	ADJ
ap-8720	261	33	positive	positive	ADJ
ap-8720	261	34	rate	rate	NOUN
ap-8720	261	35	(	(	PUNCT
ap-8720	261	36	fpr	fpr	NOUN
ap-8720	261	37	):	):	PUNCT
ap-8720	261	38	fpr	fpr	X
ap-8720	261	39	=	=	PUNCT
ap-8720	261	40	fp	fp	PROPN
ap-8720	261	41	fp	fp	X
ap-8720	262	1	+	+	PROPN
ap-8720	262	2	v	v	NOUN
ap-8720	262	3	n	n	NOUN
ap-8720	262	4	.	.	PUNCT
ap-8720	263	1	(	(	PUNCT
ap-8720	263	2	5	5	NUM
ap-8720	263	3	)	)	PUNCT
ap-8720	263	4	4.3	4.3	NUM
ap-8720	263	5	.	.	PUNCT
ap-8720	264	1	training	training	NOUN
ap-8720	264	2	and	and	CCONJ
ap-8720	264	3	validation	validation	NOUN
ap-8720	264	4	the	the	DET
ap-8720	264	5	proposed	propose	VERB
ap-8720	264	6	multi	multi	ADJ
ap-8720	264	7	-	-	ADJ
ap-8720	264	8	discriminator	discriminator	ADJ
ap-8720	264	9	architecture	architecture	NOUN
ap-8720	264	10	(	(	PUNCT
ap-8720	264	11	mddcnn	mddcnn	PROPN
ap-8720	264	12	)	)	PUNCT
ap-8720	264	13	requires	require	VERB
ap-8720	264	14	training	training	NOUN
ap-8720	264	15	in	in	ADP
ap-8720	264	16	three	three	NUM
ap-8720	264	17	steps	step	NOUN
ap-8720	264	18	,	,	PUNCT
ap-8720	264	19	with	with	ADP
ap-8720	264	20	each	each	DET
ap-8720	264	21	discriminator	discriminator	NOUN
ap-8720	264	22	undergoing	undergo	VERB
ap-8720	264	23	a	a	DET
ap-8720	264	24	different	different	ADJ
ap-8720	264	25	training	training	NOUN
ap-8720	264	26	.	.	PUNCT
ap-8720	265	1	discriminator	discriminator	NOUN
ap-8720	265	2	1	1	NUM
ap-8720	265	3	undergoes	undergo	VERB
ap-8720	265	4	adversarial	adversarial	ADJ
ap-8720	265	5	training	training	NOUN
ap-8720	265	6	,	,	PUNCT
ap-8720	265	7	discriminator	discriminator	NOUN
ap-8720	265	8	2	2	NUM
ap-8720	265	9	undergoes	undergoe	NOUN
ap-8720	265	10	training	training	NOUN
ap-8720	265	11	by	by	ADP
ap-8720	265	12	transfer	transfer	NOUN
ap-8720	265	13	learning	learning	NOUN
ap-8720	265	14	,	,	PUNCT
ap-8720	265	15	and	and	CCONJ
ap-8720	265	16	discriminator	discriminator	NOUN
ap-8720	265	17	3	3	NUM
ap-8720	265	18	undergoes	undergo	VERB
ap-8720	265	19	a	a	DET
ap-8720	265	20	standard	standard	ADJ
ap-8720	265	21	training	training	NOUN
ap-8720	265	22	of	of	ADP
ap-8720	265	23	a	a	DET
ap-8720	265	24	cnn	cnn	NOUN
ap-8720	265	25	with	with	ADP
ap-8720	265	26	examples	example	NOUN
ap-8720	265	27	and	and	CCONJ
ap-8720	265	28	counterexamples	counterexample	NOUN
ap-8720	265	29	.	.	PUNCT
ap-8720	266	1	during	during	ADP
ap-8720	266	2	training	training	NOUN
ap-8720	266	3	,	,	PUNCT
ap-8720	266	4	certain	certain	ADJ
ap-8720	266	5	parameters	parameter	NOUN
ap-8720	266	6	that	that	PRON
ap-8720	266	7	are	be	AUX
ap-8720	266	8	crucial	crucial	ADJ
ap-8720	266	9	for	for	SCONJ
ap-8720	266	10	convergence	convergence	NOUN
ap-8720	266	11	to	to	ADP
ap-8720	266	12	the	the	DET
ap-8720	266	13	optimal	optimal	ADJ
ap-8720	266	14	solution	solution	NOUN
ap-8720	266	15	were	be	AUX
ap-8720	266	16	fixed	fix	VERB
ap-8720	266	17	after	after	ADP
ap-8720	266	18	a	a	DET
ap-8720	266	19	series	series	NOUN
ap-8720	266	20	of	of	ADP
ap-8720	266	21	tests	test	NOUN
ap-8720	266	22	.	.	PUNCT
ap-8720	267	1	these	these	DET
ap-8720	267	2	parameters	parameter	NOUN
ap-8720	267	3	include	include	VERB
ap-8720	267	4	the	the	DET
ap-8720	267	5	batch	batch	NOUN
ap-8720	267	6	size	size	NOUN
ap-8720	267	7	and	and	CCONJ
ap-8720	267	8	the	the	DET
ap-8720	267	9	number	number	NOUN
ap-8720	267	10	of	of	ADP
ap-8720	267	11	epochs	epoch	NOUN
ap-8720	267	12	.	.	PUNCT
ap-8720	267	13	batch	batch	NOUN
ap-8720	267	14	size	size	NOUN
ap-8720	267	15	:	:	PUNCT
ap-8720	267	16	the	the	DET
ap-8720	267	17	batch	batch	NOUN
ap-8720	267	18	size	size	NOUN
ap-8720	267	19	refers	refer	VERB
ap-8720	267	20	to	to	ADP
ap-8720	267	21	the	the	DET
ap-8720	267	22	number	number	NOUN
ap-8720	267	23	of	of	ADP
ap-8720	267	24	training	training	NOUN
ap-8720	267	25	examples	example	NOUN
ap-8720	267	26	presented	present	VERB
ap-8720	267	27	to	to	ADP
ap-8720	267	28	the	the	DET
ap-8720	267	29	model	model	NOUN
ap-8720	267	30	in	in	ADP
ap-8720	267	31	each	each	DET
ap-8720	267	32	iteration	iteration	NOUN
ap-8720	267	33	.	.	PUNCT
ap-8720	268	1	it	it	PRON
ap-8720	268	2	has	have	VERB
ap-8720	268	3	a	a	DET
ap-8720	268	4	significant	significant	ADJ
ap-8720	268	5	influence	influence	NOUN
ap-8720	268	6	on	on	ADP
ap-8720	268	7	the	the	DET
ap-8720	268	8	training	training	NOUN
ap-8720	268	9	process	process	NOUN
ap-8720	268	10	.	.	PUNCT
ap-8720	269	1	•	•	NUM
ap-8720	269	2	computational	computational	ADJ
ap-8720	269	3	efficiency	efficiency	NOUN
ap-8720	269	4	:	:	PUNCT
ap-8720	269	5	a	a	DET
ap-8720	269	6	larger	large	ADJ
ap-8720	269	7	batch	batch	NOUN
ap-8720	269	8	size	size	NOUN
ap-8720	269	9	enables	enable	VERB
ap-8720	269	10	parallelism	parallelism	NOUN
ap-8720	269	11	and	and	CCONJ
ap-8720	269	12	can	can	AUX
ap-8720	269	13	better	well	ADV
ap-8720	269	14	utilise	utilise	VERB
ap-8720	269	15	hardware	hardware	NOUN
ap-8720	269	16	acceleration	acceleration	NOUN
ap-8720	269	17	such	such	ADJ
ap-8720	269	18	as	as	ADP
ap-8720	269	19	gpus	gpu	NOUN
ap-8720	269	20	.	.	PUNCT
ap-8720	270	1	training	training	NOUN
ap-8720	270	2	with	with	ADP
ap-8720	270	3	larger	large	ADJ
ap-8720	270	4	batch	batch	NOUN
ap-8720	270	5	sizes	size	NOUN
ap-8720	270	6	can	can	AUX
ap-8720	270	7	speed	speed	VERB
ap-8720	270	8	up	up	ADP
ap-8720	270	9	the	the	DET
ap-8720	270	10	training	training	NOUN
ap-8720	270	11	process	process	NOUN
ap-8720	270	12	.	.	PUNCT
ap-8720	271	1	•	•	NUM
ap-8720	271	2	generalisation	generalisation	NOUN
ap-8720	271	3	:	:	PUNCT
ap-8720	271	4	smaller	small	ADJ
ap-8720	271	5	batch	batch	NOUN
ap-8720	271	6	sizes	size	NOUN
ap-8720	271	7	often	often	ADV
ap-8720	271	8	lead	lead	VERB
ap-8720	271	9	to	to	ADP
ap-8720	271	10	better	well	ADJ
ap-8720	271	11	generalisation	generalisation	NOUN
ap-8720	271	12	but	but	CCONJ
ap-8720	271	13	may	may	AUX
ap-8720	271	14	require	require	VERB
ap-8720	271	15	more	more	ADJ
ap-8720	271	16	iterations	iteration	NOUN
ap-8720	271	17	to	to	PART
ap-8720	271	18	converge	converge	VERB
ap-8720	271	19	.	.	PUNCT
ap-8720	272	1	•	•	NUM
ap-8720	272	2	memory	memory	NOUN
ap-8720	272	3	constraints	constraint	NOUN
ap-8720	272	4	:	:	PUNCT
ap-8720	272	5	the	the	DET
ap-8720	272	6	batch	batch	NOUN
ap-8720	272	7	size	size	NOUN
ap-8720	272	8	must	must	AUX
ap-8720	272	9	fit	fit	VERB
ap-8720	272	10	within	within	ADP
ap-8720	272	11	the	the	DET
ap-8720	272	12	available	available	ADJ
ap-8720	272	13	memory	memory	NOUN
ap-8720	272	14	of	of	ADP
ap-8720	272	15	the	the	DET
ap-8720	272	16	training	training	NOUN
ap-8720	272	17	hardware	hardware	NOUN
ap-8720	272	18	.	.	PUNCT
ap-8720	273	1	using	use	VERB
ap-8720	273	2	excessively	excessively	ADV
ap-8720	273	3	large	large	ADJ
ap-8720	273	4	batch	batch	NOUN
ap-8720	273	5	sizes	size	NOUN
ap-8720	273	6	can	can	AUX
ap-8720	273	7	exceed	exceed	VERB
ap-8720	273	8	the	the	DET
ap-8720	273	9	memory	memory	NOUN
ap-8720	273	10	capacity	capacity	NOUN
ap-8720	273	11	and	and	CCONJ
ap-8720	273	12	result	result	VERB
ap-8720	273	13	in	in	ADP
ap-8720	273	14	out	out	ADV
ap-8720	273	15	-	-	PUNCT
ap-8720	273	16	of	of	ADP
ap-8720	273	17	-	-	PUNCT
ap-8720	273	18	memory	memory	NOUN
ap-8720	273	19	errors	error	NOUN
ap-8720	273	20	.	.	PUNCT
ap-8720	274	1	number	number	NOUN
ap-8720	274	2	of	of	ADP
ap-8720	274	3	epochs	epoch	NOUN
ap-8720	274	4	:	:	PUNCT
ap-8720	274	5	an	an	DET
ap-8720	274	6	epoch	epoch	NOUN
ap-8720	274	7	signifies	signify	VERB
ap-8720	274	8	a	a	DET
ap-8720	274	9	complete	complete	ADJ
ap-8720	274	10	pass	pass	NOUN
ap-8720	274	11	through	through	ADP
ap-8720	274	12	the	the	DET
ap-8720	274	13	entire	entire	ADJ
ap-8720	274	14	training	training	NOUN
ap-8720	274	15	dataset	dataset	NOUN
ap-8720	274	16	.	.	PUNCT
ap-8720	275	1	the	the	DET
ap-8720	275	2	number	number	NOUN
ap-8720	275	3	of	of	ADP
ap-8720	275	4	epochs	epoch	NOUN
ap-8720	275	5	determines	determine	VERB
ap-8720	275	6	how	how	SCONJ
ap-8720	275	7	many	many	ADJ
ap-8720	275	8	times	time	NOUN
ap-8720	275	9	the	the	DET
ap-8720	275	10	model	model	NOUN
ap-8720	275	11	will	will	AUX
ap-8720	275	12	go	go	VERB
ap-8720	275	13	through	through	ADP
ap-8720	275	14	the	the	DET
ap-8720	275	15	entire	entire	ADJ
ap-8720	275	16	dataset	dataset	NOUN
ap-8720	275	17	during	during	ADP
ap-8720	275	18	training	training	NOUN
ap-8720	275	19	.	.	PUNCT
ap-8720	276	1	the	the	DET
ap-8720	276	2	number	number	NOUN
ap-8720	276	3	of	of	ADP
ap-8720	276	4	epochs	epoch	NOUN
ap-8720	276	5	can	can	AUX
ap-8720	276	6	influence	influence	VERB
ap-8720	276	7	training	training	NOUN
ap-8720	276	8	in	in	ADP
ap-8720	276	9	various	various	ADJ
ap-8720	276	10	ways	way	NOUN
ap-8720	276	11	:	:	PUNCT
ap-8720	276	12	•	•	NOUN
ap-8720	276	13	underfitting	underfitting	NOUN
ap-8720	276	14	vs.	vs.	ADP
ap-8720	276	15	overfitting	overfitte	VERB
ap-8720	276	16	:	:	PUNCT
ap-8720	276	17	training	training	NOUN
ap-8720	276	18	for	for	ADP
ap-8720	276	19	too	too	ADV
ap-8720	276	20	few	few	ADJ
ap-8720	276	21	epochs	epoch	NOUN
ap-8720	276	22	can	can	AUX
ap-8720	276	23	result	result	VERB
ap-8720	276	24	in	in	ADP
ap-8720	276	25	underfitting	underfitting	NOUN
ap-8720	276	26	,	,	PUNCT
ap-8720	276	27	where	where	SCONJ
ap-8720	276	28	the	the	DET
ap-8720	276	29	model	model	NOUN
ap-8720	276	30	fails	fail	VERB
ap-8720	276	31	to	to	PART
ap-8720	276	32	learn	learn	VERB
ap-8720	276	33	enough	enough	ADV
ap-8720	276	34	from	from	ADP
ap-8720	276	35	the	the	DET
ap-8720	276	36	data	datum	NOUN
ap-8720	276	37	.	.	PUNCT
ap-8720	277	1	however	however	ADV
ap-8720	277	2	,	,	PUNCT
ap-8720	277	3	training	train	VERB
ap-8720	277	4	for	for	ADP
ap-8720	277	5	too	too	ADV
ap-8720	277	6	many	many	ADJ
ap-8720	277	7	epochs	epoch	NOUN
ap-8720	277	8	can	can	AUX
ap-8720	277	9	lead	lead	VERB
ap-8720	277	10	to	to	ADP
ap-8720	277	11	overfitting	overfitte	VERB
ap-8720	277	12	,	,	PUNCT
ap-8720	277	13	where	where	SCONJ
ap-8720	277	14	the	the	DET
ap-8720	277	15	model	model	NOUN
ap-8720	277	16	becomes	become	VERB
ap-8720	277	17	overly	overly	ADV
ap-8720	277	18	specialised	specialised	ADJ
ap-8720	277	19	to	to	ADP
ap-8720	277	20	the	the	DET
ap-8720	277	21	training	training	NOUN
ap-8720	277	22	data	datum	NOUN
ap-8720	277	23	and	and	CCONJ
ap-8720	277	24	fails	fail	VERB
ap-8720	277	25	to	to	PART
ap-8720	277	26	generalise	generalise	VERB
ap-8720	277	27	well	well	ADV
ap-8720	277	28	to	to	ADP
ap-8720	277	29	new	new	ADJ
ap-8720	277	30	data	datum	NOUN
ap-8720	277	31	.	.	PUNCT
ap-8720	278	1	•	•	NOUN
ap-8720	278	2	training	training	NOUN
ap-8720	278	3	time	time	NOUN
ap-8720	278	4	:	:	PUNCT
ap-8720	278	5	increasing	increase	VERB
ap-8720	278	6	the	the	DET
ap-8720	278	7	number	number	NOUN
ap-8720	278	8	of	of	ADP
ap-8720	278	9	epochs	epoch	NOUN
ap-8720	278	10	increases	increase	VERB
ap-8720	278	11	the	the	DET
ap-8720	278	12	overall	overall	ADJ
ap-8720	278	13	training	training	NOUN
ap-8720	278	14	time	time	NOUN
ap-8720	278	15	.	.	PUNCT
ap-8720	279	1	training	train	VERB
ap-8720	279	2	deep	deep	ADJ
ap-8720	279	3	learning	learning	NOUN
ap-8720	279	4	models	model	NOUN
ap-8720	279	5	can	can	AUX
ap-8720	279	6	be	be	AUX
ap-8720	279	7	time	time	NOUN
ap-8720	279	8	consuming	consume	VERB
ap-8720	279	9	,	,	PUNCT
ap-8720	279	10	so	so	CCONJ
ap-8720	279	11	there	there	PRON
ap-8720	279	12	is	be	VERB
ap-8720	279	13	typically	typically	ADV
ap-8720	279	14	a	a	DET
ap-8720	279	15	trade	trade	NOUN
ap-8720	279	16	-	-	PUNCT
ap-8720	279	17	off	off	NOUN
ap-8720	279	18	between	between	ADP
ap-8720	279	19	computational	computational	ADJ
ap-8720	279	20	resources	resource	NOUN
ap-8720	279	21	and	and	CCONJ
ap-8720	279	22	achieving	achieve	VERB
ap-8720	279	23	the	the	DET
ap-8720	279	24	optimal	optimal	ADJ
ap-8720	279	25	performance	performance	NOUN
ap-8720	279	26	.	.	PUNCT
ap-8720	280	1	•	•	NUM
ap-8720	280	2	learning	learn	VERB
ap-8720	280	3	rate	rate	NOUN
ap-8720	280	4	schedule	schedule	NOUN
ap-8720	280	5	:	:	PUNCT
ap-8720	280	6	the	the	DET
ap-8720	280	7	number	number	NOUN
ap-8720	280	8	of	of	ADP
ap-8720	280	9	epochs	epoch	NOUN
ap-8720	280	10	can	can	AUX
ap-8720	280	11	also	also	ADV
ap-8720	280	12	impact	impact	VERB
ap-8720	280	13	the	the	DET
ap-8720	280	14	learning	learning	NOUN
ap-8720	280	15	rate	rate	NOUN
ap-8720	280	16	schedule	schedule	NOUN
ap-8720	280	17	.	.	PUNCT
ap-8720	281	1	certain	certain	ADJ
ap-8720	281	2	training	training	NOUN
ap-8720	281	3	strategies	strategy	NOUN
ap-8720	281	4	,	,	PUNCT
ap-8720	281	5	such	such	ADJ
ap-8720	281	6	as	as	ADP
ap-8720	281	7	learning	learn	VERB
ap-8720	281	8	rate	rate	NOUN
ap-8720	281	9	decay	decay	NOUN
ap-8720	281	10	or	or	CCONJ
ap-8720	281	11	figure	figure	NOUN
ap-8720	281	12	10	10	NUM
ap-8720	281	13	.	.	PUNCT
ap-8720	282	1	training	training	NOUN
ap-8720	282	2	of	of	ADP
ap-8720	282	3	discriminator	discriminator	NOUN
ap-8720	282	4	1	1	NUM
ap-8720	282	5	.	.	PUNCT
ap-8720	282	6	class	class	NOUN
ap-8720	282	7	training	training	NOUN
ap-8720	282	8	test	test	NOUN
ap-8720	282	9	real	real	ADJ
ap-8720	282	10	faces	face	VERB
ap-8720	282	11	50	50	NUM
ap-8720	282	12	000	000	NUM
ap-8720	282	13	10	10	NUM
ap-8720	282	14	000	000	NUM
ap-8720	282	15	fake	fake	ADJ
ap-8720	282	16	faces	face	NOUN
ap-8720	282	17	–	–	PUNCT
ap-8720	282	18	10	10	NUM
ap-8720	282	19	000	000	NUM
ap-8720	282	20	table	table	NOUN
ap-8720	282	21	3	3	NUM
ap-8720	282	22	.	.	PUNCT
ap-8720	282	23	separation	separation	NOUN
ap-8720	282	24	of	of	ADP
ap-8720	282	25	training	training	NOUN
ap-8720	282	26	and	and	CCONJ
ap-8720	282	27	test	test	NOUN
ap-8720	282	28	data	datum	NOUN
ap-8720	282	29	sets	set	NOUN
ap-8720	282	30	.	.	PUNCT
ap-8720	283	1	learning	learn	VERB
ap-8720	283	2	rate	rate	NOUN
ap-8720	283	3	warm	warm	NOUN
ap-8720	283	4	-	-	PUNCT
ap-8720	283	5	up	up	NOUN
ap-8720	283	6	,	,	PUNCT
ap-8720	283	7	adjust	adjust	VERB
ap-8720	283	8	the	the	DET
ap-8720	283	9	learning	learning	NOUN
ap-8720	283	10	rate	rate	NOUN
ap-8720	283	11	as	as	SCONJ
ap-8720	283	12	the	the	DET
ap-8720	283	13	number	number	NOUN
ap-8720	283	14	of	of	ADP
ap-8720	283	15	epochs	epoch	NOUN
ap-8720	283	16	progresses	progress	VERB
ap-8720	283	17	.	.	PUNCT
ap-8720	284	1	determining	determine	VERB
ap-8720	284	2	the	the	DET
ap-8720	284	3	optimal	optimal	ADJ
ap-8720	284	4	batch	batch	NOUN
ap-8720	284	5	size	size	NOUN
ap-8720	284	6	and	and	CCONJ
ap-8720	284	7	number	number	NOUN
ap-8720	284	8	of	of	ADP
ap-8720	284	9	epochs	epoch	NOUN
ap-8720	284	10	often	often	ADV
ap-8720	284	11	requires	require	VERB
ap-8720	284	12	experimentation	experimentation	NOUN
ap-8720	284	13	and	and	CCONJ
ap-8720	284	14	tuning	tuning	NOUN
ap-8720	284	15	based	base	VERB
ap-8720	284	16	on	on	ADP
ap-8720	284	17	the	the	DET
ap-8720	284	18	specific	specific	ADJ
ap-8720	284	19	dataset	dataset	NOUN
ap-8720	284	20	,	,	PUNCT
ap-8720	284	21	model	model	NOUN
ap-8720	284	22	architecture	architecture	NOUN
ap-8720	284	23	,	,	PUNCT
ap-8720	284	24	and	and	CCONJ
ap-8720	284	25	problem	problem	NOUN
ap-8720	284	26	domain	domain	NOUN
ap-8720	284	27	.	.	PUNCT
ap-8720	285	1	it	it	PRON
ap-8720	285	2	is	be	AUX
ap-8720	285	3	common	common	ADJ
ap-8720	285	4	to	to	PART
ap-8720	285	5	begin	begin	VERB
ap-8720	285	6	with	with	ADP
ap-8720	285	7	default	default	NOUN
ap-8720	285	8	values	value	NOUN
ap-8720	285	9	and	and	CCONJ
ap-8720	285	10	iteratively	iteratively	ADV
ap-8720	285	11	adjust	adjust	VERB
ap-8720	285	12	them	they	PRON
ap-8720	285	13	based	base	VERB
ap-8720	285	14	on	on	ADP
ap-8720	285	15	the	the	DET
ap-8720	285	16	observed	observed	ADJ
ap-8720	285	17	performance	performance	NOUN
ap-8720	285	18	.	.	PUNCT
ap-8720	286	1	4.3.1	4.3.1	X
ap-8720	286	2	.	.	PUNCT
ap-8720	287	1	training	training	NOUN
ap-8720	287	2	of	of	ADP
ap-8720	287	3	discriminator	discriminator	NOUN
ap-8720	287	4	1	1	NUM
ap-8720	287	5	to	to	PART
ap-8720	287	6	develop	develop	VERB
ap-8720	287	7	the	the	DET
ap-8720	287	8	model	model	NOUN
ap-8720	287	9	of	of	ADP
ap-8720	287	10	discriminator	discriminator	NOUN
ap-8720	287	11	1	1	NUM
ap-8720	287	12	,	,	PUNCT
ap-8720	287	13	we	we	PRON
ap-8720	287	14	used	use	VERB
ap-8720	287	15	adversarial	adversarial	ADJ
ap-8720	287	16	learning	learning	NOUN
ap-8720	287	17	between	between	ADP
ap-8720	287	18	generator	generator	NOUN
ap-8720	287	19	and	and	CCONJ
ap-8720	287	20	discriminator	discriminator	NOUN
ap-8720	287	21	networks	network	NOUN
ap-8720	287	22	(	(	PUNCT
ap-8720	287	23	figure	figure	NOUN
ap-8720	287	24	10	10	NUM
ap-8720	287	25	)	)	PUNCT
ap-8720	287	26	.	.	PUNCT
ap-8720	288	1	where	where	SCONJ
ap-8720	288	2	the	the	DET
ap-8720	288	3	real	real	ADJ
ap-8720	288	4	faces	face	NOUN
ap-8720	288	5	used	use	VERB
ap-8720	288	6	(	(	PUNCT
ap-8720	288	7	60	60	NUM
ap-8720	288	8	000	000	NUM
ap-8720	288	9	)	)	PUNCT
ap-8720	288	10	are	be	AUX
ap-8720	288	11	from	from	ADP
ap-8720	288	12	the	the	DET
ap-8720	288	13	flickr	flickr	PROPN
ap-8720	288	14	dataset	dataset	VERB
ap-8720	289	1	[	[	X
ap-8720	289	2	18	18	NUM
ap-8720	289	3	]	]	PUNCT
ap-8720	289	4	and	and	CCONJ
ap-8720	289	5	the	the	DET
ap-8720	289	6	fake	fake	ADJ
ap-8720	289	7	faces	face	NOUN
ap-8720	289	8	are	be	AUX
ap-8720	289	9	from	from	ADP
ap-8720	289	10	the	the	DET
ap-8720	289	11	stylegan	stylegan	NOUN
ap-8720	289	12	dataset	dataset	NOUN
ap-8720	289	13	[	[	X
ap-8720	289	14	19	19	NUM
ap-8720	289	15	]	]	PUNCT
ap-8720	289	16	.	.	PUNCT
ap-8720	290	1	the	the	DET
ap-8720	290	2	datasets	dataset	NOUN
ap-8720	290	3	are	be	AUX
ap-8720	290	4	divided	divide	VERB
ap-8720	290	5	into	into	ADP
ap-8720	290	6	training	training	NOUN
ap-8720	290	7	and	and	CCONJ
ap-8720	290	8	test	test	NOUN
ap-8720	290	9	as	as	SCONJ
ap-8720	290	10	shown	show	VERB
ap-8720	290	11	in	in	ADP
ap-8720	290	12	table	table	NOUN
ap-8720	290	13	3	3	NUM
ap-8720	290	14	.	.	PUNCT
ap-8720	291	1	during	during	ADP
ap-8720	291	2	the	the	DET
ap-8720	291	3	training	training	NOUN
ap-8720	291	4	of	of	ADP
ap-8720	291	5	this	this	DET
ap-8720	291	6	network	network	NOUN
ap-8720	291	7	,	,	PUNCT
ap-8720	291	8	we	we	PRON
ap-8720	291	9	kept	keep	VERB
ap-8720	291	10	the	the	DET
ap-8720	291	11	following	follow	VERB
ap-8720	291	12	configuration	configuration	NOUN
ap-8720	291	13	:	:	PUNCT
ap-8720	291	14	•	•	NUM
ap-8720	291	15	optimiser	optimiser	NOUN
ap-8720	291	16	=	=	NOUN
ap-8720	291	17	rmsprop	rmsprop	NOUN
ap-8720	291	18	which	which	PRON
ap-8720	291	19	generates	generate	VERB
ap-8720	291	20	more	more	ADV
ap-8720	291	21	realistic	realistic	ADJ
ap-8720	291	22	fake	fake	ADJ
ap-8720	291	23	images	image	NOUN
ap-8720	291	24	compared	compare	VERB
ap-8720	291	25	to	to	ADP
ap-8720	291	26	adam	adam	PROPN
ap-8720	291	27	.	.	PUNCT
ap-8720	292	1	•	•	NUM
ap-8720	292	2	batch	batch	NOUN
ap-8720	292	3	size	size	NOUN
ap-8720	292	4	=	=	SYM
ap-8720	292	5	128	128	NUM
ap-8720	292	6	.	.	NOUN
ap-8720	293	1	•	•	NUM
ap-8720	293	2	number	number	NOUN
ap-8720	293	3	of	of	ADP
ap-8720	293	4	iterations	iteration	NOUN
ap-8720	293	5	=	=	SYM
ap-8720	293	6	10	10	NUM
ap-8720	293	7	000	000	NUM
ap-8720	293	8	.	.	PUNCT
ap-8720	294	1	during	during	ADP
ap-8720	294	2	the	the	DET
ap-8720	294	3	training	training	NOUN
ap-8720	294	4	,	,	PUNCT
ap-8720	294	5	the	the	DET
ap-8720	294	6	discriminator	discriminator	NOUN
ap-8720	294	7	tries	try	VERB
ap-8720	294	8	to	to	PART
ap-8720	294	9	maximise	maximise	VERB
ap-8720	294	10	the	the	DET
ap-8720	294	11	logarithmic	logarithmic	ADJ
ap-8720	294	12	probability	probability	NOUN
ap-8720	294	13	that	that	SCONJ
ap-8720	294	14	the	the	DET
ap-8720	294	15	data	data	NOUN
ap-8720	294	16	is	be	AUX
ap-8720	294	17	real	real	ADJ
ap-8720	294	18	or	or	CCONJ
ap-8720	294	19	false	false	ADJ
ap-8720	294	20	.	.	PUNCT
ap-8720	295	1	conversely	conversely	ADV
ap-8720	295	2	,	,	PUNCT
ap-8720	295	3	the	the	DET
ap-8720	295	4	generator	generator	NOUN
ap-8720	295	5	tries	try	VERB
ap-8720	295	6	to	to	PART
ap-8720	295	7	minimise	minimise	VERB
ap-8720	295	8	the	the	DET
ap-8720	295	9	logarithmic	logarithmic	ADJ
ap-8720	295	10	probability	probability	NOUN
ap-8720	295	11	that	that	SCONJ
ap-8720	295	12	the	the	DET
ap-8720	295	13	discriminator	discriminator	NOUN
ap-8720	295	14	is	be	AUX
ap-8720	295	15	correct	correct	ADJ
ap-8720	295	16	.	.	PUNCT
ap-8720	296	1	it	it	PRON
ap-8720	296	2	is	be	AUX
ap-8720	296	3	therefore	therefore	ADV
ap-8720	296	4	considered	consider	VERB
ap-8720	296	5	a	a	DET
ap-8720	296	6	min	min	ADJ
ap-8720	296	7	-	-	ADJ
ap-8720	296	8	max	max	PROPN
ap-8720	296	9	game	game	NOUN
ap-8720	296	10	.	.	PUNCT
ap-8720	297	1	313	313	NUM
ap-8720	297	2	ch	ch	NOUN
ap-8720	297	3	.	.	PUNCT
ap-8720	297	4	e.	e.	PROPN
ap-8720	297	5	bencheriet	bencheriet	PROPN
ap-8720	297	6	,	,	PUNCT
ap-8720	297	7	h.	h.	PROPN
ap-8720	297	8	abdelmoumène	abdelmoumène	PROPN
ap-8720	297	9	,	,	PUNCT
ap-8720	297	10	a.	a.	PROPN
ap-8720	297	11	sebbagh	sebbagh	PROPN
ap-8720	297	12	et	et	PROPN
ap-8720	298	1	al	al	PROPN
ap-8720	298	2	.	.	PROPN
ap-8720	299	1	acta	acta	PROPN
ap-8720	299	2	polytechnica	polytechnica	PROPN
ap-8720	299	3	the	the	DET
ap-8720	299	4	standard	standard	ADJ
ap-8720	299	5	gan	gan	ADJ
ap-8720	299	6	loss	loss	NOUN
ap-8720	299	7	function	function	NOUN
ap-8720	299	8	,	,	PUNCT
ap-8720	299	9	also	also	ADV
ap-8720	299	10	called	call	VERB
ap-8720	299	11	the	the	DET
ap-8720	299	12	min	min	ADJ
ap-8720	299	13	-	-	ADJ
ap-8720	299	14	max	max	PROPN
ap-8720	299	15	loss	loss	NOUN
ap-8720	299	16	function	function	NOUN
ap-8720	299	17	,	,	PUNCT
ap-8720	299	18	is	be	AUX
ap-8720	299	19	given	give	VERB
ap-8720	299	20	by	by	ADP
ap-8720	299	21	the	the	DET
ap-8720	299	22	following	follow	VERB
ap-8720	299	23	formula	formula	NOUN
ap-8720	299	24	(	(	PUNCT
ap-8720	299	25	6	6	NUM
ap-8720	299	26	)	)	PUNCT
ap-8720	300	1	[	[	X
ap-8720	300	2	31	31	NUM
ap-8720	300	3	]	]	SYM
ap-8720	300	4	:	:	PUNCT
ap-8720	300	5	ex[log(d(x	ex[log(d(x	NOUN
ap-8720	300	6	)	)	PUNCT
ap-8720	300	7	)	)	PUNCT
ap-8720	300	8	]	]	PUNCT
ap-8720	301	1	+	+	CCONJ
ap-8720	301	2	ez[log(1	ez[log(1	ADP
ap-8720	301	3	−	−	NOUN
ap-8720	301	4	d(g(z	d(g(z	NOUN
ap-8720	301	5	)	)	PUNCT
ap-8720	301	6	)	)	PUNCT
ap-8720	301	7	)	)	PUNCT
ap-8720	301	8	]	]	PUNCT
ap-8720	301	9	,	,	PUNCT
ap-8720	301	10	(	(	PUNCT
ap-8720	301	11	6	6	NUM
ap-8720	301	12	)	)	PUNCT
ap-8720	301	13	where	where	SCONJ
ap-8720	301	14	:	:	PUNCT
ap-8720	301	15	d(x	d(x	PROPN
ap-8720	301	16	)	)	PUNCT
ap-8720	301	17	is	be	AUX
ap-8720	301	18	an	an	DET
ap-8720	301	19	estimate	estimate	NOUN
ap-8720	301	20	of	of	ADP
ap-8720	301	21	the	the	DET
ap-8720	301	22	probability	probability	NOUN
ap-8720	301	23	that	that	SCONJ
ap-8720	301	24	the	the	DET
ap-8720	301	25	actual	actual	ADJ
ap-8720	301	26	data	data	NOUN
ap-8720	301	27	image	image	NOUN
ap-8720	301	28	x	x	PUNCT
ap-8720	301	29	is	be	AUX
ap-8720	301	30	real	real	ADJ
ap-8720	301	31	.	.	PUNCT
ap-8720	302	1	ex	ex	X
ap-8720	302	2	is	be	AUX
ap-8720	302	3	the	the	DET
ap-8720	302	4	expected	expect	VERB
ap-8720	302	5	value	value	NOUN
ap-8720	302	6	over	over	ADP
ap-8720	302	7	all	all	DET
ap-8720	302	8	actual	actual	ADJ
ap-8720	302	9	data	datum	NOUN
ap-8720	302	10	instances	instance	NOUN
ap-8720	302	11	.	.	PUNCT
ap-8720	303	1	g(z	g(z	ADJ
ap-8720	303	2	)	)	PUNCT
ap-8720	303	3	is	be	AUX
ap-8720	303	4	the	the	DET
ap-8720	303	5	output	output	NOUN
ap-8720	303	6	of	of	ADP
ap-8720	303	7	the	the	DET
ap-8720	303	8	generator	generator	NOUN
ap-8720	303	9	when	when	SCONJ
ap-8720	303	10	it	it	PRON
ap-8720	303	11	receives	receive	VERB
ap-8720	303	12	a	a	DET
ap-8720	303	13	value	value	NOUN
ap-8720	303	14	of	of	ADP
ap-8720	303	15	z	z	NOUN
ap-8720	303	16	for	for	ADP
ap-8720	303	17	the	the	DET
ap-8720	303	18	noise	noise	NOUN
ap-8720	303	19	.	.	PUNCT
ap-8720	304	1	d(g(z	d(g(z	NOUN
ap-8720	304	2	)	)	PUNCT
ap-8720	304	3	)	)	PUNCT
ap-8720	304	4	is	be	AUX
ap-8720	304	5	the	the	DET
ap-8720	304	6	estimate	estimate	NOUN
ap-8720	304	7	of	of	ADP
ap-8720	304	8	the	the	DET
ap-8720	304	9	probability	probability	NOUN
ap-8720	304	10	that	that	SCONJ
ap-8720	304	11	a	a	DET
ap-8720	304	12	false	false	ADJ
ap-8720	304	13	instance	instance	NOUN
ap-8720	304	14	is	be	AUX
ap-8720	304	15	real	real	ADJ
ap-8720	304	16	.	.	PUNCT
ap-8720	305	1	ez	ez	PROPN
ap-8720	305	2	is	be	AUX
ap-8720	305	3	the	the	DET
ap-8720	305	4	expected	expect	VERB
ap-8720	305	5	value	value	NOUN
ap-8720	305	6	over	over	ADP
ap-8720	305	7	all	all	DET
ap-8720	305	8	random	random	ADJ
ap-8720	305	9	inputs	input	NOUN
ap-8720	305	10	of	of	ADP
ap-8720	305	11	the	the	DET
ap-8720	305	12	generator	generator	NOUN
ap-8720	305	13	(	(	PUNCT
ap-8720	305	14	in	in	ADP
ap-8720	305	15	fact	fact	NOUN
ap-8720	305	16	,	,	PUNCT
ap-8720	305	17	it	it	PRON
ap-8720	305	18	is	be	AUX
ap-8720	305	19	the	the	DET
ap-8720	305	20	expected	expect	VERB
ap-8720	305	21	value	value	NOUN
ap-8720	305	22	over	over	ADP
ap-8720	305	23	all	all	PRON
ap-8720	305	24	generated	generate	VERB
ap-8720	305	25	false	false	ADJ
ap-8720	305	26	instances	instance	NOUN
ap-8720	305	27	,	,	PUNCT
ap-8720	305	28	g(z	g(z	PROPN
ap-8720	305	29	)	)	PUNCT
ap-8720	305	30	)	)	PUNCT
ap-8720	305	31	.	.	PUNCT
ap-8720	306	1	the	the	DET
ap-8720	306	2	standard	standard	ADJ
ap-8720	306	3	gan	gan	ADJ
ap-8720	306	4	loss	loss	NOUN
ap-8720	306	5	function	function	NOUN
ap-8720	306	6	can	can	AUX
ap-8720	306	7	be	be	AUX
ap-8720	306	8	further	far	ADV
ap-8720	306	9	divided	divide	VERB
ap-8720	306	10	into	into	ADP
ap-8720	306	11	two	two	NUM
ap-8720	306	12	parts	part	NOUN
ap-8720	306	13	:	:	PUNCT
ap-8720	306	14	discriminator	discriminator	NOUN
ap-8720	306	15	loss	loss	NOUN
ap-8720	306	16	and	and	CCONJ
ap-8720	306	17	generator	generator	NOUN
ap-8720	306	18	loss	loss	NOUN
ap-8720	306	19	.	.	PUNCT
ap-8720	307	1	discriminator	discriminator	NOUN
ap-8720	307	2	loss	loss	NOUN
ap-8720	307	3	during	during	ADP
ap-8720	307	4	training	training	NOUN
ap-8720	308	1	,	,	PUNCT
ap-8720	308	2	the	the	DET
ap-8720	308	3	discriminator	discriminator	NOUN
ap-8720	308	4	classifies	classify	VERB
ap-8720	308	5	both	both	CCONJ
ap-8720	308	6	the	the	DET
ap-8720	308	7	real	real	ADJ
ap-8720	308	8	face	face	NOUN
ap-8720	308	9	and	and	CCONJ
ap-8720	308	10	the	the	DET
ap-8720	308	11	fake	fake	ADJ
ap-8720	308	12	face	face	NOUN
ap-8720	308	13	from	from	ADP
ap-8720	308	14	the	the	DET
ap-8720	308	15	generator	generator	NOUN
ap-8720	308	16	and	and	CCONJ
ap-8720	308	17	penalises	penalise	NOUN
ap-8720	308	18	itself	itself	PRON
ap-8720	308	19	for	for	ADP
ap-8720	308	20	misclassifying	misclassifye	VERB
ap-8720	308	21	a	a	DET
ap-8720	308	22	real	real	ADJ
ap-8720	308	23	face	face	NOUN
ap-8720	308	24	as	as	ADP
ap-8720	308	25	fake	fake	ADJ
ap-8720	308	26	,	,	PUNCT
ap-8720	308	27	or	or	CCONJ
ap-8720	308	28	a	a	DET
ap-8720	308	29	fake	fake	ADJ
ap-8720	308	30	face	face	NOUN
ap-8720	308	31	as	as	ADP
ap-8720	308	32	real	real	ADJ
ap-8720	308	33	,	,	PUNCT
ap-8720	308	34	by	by	ADP
ap-8720	308	35	maximising	maximise	VERB
ap-8720	308	36	the	the	DET
ap-8720	308	37	function	function	NOUN
ap-8720	308	38	(	(	PUNCT
ap-8720	308	39	7	7	NUM
ap-8720	308	40	):	):	PUNCT
ap-8720	308	41	∇θd	∇θd	NOUN
ap-8720	308	42	1	1	NUM
ap-8720	308	43	m	m	NOUN
ap-8720	308	44	m∑	m∑	NOUN
ap-8720	308	45	i=1	i=1	X
ap-8720	309	1	[	[	PUNCT
ap-8720	309	2	log	log	NOUN
ap-8720	309	3	d(x(i	d(x(i	PROPN
ap-8720	309	4	)	)	PUNCT
ap-8720	309	5	)	)	PUNCT
ap-8720	310	1	+	+	CCONJ
ap-8720	310	2	log(1	log(1	NOUN
ap-8720	310	3	−	−	ADP
ap-8720	310	4	d(g(z(i	d(g(z(i	PROPN
ap-8720	310	5	)	)	PUNCT
ap-8720	310	6	)	)	PUNCT
ap-8720	310	7	)	)	PUNCT
ap-8720	310	8	)	)	PUNCT
ap-8720	310	9	]	]	PUNCT
ap-8720	310	10	,	,	PUNCT
ap-8720	310	11	(	(	PUNCT
ap-8720	310	12	7	7	X
ap-8720	310	13	)	)	PUNCT
ap-8720	310	14	where	where	SCONJ
ap-8720	310	15	:	:	PUNCT
ap-8720	310	16	•	•	NUM
ap-8720	310	17	log(d(x	log(d(x	ADJ
ap-8720	310	18	)	)	PUNCT
ap-8720	310	19	)	)	PUNCT
ap-8720	310	20	refers	refer	VERB
ap-8720	310	21	to	to	ADP
ap-8720	310	22	the	the	DET
ap-8720	310	23	probability	probability	NOUN
ap-8720	310	24	that	that	SCONJ
ap-8720	310	25	the	the	DET
ap-8720	310	26	generator	generator	NOUN
ap-8720	310	27	is	be	AUX
ap-8720	310	28	correctly	correctly	ADV
ap-8720	310	29	classifying	classify	VERB
ap-8720	310	30	the	the	DET
ap-8720	310	31	real	real	ADJ
ap-8720	310	32	face	face	NOUN
ap-8720	310	33	,	,	PUNCT
ap-8720	310	34	•	•	ADV
ap-8720	310	35	maximising	maximise	VERB
ap-8720	310	36	log(1	log(1	NOUN
ap-8720	310	37	−	−	PUNCT
ap-8720	311	1	d(g(z	d(g(z	NOUN
ap-8720	311	2	)	)	PUNCT
ap-8720	311	3	)	)	PUNCT
ap-8720	311	4	)	)	PUNCT
ap-8720	311	5	would	would	AUX
ap-8720	311	6	help	help	VERB
ap-8720	311	7	it	it	PRON
ap-8720	311	8	to	to	PART
ap-8720	311	9	correctly	correctly	ADV
ap-8720	311	10	label	label	VERB
ap-8720	311	11	the	the	DET
ap-8720	311	12	fake	fake	ADJ
ap-8720	311	13	face	face	NOUN
ap-8720	311	14	that	that	PRON
ap-8720	311	15	comes	come	VERB
ap-8720	311	16	from	from	ADP
ap-8720	311	17	the	the	DET
ap-8720	311	18	generator	generator	NOUN
ap-8720	311	19	.	.	PUNCT
ap-8720	312	1	generator	generator	PROPN
ap-8720	312	2	loss	loss	NOUN
ap-8720	312	3	the	the	DET
ap-8720	312	4	discriminator	discriminator	NOUN
ap-8720	312	5	’s	’s	PART
ap-8720	312	6	classification	classification	NOUN
ap-8720	312	7	is	be	AUX
ap-8720	312	8	used	use	VERB
ap-8720	312	9	to	to	PART
ap-8720	312	10	calculate	calculate	VERB
ap-8720	312	11	the	the	DET
ap-8720	312	12	generator	generator	NOUN
ap-8720	312	13	loss	loss	NOUN
ap-8720	312	14	.	.	PUNCT
ap-8720	313	1	if	if	SCONJ
ap-8720	313	2	succeeds	succeed	VERB
ap-8720	313	3	in	in	ADP
ap-8720	313	4	fooling	fool	VERB
ap-8720	313	5	the	the	DET
ap-8720	313	6	discriminator	discriminator	NOUN
ap-8720	313	7	,	,	PUNCT
ap-8720	313	8	it	it	PRON
ap-8720	313	9	is	be	AUX
ap-8720	313	10	rewarded	reward	VERB
ap-8720	313	11	;	;	PUNCT
ap-8720	313	12	if	if	SCONJ
ap-8720	313	13	not	not	PART
ap-8720	313	14	,	,	PUNCT
ap-8720	313	15	it	it	PRON
ap-8720	313	16	is	be	AUX
ap-8720	313	17	penalised	penalise	VERB
ap-8720	313	18	.	.	PUNCT
ap-8720	314	1	equation	equation	NOUN
ap-8720	314	2	(	(	PUNCT
ap-8720	314	3	8)	8)	NUM
ap-8720	314	4	is	be	AUX
ap-8720	314	5	minimised	minimise	VERB
ap-8720	314	6	to	to	PART
ap-8720	314	7	train	train	VERB
ap-8720	314	8	the	the	DET
ap-8720	314	9	generator	generator	NOUN
ap-8720	314	10	.	.	PUNCT
ap-8720	315	1	∇θg	∇θg	NOUN
ap-8720	315	2	1	1	NUM
ap-8720	315	3	m	m	VERB
ap-8720	315	4	m∑	m∑	X
ap-8720	315	5	i=1	i=1	NOUN
ap-8720	315	6	log(1	log(1	NOUN
ap-8720	315	7	−	−	PROPN
ap-8720	315	8	d(g(z(i	d(g(z(i	PROPN
ap-8720	315	9	)	)	PUNCT
ap-8720	315	10	)	)	PUNCT
ap-8720	315	11	)	)	PUNCT
ap-8720	315	12	)	)	PUNCT
ap-8720	315	13	.	.	PUNCT
ap-8720	316	1	(	(	PUNCT
ap-8720	316	2	8)	8)	NUM
ap-8720	316	3	figure	figure	NOUN
ap-8720	316	4	11	11	NUM
ap-8720	316	5	shows	show	VERB
ap-8720	316	6	the	the	DET
ap-8720	316	7	training	training	NOUN
ap-8720	316	8	results	result	NOUN
ap-8720	316	9	of	of	ADP
ap-8720	316	10	the	the	DET
ap-8720	316	11	generator	generator	NOUN
ap-8720	316	12	and	and	CCONJ
ap-8720	316	13	the	the	DET
ap-8720	316	14	discriminator	discriminator	NOUN
ap-8720	316	15	.	.	PUNCT
ap-8720	317	1	we	we	PRON
ap-8720	317	2	can	can	AUX
ap-8720	317	3	see	see	VERB
ap-8720	317	4	that	that	SCONJ
ap-8720	317	5	the	the	DET
ap-8720	317	6	discriminator	discriminator	NOUN
ap-8720	317	7	network	network	NOUN
ap-8720	317	8	achieved	achieve	VERB
ap-8720	317	9	a	a	DET
ap-8720	317	10	loss	loss	NOUN
ap-8720	317	11	for	for	ADP
ap-8720	317	12	real	real	ADJ
ap-8720	317	13	and	and	CCONJ
ap-8720	317	14	fake	fake	ADJ
ap-8720	317	15	samples	sample	NOUN
ap-8720	317	16	of	of	ADP
ap-8720	317	17	around	around	ADP
ap-8720	317	18	47	47	NUM
ap-8720	317	19	%	%	NOUN
ap-8720	317	20	and	and	CCONJ
ap-8720	317	21	the	the	DET
ap-8720	317	22	generator	generator	NOUN
ap-8720	317	23	of	of	ADP
ap-8720	317	24	around	around	ADV
ap-8720	317	25	96	96	NUM
ap-8720	317	26	%	%	NOUN
ap-8720	317	27	.	.	PUNCT
ap-8720	318	1	the	the	DET
ap-8720	318	2	discriminator	discriminator	NOUN
ap-8720	318	3	achieved	achieve	VERB
ap-8720	318	4	an	an	DET
ap-8720	318	5	accuracy	accuracy	NOUN
ap-8720	318	6	of	of	ADP
ap-8720	318	7	between	between	ADP
ap-8720	318	8	70	70	NUM
ap-8720	318	9	%	%	NOUN
ap-8720	318	10	and	and	CCONJ
ap-8720	318	11	79	79	NUM
ap-8720	318	12	%	%	NOUN
ap-8720	318	13	for	for	ADP
ap-8720	318	14	real	real	ADJ
ap-8720	318	15	and	and	CCONJ
ap-8720	318	16	fake	fake	ADJ
ap-8720	318	17	faces	face	NOUN
ap-8720	318	18	.	.	PUNCT
ap-8720	319	1	the	the	DET
ap-8720	319	2	learning	learning	NOUN
ap-8720	319	3	results	result	NOUN
ap-8720	319	4	are	be	AUX
ap-8720	319	5	acceptable	acceptable	ADJ
ap-8720	319	6	because	because	SCONJ
ap-8720	319	7	the	the	DET
ap-8720	319	8	final	final	ADJ
ap-8720	319	9	decision	decision	NOUN
ap-8720	319	10	is	be	AUX
ap-8720	319	11	a	a	DET
ap-8720	319	12	contribution	contribution	NOUN
ap-8720	319	13	by	by	ADP
ap-8720	319	14	voting	voting	NOUN
ap-8720	319	15	of	of	ADP
ap-8720	319	16	the	the	DET
ap-8720	319	17	decisions	decision	NOUN
ap-8720	319	18	of	of	ADP
ap-8720	319	19	the	the	DET
ap-8720	319	20	three	three	NUM
ap-8720	319	21	discriminators	discriminator	NOUN
ap-8720	319	22	.	.	PUNCT
ap-8720	320	1	therefore	therefore	ADV
ap-8720	320	2	,	,	PUNCT
ap-8720	320	3	the	the	DET
ap-8720	320	4	decision	decision	NOUN
ap-8720	320	5	of	of	ADP
ap-8720	320	6	discriminator	discriminator	NOUN
ap-8720	320	7	1	1	NUM
ap-8720	320	8	is	be	AUX
ap-8720	320	9	either	either	CCONJ
ap-8720	320	10	confirmed	confirm	VERB
ap-8720	320	11	or	or	CCONJ
ap-8720	320	12	contradicted	contradict	VERB
ap-8720	320	13	by	by	ADP
ap-8720	320	14	the	the	DET
ap-8720	320	15	other	other	ADJ
ap-8720	320	16	two	two	NUM
ap-8720	320	17	discriminators	discriminator	NOUN
ap-8720	320	18	,	,	PUNCT
ap-8720	320	19	which	which	PRON
ap-8720	320	20	form	form	VERB
ap-8720	320	21	the	the	DET
ap-8720	320	22	overall	overall	ADJ
ap-8720	320	23	architecture	architecture	NOUN
ap-8720	320	24	of	of	ADP
ap-8720	320	25	our	our	PRON
ap-8720	320	26	mdd	mdd	PROPN
ap-8720	320	27	-	-	PUNCT
ap-8720	320	28	cnn	cnn	PROPN
ap-8720	320	29	system	system	NOUN
ap-8720	320	30	.	.	PUNCT
ap-8720	321	1	class	class	NOUN
ap-8720	321	2	training	training	NOUN
ap-8720	321	3	validation	validation	NOUN
ap-8720	321	4	test	test	NOUN
ap-8720	321	5	real	real	ADJ
ap-8720	321	6	faces	face	VERB
ap-8720	321	7	50	50	NUM
ap-8720	321	8	000	000	NUM
ap-8720	321	9	10	10	NUM
ap-8720	321	10	000	000	NUM
ap-8720	321	11	10	10	NUM
ap-8720	321	12	000	000	NUM
ap-8720	321	13	fake	fake	ADJ
ap-8720	321	14	faces	face	NOUN
ap-8720	321	15	50	50	NUM
ap-8720	321	16	000	000	NUM
ap-8720	321	17	10	10	NUM
ap-8720	321	18	000	000	NUM
ap-8720	321	19	10	10	NUM
ap-8720	321	20	000	000	NUM
ap-8720	321	21	total	total	NOUN
ap-8720	321	22	100	100	NUM
ap-8720	321	23	000	000	NUM
ap-8720	321	24	20	20	NUM
ap-8720	321	25	000	000	NUM
ap-8720	321	26	20	20	NUM
ap-8720	321	27	000	000	NUM
ap-8720	321	28	table	table	NOUN
ap-8720	321	29	4	4	NUM
ap-8720	321	30	.	.	PUNCT
ap-8720	321	31	discriminator	discriminator	NOUN
ap-8720	321	32	2	2	NUM
ap-8720	321	33	training	training	NOUN
ap-8720	321	34	and	and	CCONJ
ap-8720	321	35	test	test	NOUN
ap-8720	321	36	data	datum	NOUN
ap-8720	321	37	separation	separation	NOUN
ap-8720	321	38	.	.	PUNCT
ap-8720	322	1	4.3.2	4.3.2	X
ap-8720	322	2	.	.	PUNCT
ap-8720	323	1	training	training	NOUN
ap-8720	323	2	of	of	ADP
ap-8720	323	3	discriminator	discriminator	NOUN
ap-8720	323	4	2	2	NUM
ap-8720	323	5	discriminator	discriminator	NOUN
ap-8720	323	6	2	2	NUM
ap-8720	323	7	is	be	AUX
ap-8720	323	8	the	the	DET
ap-8720	323	9	model	model	NOUN
ap-8720	323	10	of	of	ADP
ap-8720	323	11	discriminator	discriminator	NOUN
ap-8720	323	12	1	1	NUM
ap-8720	323	13	,	,	PUNCT
ap-8720	323	14	previously	previously	ADV
ap-8720	323	15	trained	train	VERB
ap-8720	323	16	with	with	ADP
ap-8720	323	17	its	its	PRON
ap-8720	323	18	opponent	opponent	NOUN
ap-8720	323	19	,	,	PUNCT
ap-8720	323	20	the	the	DET
ap-8720	323	21	generator	generator	NOUN
ap-8720	323	22	,	,	PUNCT
ap-8720	323	23	and	and	CCONJ
ap-8720	323	24	further	far	ADV
ap-8720	323	25	trained	train	VERB
ap-8720	323	26	by	by	ADP
ap-8720	323	27	transfer	transfer	NOUN
ap-8720	323	28	learning	learning	NOUN
ap-8720	323	29	on	on	ADP
ap-8720	323	30	a	a	DET
ap-8720	323	31	dataset	dataset	NOUN
ap-8720	323	32	made	make	VERB
ap-8720	323	33	up	up	ADP
ap-8720	323	34	of	of	ADP
ap-8720	323	35	real	real	ADJ
ap-8720	323	36	and	and	CCONJ
ap-8720	323	37	fake	fake	ADJ
ap-8720	323	38	faces	face	NOUN
ap-8720	323	39	(	(	PUNCT
ap-8720	323	40	figure	figure	NOUN
ap-8720	323	41	12	12	NUM
ap-8720	323	42	)	)	PUNCT
ap-8720	323	43	.	.	PUNCT
ap-8720	324	1	during	during	ADP
ap-8720	324	2	the	the	DET
ap-8720	324	3	training	training	NOUN
ap-8720	324	4	of	of	ADP
ap-8720	324	5	this	this	DET
ap-8720	324	6	network	network	NOUN
ap-8720	324	7	,	,	PUNCT
ap-8720	324	8	we	we	PRON
ap-8720	324	9	kept	keep	VERB
ap-8720	324	10	the	the	DET
ap-8720	324	11	following	follow	VERB
ap-8720	324	12	configuration	configuration	NOUN
ap-8720	324	13	:	:	PUNCT
ap-8720	324	14	optimizer	optimizer	NOUN
ap-8720	324	15	=	=	SYM
ap-8720	324	16	rmsprop	rmsprop	NOUN
ap-8720	324	17	.	.	PUNCT
ap-8720	325	1	batch	batch	NOUN
ap-8720	325	2	size	size	NOUN
ap-8720	325	3	=	=	SYM
ap-8720	325	4	128	128	NUM
ap-8720	325	5	.	.	PUNCT
ap-8720	325	6	number	number	NOUN
ap-8720	325	7	of	of	ADP
ap-8720	325	8	epochs	epoch	NOUN
ap-8720	325	9	=	=	SYM
ap-8720	325	10	30	30	X
ap-8720	325	11	.	.	PUNCT
ap-8720	326	1	figure	figure	NOUN
ap-8720	326	2	12	12	NUM
ap-8720	326	3	shows	show	VERB
ap-8720	326	4	the	the	DET
ap-8720	326	5	main	main	ADJ
ap-8720	326	6	training	training	NOUN
ap-8720	326	7	phases	phase	NOUN
ap-8720	326	8	of	of	ADP
ap-8720	326	9	the	the	DET
ap-8720	326	10	discriminator	discriminator	NOUN
ap-8720	326	11	.	.	PUNCT
ap-8720	327	1	the	the	DET
ap-8720	327	2	distribution	distribution	NOUN
ap-8720	327	3	of	of	ADP
ap-8720	327	4	the	the	DET
ap-8720	327	5	dataset	dataset	NOUN
ap-8720	327	6	for	for	ADP
ap-8720	327	7	training	training	NOUN
ap-8720	327	8	and	and	CCONJ
ap-8720	327	9	test	test	NOUN
ap-8720	327	10	is	be	AUX
ap-8720	327	11	described	describe	VERB
ap-8720	327	12	according	accord	VERB
ap-8720	327	13	to	to	ADP
ap-8720	327	14	table	table	NOUN
ap-8720	327	15	4	4	NUM
ap-8720	327	16	.	.	PUNCT
ap-8720	328	1	figure	figure	VERB
ap-8720	328	2	13	13	NUM
ap-8720	328	3	shows	show	VERB
ap-8720	328	4	the	the	DET
ap-8720	328	5	training	training	NOUN
ap-8720	328	6	results	result	NOUN
ap-8720	328	7	of	of	ADP
ap-8720	328	8	the	the	DET
ap-8720	328	9	discriminator	discriminator	NOUN
ap-8720	328	10	.	.	PUNCT
ap-8720	329	1	we	we	PRON
ap-8720	329	2	can	can	AUX
ap-8720	329	3	see	see	VERB
ap-8720	329	4	that	that	SCONJ
ap-8720	329	5	the	the	DET
ap-8720	329	6	network	network	NOUN
ap-8720	329	7	achieved	achieve	VERB
ap-8720	329	8	an	an	DET
ap-8720	329	9	accuracy	accuracy	NOUN
ap-8720	329	10	rate	rate	NOUN
ap-8720	329	11	of	of	ADP
ap-8720	329	12	98	98	NUM
ap-8720	329	13	%	%	NOUN
ap-8720	329	14	during	during	ADP
ap-8720	329	15	training	training	NOUN
ap-8720	329	16	and	and	CCONJ
ap-8720	329	17	94	94	NUM
ap-8720	329	18	%	%	NOUN
ap-8720	329	19	during	during	ADP
ap-8720	329	20	testing	testing	NOUN
ap-8720	329	21	,	,	PUNCT
ap-8720	329	22	with	with	ADP
ap-8720	329	23	a	a	DET
ap-8720	329	24	loss	loss	NOUN
ap-8720	329	25	of	of	ADP
ap-8720	329	26	0.05	0.05	NUM
ap-8720	329	27	%	%	NOUN
ap-8720	329	28	during	during	ADP
ap-8720	329	29	training	training	NOUN
ap-8720	329	30	and	and	CCONJ
ap-8720	329	31	0.12	0.12	NUM
ap-8720	329	32	%	%	NOUN
ap-8720	329	33	during	during	ADP
ap-8720	329	34	testing	testing	NOUN
ap-8720	329	35	,	,	PUNCT
ap-8720	329	36	which	which	PRON
ap-8720	329	37	is	be	AUX
ap-8720	329	38	considered	consider	VERB
ap-8720	329	39	as	as	ADP
ap-8720	329	40	acceptable	acceptable	ADJ
ap-8720	329	41	for	for	ADP
ap-8720	329	42	the	the	DET
ap-8720	329	43	detection	detection	NOUN
ap-8720	329	44	of	of	ADP
ap-8720	329	45	fake	fake	ADJ
ap-8720	329	46	faces	face	NOUN
ap-8720	329	47	.	.	PUNCT
ap-8720	330	1	4.3.3	4.3.3	X
ap-8720	330	2	.	.	PUNCT
ap-8720	331	1	training	training	NOUN
ap-8720	331	2	of	of	ADP
ap-8720	331	3	discriminator	discriminator	NOUN
ap-8720	331	4	3	3	NUM
ap-8720	331	5	the	the	DET
ap-8720	331	6	architecture	architecture	NOUN
ap-8720	331	7	of	of	ADP
ap-8720	331	8	discriminator	discriminator	NOUN
ap-8720	331	9	3	3	NUM
ap-8720	331	10	is	be	AUX
ap-8720	331	11	the	the	DET
ap-8720	331	12	same	same	ADJ
ap-8720	331	13	as	as	ADP
ap-8720	331	14	that	that	PRON
ap-8720	331	15	of	of	ADP
ap-8720	331	16	discriminator	discriminator	NOUN
ap-8720	331	17	1	1	NUM
ap-8720	331	18	,	,	PUNCT
ap-8720	331	19	trained	train	VERB
ap-8720	331	20	from	from	ADP
ap-8720	331	21	scratch	scratch	NOUN
ap-8720	331	22	on	on	ADP
ap-8720	331	23	real	real	ADJ
ap-8720	331	24	faces	face	NOUN
ap-8720	331	25	and	and	CCONJ
ap-8720	331	26	fake	fake	ADJ
ap-8720	331	27	faces	face	NOUN
ap-8720	331	28	without	without	ADP
ap-8720	331	29	its	its	PRON
ap-8720	331	30	generator	generator	NOUN
ap-8720	331	31	.	.	PUNCT
ap-8720	332	1	during	during	ADP
ap-8720	332	2	the	the	DET
ap-8720	332	3	training	training	NOUN
ap-8720	332	4	of	of	ADP
ap-8720	332	5	this	this	DET
ap-8720	332	6	network	network	NOUN
ap-8720	332	7	,	,	PUNCT
ap-8720	332	8	we	we	PRON
ap-8720	332	9	kept	keep	VERB
ap-8720	332	10	the	the	DET
ap-8720	332	11	following	follow	VERB
ap-8720	332	12	configuration	configuration	NOUN
ap-8720	332	13	:	:	PUNCT
ap-8720	332	14	optimizer	optimizer	NOUN
ap-8720	332	15	=	=	SYM
ap-8720	332	16	rmsprop	rmsprop	NOUN
ap-8720	332	17	.	.	PUNCT
ap-8720	333	1	batch	batch	NOUN
ap-8720	333	2	size	size	NOUN
ap-8720	333	3	=	=	SYM
ap-8720	333	4	128	128	NUM
ap-8720	333	5	.	.	PUNCT
ap-8720	333	6	number	number	NOUN
ap-8720	333	7	of	of	ADP
ap-8720	333	8	epochs	epoch	NOUN
ap-8720	333	9	=	=	SYM
ap-8720	333	10	45	45	NUM
ap-8720	333	11	.	.	PUNCT
ap-8720	333	12	figure	figure	NOUN
ap-8720	333	13	14	14	NUM
ap-8720	333	14	shows	show	VERB
ap-8720	333	15	the	the	DET
ap-8720	333	16	main	main	ADJ
ap-8720	333	17	training	training	NOUN
ap-8720	333	18	phases	phase	NOUN
ap-8720	333	19	of	of	ADP
ap-8720	333	20	the	the	DET
ap-8720	333	21	discriminator	discriminator	NOUN
ap-8720	333	22	3	3	NUM
ap-8720	333	23	.	.	PUNCT
ap-8720	334	1	the	the	DET
ap-8720	334	2	distribution	distribution	NOUN
ap-8720	334	3	of	of	ADP
ap-8720	334	4	the	the	DET
ap-8720	334	5	training	training	NOUN
ap-8720	334	6	dataset	dataset	NOUN
ap-8720	334	7	is	be	AUX
ap-8720	334	8	the	the	DET
ap-8720	334	9	same	same	ADJ
ap-8720	334	10	as	as	ADP
ap-8720	334	11	for	for	ADP
ap-8720	334	12	the	the	DET
ap-8720	334	13	discriminator	discriminator	NOUN
ap-8720	334	14	2	2	NUM
ap-8720	334	15	(	(	PUNCT
ap-8720	334	16	table	table	NOUN
ap-8720	334	17	4	4	NUM
ap-8720	334	18	)	)	PUNCT
ap-8720	334	19	.	.	PUNCT
ap-8720	335	1	from	from	ADP
ap-8720	335	2	the	the	DET
ap-8720	335	3	graphs	graph	NOUN
ap-8720	335	4	in	in	ADP
ap-8720	335	5	figure	figure	NOUN
ap-8720	335	6	15	15	NUM
ap-8720	335	7	,	,	PUNCT
ap-8720	335	8	we	we	PRON
ap-8720	335	9	can	can	AUX
ap-8720	335	10	see	see	VERB
ap-8720	335	11	that	that	SCONJ
ap-8720	335	12	the	the	DET
ap-8720	335	13	network	network	NOUN
ap-8720	335	14	achieved	achieve	VERB
ap-8720	335	15	an	an	DET
ap-8720	335	16	accuracy	accuracy	NOUN
ap-8720	335	17	rate	rate	NOUN
ap-8720	335	18	of	of	ADP
ap-8720	335	19	97	97	NUM
ap-8720	335	20	%	%	NOUN
ap-8720	335	21	during	during	ADP
ap-8720	335	22	training	training	NOUN
ap-8720	335	23	and	and	CCONJ
ap-8720	335	24	94	94	NUM
ap-8720	335	25	%	%	NOUN
ap-8720	335	26	during	during	ADP
ap-8720	335	27	testing	testing	NOUN
ap-8720	335	28	,	,	PUNCT
ap-8720	335	29	with	with	ADP
ap-8720	335	30	a	a	DET
ap-8720	335	31	loss	loss	NOUN
ap-8720	335	32	of	of	ADP
ap-8720	335	33	0.06	0.06	NUM
ap-8720	335	34	%	%	NOUN
ap-8720	335	35	during	during	ADP
ap-8720	335	36	training	training	NOUN
ap-8720	335	37	while	while	SCONJ
ap-8720	335	38	the	the	DET
ap-8720	335	39	estimated	estimate	VERB
ap-8720	335	40	test	test	NOUN
ap-8720	335	41	loss	loss	NOUN
ap-8720	335	42	remained	remain	VERB
ap-8720	335	43	within	within	ADP
ap-8720	335	44	an	an	DET
ap-8720	335	45	acceptable	acceptable	ADJ
ap-8720	335	46	range	range	NOUN
ap-8720	335	47	of	of	ADP
ap-8720	335	48	0.14	0.14	NUM
ap-8720	335	49	%	%	NOUN
ap-8720	335	50	for	for	ADP
ap-8720	335	51	effective	effective	ADJ
ap-8720	335	52	detection	detection	NOUN
ap-8720	335	53	of	of	ADP
ap-8720	335	54	fake	fake	ADJ
ap-8720	335	55	faces	face	NOUN
ap-8720	335	56	.	.	PUNCT
ap-8720	336	1	4.4	4.4	NUM
ap-8720	336	2	.	.	PUNCT
ap-8720	336	3	test	test	NOUN
ap-8720	336	4	and	and	CCONJ
ap-8720	336	5	analysis	analysis	NOUN
ap-8720	336	6	the	the	DET
ap-8720	336	7	tests	test	NOUN
ap-8720	336	8	with	with	ADP
ap-8720	336	9	the	the	DET
ap-8720	336	10	three	three	NUM
ap-8720	336	11	discriminators	discriminator	NOUN
ap-8720	336	12	were	be	AUX
ap-8720	336	13	carried	carry	VERB
ap-8720	336	14	out	out	ADP
ap-8720	336	15	with	with	ADP
ap-8720	336	16	datasets	dataset	NOUN
ap-8720	336	17	of	of	ADP
ap-8720	336	18	20	20	NUM
ap-8720	336	19	000	000	NUM
ap-8720	336	20	images	image	NOUN
ap-8720	336	21	(	(	PUNCT
ap-8720	336	22	fake	fake	ADJ
ap-8720	336	23	and	and	CCONJ
ap-8720	336	24	real	real	ADJ
ap-8720	336	25	)	)	PUNCT
ap-8720	336	26	,	,	PUNCT
ap-8720	336	27	that	that	PRON
ap-8720	336	28	had	have	AUX
ap-8720	336	29	not	not	PART
ap-8720	336	30	been	be	AUX
ap-8720	336	31	presented	present	VERB
ap-8720	336	32	to	to	ADP
ap-8720	336	33	the	the	DET
ap-8720	336	34	networks	network	NOUN
ap-8720	336	35	before	before	ADV
ap-8720	336	36	.	.	PUNCT
ap-8720	337	1	the	the	DET
ap-8720	337	2	results	result	NOUN
ap-8720	337	3	obtained	obtain	VERB
ap-8720	337	4	by	by	ADP
ap-8720	337	5	each	each	DET
ap-8720	337	6	network	network	NOUN
ap-8720	337	7	are	be	AUX
ap-8720	337	8	shown	show	VERB
ap-8720	337	9	in	in	ADP
ap-8720	337	10	tables	table	NOUN
ap-8720	337	11	5	5	NUM
ap-8720	337	12	.	.	PUNCT
ap-8720	337	13	table	table	NOUN
ap-8720	337	14	6	6	NUM
ap-8720	337	15	shows	show	VERB
ap-8720	337	16	the	the	DET
ap-8720	337	17	true	true	ADJ
ap-8720	337	18	and	and	CCONJ
ap-8720	337	19	false	false	ADJ
ap-8720	337	20	positive	positive	ADJ
ap-8720	337	21	and	and	CCONJ
ap-8720	337	22	negative	negative	ADJ
ap-8720	337	23	examples	example	NOUN
ap-8720	337	24	.	.	PUNCT
ap-8720	338	1	we	we	PRON
ap-8720	338	2	observed	observe	VERB
ap-8720	338	3	that	that	SCONJ
ap-8720	338	4	discriminator	discriminator	NOUN
ap-8720	338	5	2	2	NUM
ap-8720	338	6	is	be	AUX
ap-8720	338	7	314	314	NUM
ap-8720	338	8	vol	vol	NOUN
ap-8720	338	9	.	.	PUNCT
ap-8720	339	1	63	63	NUM
ap-8720	339	2	no	no	NOUN
ap-8720	339	3	.	.	PUNCT
ap-8720	340	1	5/2023	5/2023	NUM
ap-8720	340	2	fake	fake	ADJ
ap-8720	340	3	face	face	NOUN
ap-8720	340	4	detection	detection	NOUN
ap-8720	340	5	based	base	VERB
ap-8720	340	6	on	on	ADP
ap-8720	340	7	a	a	DET
ap-8720	340	8	multi	multi	ADJ
ap-8720	340	9	discriminator	discriminator	NOUN
ap-8720	340	10	deep	deep	ADJ
ap-8720	340	11	cnn	cnn	PROPN
ap-8720	340	12	architecture	architecture	NOUN
ap-8720	340	13	figure	figure	NOUN
ap-8720	340	14	11	11	NUM
ap-8720	340	15	.	.	PUNCT
ap-8720	341	1	discriminator	discriminator	NOUN
ap-8720	341	2	and	and	CCONJ
ap-8720	341	3	generator	generator	NOUN
ap-8720	341	4	loss	loss	NOUN
ap-8720	341	5	.	.	PUNCT
ap-8720	342	1	figure	figure	NOUN
ap-8720	342	2	12	12	NUM
ap-8720	342	3	.	.	PUNCT
ap-8720	343	1	training	training	NOUN
ap-8720	343	2	of	of	ADP
ap-8720	343	3	discriminator	discriminator	NOUN
ap-8720	343	4	2	2	NUM
ap-8720	343	5	.	.	PUNCT
ap-8720	343	6	better	well	ADJ
ap-8720	343	7	at	at	ADP
ap-8720	343	8	discriminating	discriminate	VERB
ap-8720	343	9	negative	negative	ADJ
ap-8720	343	10	examples	example	NOUN
ap-8720	343	11	(	(	PUNCT
ap-8720	343	12	fake	fake	ADJ
ap-8720	343	13	face	face	NOUN
ap-8720	343	14	)	)	PUNCT
ap-8720	343	15	with	with	ADP
ap-8720	343	16	a	a	DET
ap-8720	343	17	tn	tn	NOUN
ap-8720	343	18	of	of	ADP
ap-8720	343	19	98	98	NUM
ap-8720	343	20	%	%	NOUN
ap-8720	343	21	than	than	ADP
ap-8720	343	22	positive	positive	ADJ
ap-8720	343	23	examples	example	NOUN
ap-8720	343	24	(	(	PUNCT
ap-8720	343	25	real	real	ADJ
ap-8720	343	26	face	face	NOUN
ap-8720	343	27	)	)	PUNCT
ap-8720	343	28	with	with	ADP
ap-8720	343	29	a	a	DET
ap-8720	343	30	tp	tp	NOUN
ap-8720	343	31	of	of	ADP
ap-8720	343	32	92	92	NUM
ap-8720	343	33	%	%	NOUN
ap-8720	343	34	.	.	PUNCT
ap-8720	344	1	discriminator	discriminator	NOUN
ap-8720	344	2	3	3	NUM
ap-8720	344	3	is	be	AUX
ap-8720	344	4	then	then	ADV
ap-8720	344	5	better	well	ADJ
ap-8720	344	6	at	at	ADP
ap-8720	344	7	discriminating	discriminate	VERB
ap-8720	344	8	real	real	ADJ
ap-8720	344	9	faces	face	NOUN
ap-8720	344	10	,	,	PUNCT
ap-8720	344	11	with	with	ADP
ap-8720	344	12	a	a	DET
ap-8720	344	13	tp	tp	NOUN
ap-8720	344	14	of	of	ADP
ap-8720	344	15	98	98	NUM
ap-8720	344	16	%	%	NOUN
ap-8720	344	17	,	,	PUNCT
ap-8720	344	18	than	than	ADP
ap-8720	344	19	fake	fake	ADJ
ap-8720	344	20	faces	face	NOUN
ap-8720	344	21	,	,	PUNCT
ap-8720	344	22	with	with	ADP
ap-8720	344	23	a	a	DET
ap-8720	344	24	tn	tn	NOUN
ap-8720	344	25	of	of	ADP
ap-8720	344	26	93	93	NUM
ap-8720	344	27	%	%	NOUN
ap-8720	344	28	.	.	PUNCT
ap-8720	345	1	table	table	NOUN
ap-8720	345	2	7	7	NUM
ap-8720	345	3	shows	show	VERB
ap-8720	345	4	the	the	DET
ap-8720	345	5	precision	precision	NOUN
ap-8720	345	6	and	and	CCONJ
ap-8720	345	7	recall	recall	NOUN
ap-8720	345	8	of	of	ADP
ap-8720	345	9	the	the	DET
ap-8720	345	10	three	three	NUM
ap-8720	345	11	networks	network	NOUN
ap-8720	345	12	obtained	obtain	VERB
ap-8720	345	13	with	with	ADP
ap-8720	345	14	real	real	ADJ
ap-8720	345	15	and	and	CCONJ
ap-8720	345	16	fake	fake	ADJ
ap-8720	345	17	faces	face	NOUN
ap-8720	345	18	datasets	dataset	NOUN
ap-8720	345	19	.	.	PUNCT
ap-8720	346	1	figure	figure	NOUN
ap-8720	346	2	13	13	NUM
ap-8720	346	3	.	.	PUNCT
ap-8720	347	1	accuracy	accuracy	NOUN
ap-8720	347	2	and	and	CCONJ
ap-8720	347	3	loss	loss	NOUN
ap-8720	347	4	during	during	ADP
ap-8720	347	5	training	training	NOUN
ap-8720	347	6	and	and	CCONJ
ap-8720	347	7	validation	validation	NOUN
ap-8720	347	8	of	of	ADP
ap-8720	347	9	discriminator	discriminator	NOUN
ap-8720	347	10	2	2	NUM
ap-8720	347	11	.	.	PUNCT
ap-8720	348	1	the	the	DET
ap-8720	348	2	discriminator	discriminator	NOUN
ap-8720	348	3	2	2	NUM
ap-8720	348	4	recognises	recognise	VERB
ap-8720	348	5	the	the	DET
ap-8720	348	6	real	real	ADJ
ap-8720	348	7	faces	face	NOUN
ap-8720	348	8	with	with	ADP
ap-8720	348	9	an	an	DET
ap-8720	348	10	accuracy	accuracy	NOUN
ap-8720	348	11	of	of	ADP
ap-8720	348	12	0.98	0.98	NUM
ap-8720	348	13	,	,	PUNCT
ap-8720	348	14	exceeding	exceed	VERB
ap-8720	348	15	that	that	PRON
ap-8720	348	16	of	of	ADP
ap-8720	348	17	fake	fake	ADJ
ap-8720	348	18	faces	face	NOUN
ap-8720	348	19	by	by	ADP
ap-8720	348	20	0.05	0.05	NUM
ap-8720	348	21	.	.	PUNCT
ap-8720	349	1	conversely	conversely	ADV
ap-8720	349	2	,	,	PUNCT
ap-8720	349	3	for	for	ADP
ap-8720	349	4	discriminator	discriminator	NOUN
ap-8720	349	5	3	3	NUM
ap-8720	349	6	,	,	PUNCT
ap-8720	349	7	the	the	DET
ap-8720	349	8	recognition	recognition	NOUN
ap-8720	349	9	accuracy	accuracy	NOUN
ap-8720	349	10	of	of	ADP
ap-8720	349	11	fake	fake	ADJ
ap-8720	349	12	faces	face	NOUN
ap-8720	349	13	(	(	PUNCT
ap-8720	349	14	0.98	0.98	NUM
ap-8720	349	15	)	)	PUNCT
ap-8720	349	16	exceeds	exceed	VERB
ap-8720	349	17	that	that	PRON
ap-8720	349	18	of	of	ADP
ap-8720	349	19	the	the	DET
ap-8720	349	20	real	real	ADJ
ap-8720	349	21	faces	face	NOUN
ap-8720	349	22	(	(	PUNCT
ap-8720	349	23	0.05	0.05	NUM
ap-8720	349	24	)	)	PUNCT
ap-8720	349	25	.	.	PUNCT
ap-8720	350	1	with	with	ADP
ap-8720	350	2	the	the	DET
ap-8720	350	3	accuracies	accuracy	NOUN
ap-8720	350	4	obtained	obtain	VERB
ap-8720	350	5	,	,	PUNCT
ap-8720	350	6	the	the	DET
ap-8720	350	7	discriminators	discriminator	NOUN
ap-8720	350	8	of	of	ADP
ap-8720	350	9	our	our	PRON
ap-8720	350	10	system	system	NOUN
ap-8720	350	11	manage	manage	VERB
ap-8720	350	12	to	to	PART
ap-8720	350	13	recognise	recognise	VERB
ap-8720	350	14	more	more	ADJ
ap-8720	350	15	than	than	ADP
ap-8720	350	16	93	93	NUM
ap-8720	350	17	%	%	NOUN
ap-8720	350	18	of	of	ADP
ap-8720	350	19	the	the	DET
ap-8720	350	20	examples	example	NOUN
ap-8720	350	21	presented	present	VERB
ap-8720	350	22	.	.	PUNCT
ap-8720	351	1	it	it	PRON
ap-8720	351	2	should	should	AUX
ap-8720	351	3	be	be	AUX
ap-8720	351	4	noted	note	VERB
ap-8720	351	5	that	that	SCONJ
ap-8720	351	6	5	5	NUM
ap-8720	351	7	%	%	NOUN
ap-8720	351	8	of	of	ADP
ap-8720	351	9	the	the	DET
ap-8720	351	10	examples	example	NOUN
ap-8720	351	11	not	not	PART
ap-8720	351	12	recognised	recognise	VERB
ap-8720	351	13	by	by	ADP
ap-8720	351	14	one	one	NUM
ap-8720	351	15	of	of	ADP
ap-8720	351	16	the	the	DET
ap-8720	351	17	discriminators	discriminator	NOUN
ap-8720	351	18	are	be	AUX
ap-8720	351	19	recognised	recognise	VERB
ap-8720	351	20	by	by	ADP
ap-8720	351	21	the	the	DET
ap-8720	351	22	others	other	NOUN
ap-8720	351	23	,	,	PUNCT
ap-8720	351	24	which	which	PRON
ap-8720	351	25	is	be	AUX
ap-8720	351	26	advantageous	advantageous	ADJ
ap-8720	351	27	for	for	ADP
ap-8720	351	28	our	our	PRON
ap-8720	351	29	multi	multi	ADJ
ap-8720	351	30	-	-	NOUN
ap-8720	351	31	discriminator	discriminator	NOUN
ap-8720	351	32	because	because	SCONJ
ap-8720	351	33	its	its	PRON
ap-8720	351	34	final	final	ADJ
ap-8720	351	35	decision	decision	NOUN
ap-8720	351	36	is	be	AUX
ap-8720	351	37	obtained	obtain	VERB
ap-8720	351	38	by	by	ADP
ap-8720	351	39	voting	vote	VERB
ap-8720	351	40	between	between	ADP
ap-8720	351	41	the	the	DET
ap-8720	351	42	decisions	decision	NOUN
ap-8720	351	43	of	of	ADP
ap-8720	351	44	the	the	DET
ap-8720	351	45	three	three	NUM
ap-8720	351	46	discriminators	discriminator	NOUN
ap-8720	351	47	,	,	PUNCT
ap-8720	351	48	where	where	SCONJ
ap-8720	351	49	the	the	DET
ap-8720	351	50	final	final	ADJ
ap-8720	351	51	result	result	NOUN
ap-8720	351	52	is	be	AUX
ap-8720	351	53	obtained	obtain	VERB
ap-8720	351	54	by	by	ADP
ap-8720	351	55	a	a	DET
ap-8720	351	56	majority	majority	NOUN
ap-8720	351	57	vote	vote	NOUN
ap-8720	351	58	which	which	PRON
ap-8720	351	59	is	be	AUX
ap-8720	351	60	not	not	PART
ap-8720	351	61	always	always	ADV
ap-8720	351	62	possible	possible	ADJ
ap-8720	351	63	in	in	ADP
ap-8720	351	64	the	the	DET
ap-8720	351	65	case	case	NOUN
ap-8720	351	66	of	of	ADP
ap-8720	351	67	an	an	DET
ap-8720	351	68	even	even	ADJ
ap-8720	351	69	number	number	NOUN
ap-8720	351	70	of	of	ADP
ap-8720	351	71	discriminators	discriminator	NOUN
ap-8720	351	72	.	.	PUNCT
ap-8720	352	1	figure	figure	NOUN
ap-8720	352	2	16	16	NUM
ap-8720	352	3	shows	show	VERB
ap-8720	352	4	the	the	DET
ap-8720	352	5	roc	roc	PROPN
ap-8720	352	6	curves	curve	NOUN
ap-8720	352	7	of	of	ADP
ap-8720	352	8	the	the	DET
ap-8720	352	9	three	three	NUM
ap-8720	352	10	discriminators	discriminator	NOUN
ap-8720	352	11	and	and	CCONJ
ap-8720	352	12	the	the	DET
ap-8720	352	13	multi	multi	ADJ
ap-8720	352	14	-	-	NOUN
ap-8720	352	15	discriminator	discriminator	ADJ
ap-8720	352	16	(	(	PUNCT
ap-8720	352	17	mdd	mdd	PROPN
ap-8720	352	18	-	-	PUNCT
ap-8720	352	19	cnn	cnn	PROPN
ap-8720	352	20	)	)	PUNCT
ap-8720	352	21	allowing	allow	VERB
ap-8720	352	22	the	the	DET
ap-8720	352	23	performances	performance	NOUN
ap-8720	352	24	of	of	ADP
ap-8720	352	25	the	the	DET
ap-8720	352	26	four	four	NUM
ap-8720	352	27	architectures	architecture	NOUN
ap-8720	352	28	to	to	PART
ap-8720	352	29	be	be	AUX
ap-8720	352	30	compared	compare	VERB
ap-8720	352	31	.	.	PUNCT
ap-8720	353	1	we	we	PRON
ap-8720	353	2	can	can	AUX
ap-8720	353	3	see	see	VERB
ap-8720	353	4	that	that	SCONJ
ap-8720	353	5	the	the	DET
ap-8720	353	6	mdd	mdd	PROPN
ap-8720	353	7	-	-	PUNCT
ap-8720	353	8	cnn	cnn	PROPN
ap-8720	353	9	curve	curve	NOUN
ap-8720	353	10	deviates	deviate	VERB
ap-8720	353	11	slightly	slightly	ADV
ap-8720	353	12	from	from	ADP
ap-8720	353	13	discriminator	discriminator	NOUN
ap-8720	353	14	2	2	NUM
ap-8720	353	15	(	(	PUNCT
ap-8720	353	16	best	good	ADJ
ap-8720	353	17	discriminator	discriminator	NOUN
ap-8720	353	18	of	of	ADP
ap-8720	353	19	fakes	fake	NOUN
ap-8720	353	20	)	)	PUNCT
ap-8720	353	21	and	and	CCONJ
ap-8720	353	22	occupies	occupy	VERB
ap-8720	353	23	the	the	DET
ap-8720	353	24	upper	upper	ADJ
ap-8720	353	25	part	part	NOUN
ap-8720	353	26	of	of	ADP
ap-8720	353	27	the	the	DET
ap-8720	353	28	roc	roc	PROPN
ap-8720	353	29	space	space	NOUN
ap-8720	353	30	indicating	indicate	VERB
ap-8720	353	31	the	the	DET
ap-8720	353	32	best	good	ADJ
ap-8720	353	33	classifier	classifier	NOUN
ap-8720	353	34	among	among	ADP
ap-8720	353	35	the	the	DET
ap-8720	353	36	four	four	NUM
ap-8720	353	37	with	with	ADP
ap-8720	353	38	a	a	DET
ap-8720	353	39	tpr	tpr	NOUN
ap-8720	353	40	around	around	ADP
ap-8720	353	41	97.8	97.8	NUM
ap-8720	353	42	%	%	NOUN
ap-8720	353	43	.	.	PUNCT
ap-8720	354	1	315	315	NUM
ap-8720	354	2	ch	ch	NOUN
ap-8720	354	3	.	.	PUNCT
ap-8720	354	4	e.	e.	PROPN
ap-8720	354	5	bencheriet	bencheriet	PROPN
ap-8720	354	6	,	,	PUNCT
ap-8720	354	7	h.	h.	PROPN
ap-8720	354	8	abdelmoumène	abdelmoumène	PROPN
ap-8720	354	9	,	,	PUNCT
ap-8720	354	10	a.	a.	PROPN
ap-8720	354	11	sebbagh	sebbagh	PROPN
ap-8720	354	12	et	et	PROPN
ap-8720	354	13	al	al	PROPN
ap-8720	354	14	.	.	PROPN
ap-8720	354	15	acta	acta	PROPN
ap-8720	354	16	polytechnica	polytechnica	PROPN
ap-8720	354	17	figure	figure	NOUN
ap-8720	354	18	14	14	NUM
ap-8720	354	19	.	.	PUNCT
ap-8720	355	1	training	training	NOUN
ap-8720	355	2	of	of	ADP
ap-8720	355	3	discriminator	discriminator	NOUN
ap-8720	355	4	3	3	NUM
ap-8720	355	5	.	.	PUNCT
ap-8720	355	6	figure	figure	NOUN
ap-8720	355	7	15	15	NUM
ap-8720	355	8	.	.	PUNCT
ap-8720	356	1	accuracy	accuracy	NOUN
ap-8720	356	2	and	and	CCONJ
ap-8720	356	3	loss	loss	NOUN
ap-8720	356	4	during	during	ADP
ap-8720	356	5	training	training	NOUN
ap-8720	356	6	and	and	CCONJ
ap-8720	356	7	validation	validation	NOUN
ap-8720	356	8	of	of	ADP
ap-8720	356	9	discriminator	discriminator	NOUN
ap-8720	356	10	3	3	NUM
ap-8720	356	11	.	.	PUNCT
ap-8720	357	1	in	in	ADP
ap-8720	357	2	order	order	NOUN
ap-8720	357	3	to	to	PART
ap-8720	357	4	test	test	VERB
ap-8720	357	5	the	the	DET
ap-8720	357	6	performance	performance	NOUN
ap-8720	357	7	of	of	ADP
ap-8720	357	8	our	our	PRON
ap-8720	357	9	mdd	mdd	PROPN
ap-8720	357	10	-	-	PUNCT
ap-8720	357	11	cnn	cnn	PROPN
ap-8720	357	12	architecture	architecture	NOUN
ap-8720	357	13	,	,	PUNCT
ap-8720	357	14	we	we	PRON
ap-8720	357	15	have	have	AUX
ap-8720	357	16	carried	carry	VERB
ap-8720	357	17	out	out	ADP
ap-8720	357	18	several	several	ADJ
ap-8720	357	19	tests	test	NOUN
ap-8720	357	20	on	on	ADP
ap-8720	357	21	images	image	NOUN
ap-8720	357	22	of	of	ADP
ap-8720	357	23	real	real	ADJ
ap-8720	357	24	faces	face	NOUN
ap-8720	357	25	and	and	CCONJ
ap-8720	357	26	false	false	ADJ
ap-8720	357	27	faces	face	NOUN
ap-8720	357	28	randomly	randomly	ADV
ap-8720	357	29	introduced	introduce	VERB
ap-8720	357	30	into	into	ADP
ap-8720	357	31	the	the	DET
ap-8720	357	32	system	system	NOUN
ap-8720	357	33	,	,	PUNCT
ap-8720	357	34	the	the	DET
ap-8720	357	35	results	result	NOUN
ap-8720	357	36	are	be	AUX
ap-8720	357	37	shown	show	VERB
ap-8720	357	38	in	in	ADP
ap-8720	357	39	figure	figure	NOUN
ap-8720	357	40	17	17	NUM
ap-8720	357	41	with	with	ADP
ap-8720	357	42	an	an	DET
ap-8720	357	43	indication	indication	NOUN
ap-8720	357	44	for	for	ADP
ap-8720	357	45	each	each	DET
ap-8720	357	46	image	image	NOUN
ap-8720	357	47	of	of	ADP
ap-8720	357	48	the	the	DET
ap-8720	357	49	prediction	prediction	NOUN
ap-8720	357	50	and	and	CCONJ
ap-8720	357	51	the	the	DET
ap-8720	357	52	ground	ground	NOUN
ap-8720	357	53	truth	truth	NOUN
ap-8720	357	54	class	class	NOUN
ap-8720	357	55	(	(	PUNCT
ap-8720	357	56	true	true	ADJ
ap-8720	357	57	or	or	CCONJ
ap-8720	357	58	false	false	ADJ
ap-8720	357	59	faces	face	NOUN
ap-8720	357	60	)	)	PUNCT
ap-8720	357	61	.	.	PUNCT
ap-8720	358	1	thanks	thank	NOUN
ap-8720	358	2	to	to	ADP
ap-8720	358	3	the	the	DET
ap-8720	358	4	combination	combination	NOUN
ap-8720	358	5	of	of	ADP
ap-8720	358	6	the	the	DET
ap-8720	358	7	performance	performance	NOUN
ap-8720	358	8	of	of	ADP
ap-8720	358	9	the	the	DET
ap-8720	358	10	three	three	NUM
ap-8720	358	11	discriminators	discriminator	NOUN
ap-8720	358	12	,	,	PUNCT
ap-8720	358	13	the	the	DET
ap-8720	358	14	mdd	mdd	PROPN
ap-8720	358	15	-	-	PUNCT
ap-8720	358	16	cnn	cnn	PROPN
ap-8720	358	17	architecture	architecture	NOUN
ap-8720	358	18	has	have	AUX
ap-8720	358	19	given	give	VERB
ap-8720	358	20	high	high	ADJ
ap-8720	358	21	discrimination	discrimination	NOUN
ap-8720	358	22	rates	rate	NOUN
ap-8720	358	23	(	(	PUNCT
ap-8720	358	24	97.8	97.8	NUM
ap-8720	358	25	%	%	NOUN
ap-8720	358	26	)	)	PUNCT
ap-8720	358	27	thus	thus	ADV
ap-8720	358	28	exceeding	exceed	VERB
ap-8720	358	29	the	the	DET
ap-8720	358	30	state	state	NOUN
ap-8720	358	31	of	of	ADP
ap-8720	358	32	the	the	DET
ap-8720	358	33	art	art	NOUN
ap-8720	358	34	with	with	ADP
ap-8720	358	35	a	a	DET
ap-8720	358	36	minimum	minimum	ADJ
ap-8720	358	37	deviation	deviation	NOUN
ap-8720	358	38	of	of	ADP
ap-8720	358	39	0.8	0.8	NUM
ap-8720	358	40	%	%	NOUN
ap-8720	358	41	and	and	CCONJ
ap-8720	358	42	a	a	DET
ap-8720	358	43	maximum	maximum	NOUN
ap-8720	358	44	of	of	ADP
ap-8720	358	45	17	17	NUM
ap-8720	358	46	%	%	NOUN
ap-8720	358	47	(	(	PUNCT
ap-8720	358	48	table	table	NOUN
ap-8720	358	49	8)	8)	NUM
ap-8720	358	50	.	.	NOUN
ap-8720	358	51	5	5	NUM
ap-8720	358	52	.	.	X
ap-8720	358	53	conclusion	conclusion	NOUN
ap-8720	358	54	in	in	ADP
ap-8720	358	55	recent	recent	ADJ
ap-8720	358	56	years	year	NOUN
ap-8720	358	57	,	,	PUNCT
ap-8720	358	58	deepfake	deepfake	NOUN
ap-8720	358	59	technologies	technology	NOUN
ap-8720	358	60	have	have	AUX
ap-8720	358	61	gained	gain	VERB
ap-8720	358	62	significant	significant	ADJ
ap-8720	358	63	popularity	popularity	NOUN
ap-8720	358	64	and	and	CCONJ
ap-8720	358	65	pose	pose	VERB
ap-8720	358	66	a	a	DET
ap-8720	358	67	threat	threat	NOUN
ap-8720	358	68	to	to	ADP
ap-8720	358	69	our	our	PRON
ap-8720	358	70	security	security	NOUN
ap-8720	358	71	as	as	SCONJ
ap-8720	358	72	they	they	PRON
ap-8720	358	73	can	can	AUX
ap-8720	358	74	be	be	AUX
ap-8720	358	75	exploited	exploit	VERB
ap-8720	358	76	by	by	ADP
ap-8720	358	77	criminals	criminal	NOUN
ap-8720	358	78	for	for	ADP
ap-8720	358	79	fraudulent	fraudulent	ADJ
ap-8720	358	80	activities	activity	NOUN
ap-8720	358	81	.	.	PUNCT
ap-8720	359	1	consequently	consequently	ADV
ap-8720	359	2	,	,	PUNCT
ap-8720	359	3	the	the	DET
ap-8720	359	4	detection	detection	NOUN
ap-8720	359	5	of	of	ADP
ap-8720	359	6	fake	fake	ADJ
ap-8720	359	7	content	content	NOUN
ap-8720	359	8	has	have	AUX
ap-8720	359	9	become	become	VERB
ap-8720	359	10	a	a	DET
ap-8720	359	11	crucial	crucial	ADJ
ap-8720	359	12	research	research	NOUN
ap-8720	359	13	area	area	NOUN
ap-8720	359	14	,	,	PUNCT
ap-8720	359	15	given	give	VERB
ap-8720	359	16	the	the	DET
ap-8720	359	17	proliferation	proliferation	NOUN
ap-8720	359	18	of	of	ADP
ap-8720	359	19	manipulated	manipulate	VERB
ap-8720	359	20	or	or	CCONJ
ap-8720	359	21	fabricated	fabricate	VERB
ap-8720	359	22	data	datum	NOUN
ap-8720	359	23	in	in	ADP
ap-8720	359	24	various	various	ADJ
ap-8720	359	25	forms	form	NOUN
ap-8720	359	26	,	,	PUNCT
ap-8720	359	27	such	such	ADJ
ap-8720	359	28	as	as	ADP
ap-8720	359	29	images	image	NOUN
ap-8720	359	30	,	,	PUNCT
ap-8720	359	31	videos	video	NOUN
ap-8720	359	32	,	,	PUNCT
ap-8720	359	33	and	and	CCONJ
ap-8720	359	34	sounds	sound	VERB
ap-8720	359	35	.	.	PUNCT
ap-8720	360	1	distinguishing	distinguish	VERB
ap-8720	360	2	fake	fake	ADJ
ap-8720	360	3	faces	face	NOUN
ap-8720	360	4	from	from	ADP
ap-8720	360	5	real	real	ADJ
ap-8720	360	6	ones	one	NOUN
ap-8720	360	7	is	be	AUX
ap-8720	360	8	both	both	CCONJ
ap-8720	360	9	a	a	DET
ap-8720	360	10	challenging	challenging	ADJ
ap-8720	360	11	and	and	CCONJ
ap-8720	360	12	a	a	DET
ap-8720	360	13	delicate	delicate	ADJ
ap-8720	360	14	task	task	NOUN
ap-8720	360	15	.	.	PUNCT
ap-8720	361	1	it	it	PRON
ap-8720	361	2	requires	require	VERB
ap-8720	361	3	a	a	DET
ap-8720	361	4	highperformance	highperformance	NOUN
ap-8720	361	5	discrimination	discrimination	NOUN
ap-8720	361	6	system	system	NOUN
ap-8720	361	7	with	with	ADP
ap-8720	361	8	considerable	considerable	ADJ
ap-8720	361	9	discriminator	discriminator	NOUN
ap-8720	361	10	1	1	NUM
ap-8720	361	11	tp	tp	NOUN
ap-8720	361	12	fn	fn	PROPN
ap-8720	361	13	tn	tn	PROPN
ap-8720	361	14	fp	fp	PROPN
ap-8720	361	15	image	image	NOUN
ap-8720	361	16	8483	8483	NUM
ap-8720	361	17	1517	1517	NUM
ap-8720	361	18	7013	7013	NUM
ap-8720	361	19	2987count	2987count	NUM
ap-8720	361	20	ratio	ratio	NOUN
ap-8720	361	21	on	on	ADP
ap-8720	361	22	0.8483	0.8483	NUM
ap-8720	361	23	0.1517	0.1517	NUM
ap-8720	361	24	0.7013	0.7013	NUM
ap-8720	361	25	0.2987total	0.2987total	ADJ
ap-8720	361	26	count	count	NOUN
ap-8720	361	27	percentage	percentage	NOUN
ap-8720	361	28	84.8	84.8	NUM
ap-8720	361	29	%	%	NOUN
ap-8720	361	30	15.2	15.2	NUM
ap-8720	361	31	%	%	NOUN
ap-8720	361	32	70.1	70.1	NUM
ap-8720	361	33	%	%	NOUN
ap-8720	361	34	29.9	29.9	NUM
ap-8720	361	35	%	%	NOUN
ap-8720	361	36	discriminator	discriminator	NOUN
ap-8720	361	37	2	2	NUM
ap-8720	361	38	tp	tp	NOUN
ap-8720	361	39	fn	fn	PROPN
ap-8720	361	40	tn	tn	PROPN
ap-8720	361	41	fp	fp	PROPN
ap-8720	361	42	image	image	NOUN
ap-8720	361	43	9823	9823	NUM
ap-8720	361	44	177	177	NUM
ap-8720	361	45	9205	9205	NUM
ap-8720	361	46	795count	795count	PROPN
ap-8720	361	47	ratio	ratio	NOUN
ap-8720	361	48	on	on	ADP
ap-8720	361	49	0.9823	0.9823	NUM
ap-8720	361	50	0.0177	0.0177	NUM
ap-8720	361	51	0.9205	0.9205	NUM
ap-8720	361	52	0.0795total	0.0795total	NUM
ap-8720	361	53	count	count	NOUN
ap-8720	361	54	percentage	percentage	NOUN
ap-8720	361	55	98	98	NUM
ap-8720	361	56	%	%	NOUN
ap-8720	361	57	2	2	NUM
ap-8720	361	58	%	%	NOUN
ap-8720	361	59	92	92	NUM
ap-8720	361	60	%	%	NOUN
ap-8720	361	61	8	8	NUM
ap-8720	361	62	%	%	NOUN
ap-8720	361	63	discriminator	discriminator	NOUN
ap-8720	361	64	3	3	NUM
ap-8720	361	65	tp	tp	ADP
ap-8720	361	66	fn	fn	PROPN
ap-8720	361	67	tn	tn	PROPN
ap-8720	361	68	fp	fp	PROPN
ap-8720	361	69	image	image	NOUN
ap-8720	361	70	9763	9763	NUM
ap-8720	361	71	237	237	NUM
ap-8720	361	72	9277	9277	NUM
ap-8720	361	73	723count	723count	PROPN
ap-8720	361	74	ratio	ratio	NOUN
ap-8720	361	75	on	on	ADP
ap-8720	361	76	0.9763	0.9763	NUM
ap-8720	361	77	0.02	0.02	NUM
ap-8720	361	78	0.9277	0.9277	NUM
ap-8720	361	79	0.0723total	0.0723total	ADJ
ap-8720	361	80	count	count	NOUN
ap-8720	361	81	percentage	percentage	NOUN
ap-8720	361	82	98	98	NUM
ap-8720	361	83	%	%	NOUN
ap-8720	361	84	2	2	NUM
ap-8720	361	85	%	%	NOUN
ap-8720	361	86	93	93	NUM
ap-8720	361	87	%	%	NOUN
ap-8720	361	88	7	7	NUM
ap-8720	361	89	%	%	NOUN
ap-8720	361	90	table	table	NOUN
ap-8720	361	91	5	5	NUM
ap-8720	361	92	.	.	PUNCT
ap-8720	361	93	performance	performance	NOUN
ap-8720	361	94	of	of	ADP
ap-8720	361	95	discriminators	discriminator	NOUN
ap-8720	361	96	1	1	NUM
ap-8720	361	97	,	,	PUNCT
ap-8720	361	98	2	2	NUM
ap-8720	361	99	and	and	CCONJ
ap-8720	361	100	3	3	NUM
ap-8720	361	101	.	.	X
ap-8720	362	1	tp	tp	ADP
ap-8720	362	2	tn	tn	PROPN
ap-8720	362	3	fp	fp	PROPN
ap-8720	362	4	fn	fn	PROPN
ap-8720	362	5	discriminator	discriminator	NOUN
ap-8720	362	6	1	1	NUM
ap-8720	362	7	84.8	84.8	NUM
ap-8720	362	8	%	%	NOUN
ap-8720	362	9	15.2	15.2	NUM
ap-8720	362	10	%	%	NOUN
ap-8720	362	11	70.1	70.1	NUM
ap-8720	362	12	%	%	NOUN
ap-8720	362	13	29.9	29.9	NUM
ap-8720	362	14	%	%	NOUN
ap-8720	362	15	discriminator	discriminator	NOUN
ap-8720	362	16	2	2	NUM
ap-8720	362	17	92	92	NUM
ap-8720	362	18	%	%	NOUN
ap-8720	362	19	98	98	NUM
ap-8720	362	20	%	%	NOUN
ap-8720	362	21	2	2	NUM
ap-8720	362	22	%	%	NOUN
ap-8720	362	23	8	8	NUM
ap-8720	362	24	%	%	NOUN
ap-8720	362	25	discriminator	discriminator	NOUN
ap-8720	362	26	3	3	NUM
ap-8720	362	27	98	98	NUM
ap-8720	362	28	%	%	NOUN
ap-8720	362	29	93	93	NUM
ap-8720	362	30	%	%	NOUN
ap-8720	362	31	7	7	NUM
ap-8720	362	32	%	%	NOUN
ap-8720	362	33	2	2	NUM
ap-8720	362	34	%	%	NOUN
ap-8720	362	35	table	table	NOUN
ap-8720	362	36	6	6	NUM
ap-8720	362	37	.	.	PUNCT
ap-8720	362	38	performance	performance	NOUN
ap-8720	362	39	comparisons	comparison	NOUN
ap-8720	362	40	of	of	ADP
ap-8720	362	41	the	the	DET
ap-8720	362	42	three	three	NUM
ap-8720	362	43	discriminators	discriminator	NOUN
ap-8720	362	44	.	.	PUNCT
ap-8720	363	1	accuracy	accuracy	NOUN
ap-8720	363	2	to	to	PART
ap-8720	363	3	distinguish	distinguish	VERB
ap-8720	363	4	fake	fake	ADJ
ap-8720	363	5	faces	face	NOUN
ap-8720	363	6	from	from	ADP
ap-8720	363	7	real	real	ADJ
ap-8720	363	8	ones	one	NOUN
ap-8720	363	9	.	.	PUNCT
ap-8720	364	1	this	this	DET
ap-8720	364	2	need	need	NOUN
ap-8720	364	3	led	lead	VERB
ap-8720	364	4	to	to	ADP
ap-8720	364	5	the	the	DET
ap-8720	364	6	development	development	NOUN
ap-8720	364	7	of	of	ADP
ap-8720	364	8	the	the	DET
ap-8720	364	9	multi	multi	ADJ
ap-8720	364	10	-	-	ADJ
ap-8720	364	11	discriminator	discriminator	ADJ
ap-8720	364	12	architecture	architecture	NOUN
ap-8720	364	13	proposed	propose	VERB
ap-8720	364	14	in	in	ADP
ap-8720	364	15	this	this	DET
ap-8720	364	16	study	study	NOUN
ap-8720	364	17	.	.	PUNCT
ap-8720	365	1	this	this	DET
ap-8720	365	2	architecture	architecture	NOUN
ap-8720	365	3	consists	consist	VERB
ap-8720	365	4	mainly	mainly	ADV
ap-8720	365	5	of	of	ADP
ap-8720	365	6	three	three	NUM
ap-8720	365	7	discriminators	discriminator	NOUN
ap-8720	365	8	whose	whose	DET
ap-8720	365	9	main	main	ADJ
ap-8720	365	10	task	task	NOUN
ap-8720	365	11	is	be	AUX
ap-8720	365	12	to	to	PART
ap-8720	365	13	detect	detect	VERB
ap-8720	365	14	false	false	ADJ
ap-8720	365	15	faces	face	NOUN
ap-8720	365	16	.	.	PUNCT
ap-8720	366	1	each	each	DET
ap-8720	366	2	network	network	NOUN
ap-8720	366	3	provides	provide	VERB
ap-8720	366	4	a	a	DET
ap-8720	366	5	decision	decision	NOUN
ap-8720	366	6	weighted	weight	VERB
ap-8720	366	7	by	by	ADP
ap-8720	366	8	the	the	DET
ap-8720	366	9	probability	probability	NOUN
ap-8720	366	10	of	of	ADP
ap-8720	366	11	belonging	belong	VERB
ap-8720	366	12	to	to	ADP
ap-8720	366	13	the	the	DET
ap-8720	366	14	class	class	NOUN
ap-8720	366	15	of	of	ADP
ap-8720	366	16	real	real	ADJ
ap-8720	366	17	or	or	CCONJ
ap-8720	366	18	fake	fake	ADJ
ap-8720	366	19	faces	face	NOUN
ap-8720	366	20	.	.	PUNCT
ap-8720	367	1	the	the	DET
ap-8720	367	2	decision	decision	NOUN
ap-8720	367	3	of	of	ADP
ap-8720	367	4	each	each	DET
ap-8720	367	5	network	network	NOUN
ap-8720	367	6	represents	represent	VERB
ap-8720	367	7	a	a	DET
ap-8720	367	8	vote	vote	NOUN
ap-8720	367	9	that	that	PRON
ap-8720	367	10	will	will	AUX
ap-8720	367	11	contribute	contribute	VERB
ap-8720	367	12	to	to	ADP
ap-8720	367	13	the	the	DET
ap-8720	367	14	voting	voting	NOUN
ap-8720	367	15	of	of	ADP
ap-8720	367	16	all	all	DET
ap-8720	367	17	the	the	DET
ap-8720	367	18	networks	network	NOUN
ap-8720	367	19	to	to	PART
ap-8720	367	20	provide	provide	VERB
ap-8720	367	21	a	a	DET
ap-8720	367	22	final	final	ADJ
ap-8720	367	23	decision	decision	NOUN
ap-8720	367	24	(	(	PUNCT
ap-8720	367	25	an	an	DET
ap-8720	367	26	opinion	opinion	NOUN
ap-8720	367	27	)	)	PUNCT
ap-8720	367	28	on	on	ADP
ap-8720	367	29	the	the	DET
ap-8720	367	30	class	class	NOUN
ap-8720	367	31	of	of	ADP
ap-8720	367	32	the	the	DET
ap-8720	367	33	image	image	NOUN
ap-8720	367	34	.	.	PUNCT
ap-8720	368	1	the	the	DET
ap-8720	368	2	three	three	NUM
ap-8720	368	3	discriminators	discriminator	NOUN
ap-8720	368	4	have	have	AUX
ap-8720	368	5	been	be	AUX
ap-8720	368	6	trained	train	VERB
ap-8720	368	7	differently	differently	ADV
ap-8720	368	8	,	,	PUNCT
ap-8720	368	9	whith	whith	ADP
ap-8720	368	10	discriminator	discriminator	NOUN
ap-8720	368	11	1	1	NUM
ap-8720	368	12	being	be	AUX
ap-8720	368	13	part	part	NOUN
ap-8720	368	14	of	of	ADP
ap-8720	368	15	the	the	DET
ap-8720	368	16	architecture	architecture	NOUN
ap-8720	368	17	that	that	PRON
ap-8720	368	18	we	we	PRON
ap-8720	368	19	have	have	AUX
ap-8720	368	20	proposed	propose	VERB
ap-8720	368	21	for	for	ADP
ap-8720	368	22	the	the	DET
ap-8720	368	23	new	new	ADJ
ap-8720	368	24	dcgan	dcgan	NOUN
ap-8720	368	25	resulting	result	VERB
ap-8720	368	26	from	from	ADP
ap-8720	368	27	the	the	DET
ap-8720	368	28	transformation	transformation	NOUN
ap-8720	368	29	of	of	ADP
ap-8720	368	30	a	a	DET
ap-8720	368	31	gan	gan	ADJ
ap-8720	368	32	network	network	NOUN
ap-8720	368	33	from	from	ADP
ap-8720	368	34	a	a	DET
ap-8720	368	35	generator	generator	NOUN
ap-8720	368	36	of	of	ADP
ap-8720	368	37	fake	fake	ADJ
ap-8720	368	38	faces	face	NOUN
ap-8720	368	39	,	,	PUNCT
ap-8720	368	40	with	with	ADP
ap-8720	368	41	a	a	DET
ap-8720	368	42	generator	generator	NOUN
ap-8720	368	43	that	that	PRON
ap-8720	368	44	manages	manage	VERB
ap-8720	368	45	to	to	PART
ap-8720	368	46	deceive	deceive	VERB
ap-8720	368	47	its	its	PRON
ap-8720	368	48	discriminator	discriminator	NOUN
ap-8720	368	49	to	to	ADP
ap-8720	368	50	a	a	DET
ap-8720	368	51	detector	detector	NOUN
ap-8720	368	52	of	of	ADP
ap-8720	368	53	fakes	fake	NOUN
ap-8720	368	54	,	,	PUNCT
ap-8720	368	55	with	with	ADP
ap-8720	368	56	a	a	DET
ap-8720	368	57	discriminator	discriminator	NOUN
ap-8720	368	58	that	that	PRON
ap-8720	368	59	manages	manage	VERB
ap-8720	368	60	to	to	PART
ap-8720	368	61	recognise	recognise	VERB
ap-8720	368	62	the	the	DET
ap-8720	368	63	fake	fake	ADJ
ap-8720	368	64	faces	face	NOUN
ap-8720	368	65	of	of	ADP
ap-8720	368	66	the	the	DET
ap-8720	368	67	generator	generator	NOUN
ap-8720	368	68	.	.	PUNCT
ap-8720	369	1	although	although	SCONJ
ap-8720	369	2	the	the	DET
ap-8720	369	3	discriminator	discriminator	NOUN
ap-8720	369	4	of	of	ADP
ap-8720	369	5	the	the	DET
ap-8720	369	6	ndcgan	ndcgan	ADJ
ap-8720	369	7	network	network	NOUN
ap-8720	369	8	gave	give	VERB
ap-8720	369	9	good	good	ADJ
ap-8720	369	10	results	result	NOUN
ap-8720	369	11	with	with	ADP
ap-8720	369	12	a	a	DET
ap-8720	369	13	precision	precision	NOUN
ap-8720	369	14	of	of	ADP
ap-8720	369	15	79	79	NUM
ap-8720	369	16	%	%	NOUN
ap-8720	369	17	,	,	PUNCT
ap-8720	369	18	it	it	PRON
ap-8720	369	19	is	be	AUX
ap-8720	369	20	still	still	ADV
ap-8720	369	21	insufficient	insufficient	ADJ
ap-8720	369	22	as	as	SCONJ
ap-8720	369	23	the	the	DET
ap-8720	369	24	loss	loss	NOUN
ap-8720	369	25	exceeds	exceed	VERB
ap-8720	369	26	40	40	NUM
ap-8720	369	27	%	%	NOUN
ap-8720	369	28	.	.	PUNCT
ap-8720	370	1	the	the	DET
ap-8720	370	2	loss	loss	NOUN
ap-8720	370	3	was	be	AUX
ap-8720	370	4	significantly	significantly	ADV
ap-8720	370	5	reduced	reduce	VERB
ap-8720	370	6	by	by	ADP
ap-8720	370	7	transfer	transfer	NOUN
ap-8720	370	8	learning	learning	NOUN
ap-8720	370	9	,	,	PUNCT
ap-8720	370	10	which	which	PRON
ap-8720	370	11	was	be	AUX
ap-8720	370	12	used	use	VERB
ap-8720	370	13	in	in	ADP
ap-8720	370	14	the	the	DET
ap-8720	370	15	case	case	NOUN
ap-8720	370	16	of	of	ADP
ap-8720	370	17	discriminator	discriminator	NOUN
ap-8720	370	18	2	2	NUM
ap-8720	370	19	,	,	PUNCT
ap-8720	370	20	resulting	result	VERB
ap-8720	370	21	in	in	ADP
ap-8720	370	22	the	the	DET
ap-8720	370	23	model	model	NOUN
ap-8720	370	24	having	have	VERB
ap-8720	370	25	the	the	DET
ap-8720	370	26	316	316	NUM
ap-8720	370	27	vol	vol	NOUN
ap-8720	370	28	.	.	PUNCT
ap-8720	371	1	63	63	NUM
ap-8720	371	2	no	no	NOUN
ap-8720	371	3	.	.	PUNCT
ap-8720	372	1	5/2023	5/2023	NUM
ap-8720	372	2	fake	fake	ADJ
ap-8720	372	3	face	face	NOUN
ap-8720	372	4	detection	detection	NOUN
ap-8720	372	5	based	base	VERB
ap-8720	372	6	on	on	ADP
ap-8720	372	7	a	a	DET
ap-8720	372	8	multi	multi	ADJ
ap-8720	372	9	discriminator	discriminator	NOUN
ap-8720	372	10	deep	deep	ADJ
ap-8720	372	11	cnn	cnn	PROPN
ap-8720	372	12	architecture	architecture	NOUN
ap-8720	372	13	support	support	NOUN
ap-8720	372	14	precision	precision	NOUN
ap-8720	372	15	recall	recall	NOUN
ap-8720	372	16	discriminator	discriminator	NOUN
ap-8720	372	17	1	1	NUM
ap-8720	372	18	fake	fake	ADJ
ap-8720	372	19	10	10	NUM
ap-8720	372	20	000	000	NUM
ap-8720	372	21	0.79	0.79	NUM
ap-8720	372	22	0.77	0.77	NUM
ap-8720	372	23	real	real	ADJ
ap-8720	372	24	10	10	NUM
ap-8720	372	25	000	000	NUM
ap-8720	372	26	0.80	0.80	NUM
ap-8720	372	27	0.82	0.82	NUM
ap-8720	372	28	discriminator	discriminator	NOUN
ap-8720	372	29	2	2	NUM
ap-8720	372	30	fake	fake	ADJ
ap-8720	372	31	10	10	NUM
ap-8720	372	32	000	000	NUM
ap-8720	372	33	0.93	0.93	NUM
ap-8720	372	34	0.98	0.98	NUM
ap-8720	372	35	real	real	ADJ
ap-8720	372	36	10	10	NUM
ap-8720	372	37	000	000	NUM
ap-8720	372	38	0.98	0.98	NUM
ap-8720	372	39	0.92	0.92	NUM
ap-8720	372	40	discriminator	discriminator	NOUN
ap-8720	372	41	3	3	NUM
ap-8720	372	42	fake	fake	ADJ
ap-8720	372	43	10	10	NUM
ap-8720	372	44	000	000	NUM
ap-8720	372	45	0.98	0.98	NUM
ap-8720	372	46	0.93	0.93	NUM
ap-8720	372	47	real	real	ADJ
ap-8720	372	48	10	10	NUM
ap-8720	372	49	000	000	NUM
ap-8720	372	50	0.93	0.93	NUM
ap-8720	372	51	0.98	0.98	NUM
ap-8720	372	52	table	table	NOUN
ap-8720	372	53	7	7	NUM
ap-8720	372	54	.	.	PUNCT
ap-8720	372	55	precision	precision	NOUN
ap-8720	372	56	and	and	CCONJ
ap-8720	372	57	recall	recall	NOUN
ap-8720	372	58	of	of	ADP
ap-8720	372	59	the	the	DET
ap-8720	372	60	three	three	NUM
ap-8720	372	61	discriminators	discriminator	NOUN
ap-8720	372	62	.	.	PUNCT
ap-8720	373	1	reference	reference	NOUN
ap-8720	373	2	dataset	dataset	NOUN
ap-8720	373	3	accuracy	accuracy	NOUN
ap-8720	373	4	z.	z.	PROPN
ap-8720	373	5	liu	liu	PROPN
ap-8720	373	6	et	et	PROPN
ap-8720	373	7	al	al	PROPN
ap-8720	373	8	.	.	PUNCT
ap-8720	374	1	[	[	X
ap-8720	374	2	17	17	NUM
ap-8720	374	3	]	]	PUNCT
ap-8720	374	4	stylegan	stylegan	NOUN
ap-8720	374	5	80.72	80.72	NUM
ap-8720	374	6	%	%	NOUN
ap-8720	374	7	celeba	celeba	PROPN
ap-8720	374	8	-	-	PUNCT
ap-8720	374	9	hq	hq	PROPN
ap-8720	374	10	t.	t.	PROPN
ap-8720	374	11	jung	jung	PROPN
ap-8720	374	12	et	et	PROPN
ap-8720	374	13	al	al	PROPN
ap-8720	374	14	.	.	PUNCT
ap-8720	375	1	[	[	X
ap-8720	375	2	22	22	NUM
ap-8720	375	3	]	]	X
ap-8720	375	4	gan	gan	X
ap-8720	375	5	df	df	PROPN
ap-8720	375	6	87.5	87.5	NUM
ap-8720	375	7	%	%	NOUN
ap-8720	375	8	p.	p.	PROPN
ap-8720	375	9	korshunov	korshunov	PROPN
ap-8720	375	10	&	&	CCONJ
ap-8720	375	11	s.	s.	PROPN
ap-8720	375	12	marcel	marcel	PROPN
ap-8720	376	1	[	[	X
ap-8720	376	2	23	23	NUM
ap-8720	376	3	]	]	X
ap-8720	376	4	vid	vid	NOUN
ap-8720	376	5	timit	timit	NOUN
ap-8720	376	6	91.03	91.03	NUM
ap-8720	376	7	%	%	NOUN
ap-8720	376	8	y.	y.	PROPN
ap-8720	376	9	li	li	PROPN
ap-8720	376	10	&	&	CCONJ
ap-8720	376	11	s.	s.	PROPN
ap-8720	376	12	lyu	lyu	X
ap-8720	377	1	[	[	X
ap-8720	377	2	26	26	NUM
ap-8720	377	3	]	]	X
ap-8720	377	4	deep	deep	ADJ
ap-8720	377	5	fake	fake	ADJ
ap-8720	377	6	uadfv	uadfv	NOUN
ap-8720	377	7	97	97	NUM
ap-8720	377	8	%	%	NOUN
ap-8720	377	9	x.	x.	NOUN
ap-8720	377	10	yang	yang	PROPN
ap-8720	377	11	et	et	PROPN
ap-8720	377	12	al	al	PROPN
ap-8720	377	13	.	.	PUNCT
ap-8720	378	1	[	[	X
ap-8720	378	2	27	27	NUM
ap-8720	378	3	]	]	X
ap-8720	378	4	hff	hff	X
ap-8720	378	5	95.17	95.17	NUM
ap-8720	378	6	%	%	NOUN
ap-8720	378	7	a.	a.	NOUN
ap-8720	378	8	chintha	chintha	PROPN
ap-8720	378	9	et	et	PROPN
ap-8720	378	10	al	al	PROPN
ap-8720	378	11	.	.	PUNCT
ap-8720	379	1	[	[	X
ap-8720	379	2	28	28	NUM
ap-8720	379	3	]	]	PUNCT
ap-8720	379	4	face	face	NOUN
ap-8720	379	5	forensics	forensic	NOUN
ap-8720	379	6	84.8	84.8	NUM
ap-8720	379	7	%	%	NOUN
ap-8720	379	8	celeb	celeb	PROPN
ap-8720	379	9	df	df	PROPN
ap-8720	379	10	s.	s.	PROPN
ap-8720	379	11	lee	lee	PROPN
ap-8720	379	12	et	et	PROPN
ap-8720	379	13	al	al	PROPN
ap-8720	379	14	.	.	PUNCT
ap-8720	380	1	[	[	X
ap-8720	380	2	30	30	NUM
ap-8720	380	3	]	]	X
ap-8720	380	4	gan	gin	VERB
ap-8720	380	5	df	df	PROPN
ap-8720	380	6	93.99	93.99	NUM
ap-8720	380	7	%	%	NOUN
ap-8720	380	8	our	our	PRON
ap-8720	380	9	work	work	NOUN
ap-8720	380	10	(	(	PUNCT
ap-8720	380	11	mdd	mdd	PROPN
ap-8720	380	12	-	-	PUNCT
ap-8720	380	13	cnn	cnn	PROPN
ap-8720	380	14	)	)	PUNCT
ap-8720	380	15	flickr	flickr	PROPN
ap-8720	380	16	-	-	PUNCT
ap-8720	380	17	faces	faces	PROPN
ap-8720	380	18	-	-	PUNCT
ap-8720	380	19	hq	hq	NOUN
ap-8720	380	20	97.8	97.8	NUM
ap-8720	380	21	%	%	NOUN
ap-8720	380	22	nvidia	nvidia	PROPN
ap-8720	380	23	style	style	NOUN
ap-8720	380	24	gan	gan	X
ap-8720	380	25	ff	ff	NOUN
ap-8720	380	26	table	table	NOUN
ap-8720	380	27	8	8	NUM
ap-8720	380	28	.	.	PUNCT
ap-8720	380	29	comparison	comparison	NOUN
ap-8720	380	30	of	of	ADP
ap-8720	380	31	mdd	mdd	PROPN
ap-8720	380	32	-	-	PUNCT
ap-8720	380	33	cnn	cnn	PROPN
ap-8720	380	34	performance	performance	NOUN
ap-8720	380	35	with	with	ADP
ap-8720	380	36	state	state	NOUN
ap-8720	380	37	of	of	ADP
ap-8720	380	38	the	the	DET
ap-8720	380	39	art	art	NOUN
ap-8720	380	40	.	.	PUNCT
ap-8720	381	1	figure	figure	NOUN
ap-8720	381	2	16	16	NUM
ap-8720	381	3	.	.	PUNCT
ap-8720	382	1	roc	roc	PROPN
ap-8720	382	2	curve	curve	NOUN
ap-8720	382	3	of	of	ADP
ap-8720	382	4	the	the	DET
ap-8720	382	5	four	four	NUM
ap-8720	382	6	classifiers	classifier	NOUN
ap-8720	382	7	.	.	PUNCT
ap-8720	383	1	best	good	ADJ
ap-8720	383	2	fake	fake	ADJ
ap-8720	383	3	face	face	NOUN
ap-8720	383	4	detection	detection	NOUN
ap-8720	383	5	rate	rate	NOUN
ap-8720	383	6	(	(	PUNCT
ap-8720	383	7	98	98	NUM
ap-8720	383	8	%	%	NOUN
ap-8720	383	9	)	)	PUNCT
ap-8720	383	10	.	.	PUNCT
ap-8720	384	1	for	for	ADP
ap-8720	384	2	the	the	DET
ap-8720	384	3	detection	detection	NOUN
ap-8720	384	4	of	of	ADP
ap-8720	384	5	real	real	ADJ
ap-8720	384	6	faces	face	NOUN
ap-8720	384	7	,	,	PUNCT
ap-8720	384	8	the	the	DET
ap-8720	384	9	best	good	ADJ
ap-8720	384	10	rates	rate	NOUN
ap-8720	384	11	were	be	AUX
ap-8720	384	12	obtained	obtain	VERB
ap-8720	384	13	by	by	ADP
ap-8720	384	14	discriminator	discriminator	NOUN
ap-8720	384	15	3	3	NUM
ap-8720	384	16	(	(	PUNCT
ap-8720	384	17	98	98	NUM
ap-8720	384	18	%	%	NOUN
ap-8720	384	19	)	)	PUNCT
ap-8720	384	20	,	,	PUNCT
ap-8720	384	21	which	which	PRON
ap-8720	384	22	underwent	undergo	VERB
ap-8720	384	23	supervised	supervised	ADJ
ap-8720	384	24	training	training	NOUN
ap-8720	384	25	with	with	ADP
ap-8720	384	26	positive	positive	ADJ
ap-8720	384	27	and	and	CCONJ
ap-8720	384	28	negative	negative	ADJ
ap-8720	384	29	examples	example	NOUN
ap-8720	384	30	.	.	PUNCT
ap-8720	385	1	the	the	DET
ap-8720	385	2	performance	performance	NOUN
ap-8720	385	3	of	of	ADP
ap-8720	385	4	the	the	DET
ap-8720	385	5	three	three	NUM
ap-8720	385	6	discriminators	discriminator	NOUN
ap-8720	385	7	combined	combine	VERB
ap-8720	385	8	in	in	ADP
ap-8720	385	9	the	the	DET
ap-8720	385	10	multi	multi	ADJ
ap-8720	385	11	-	-	ADJ
ap-8720	385	12	discriminator	discriminator	ADJ
ap-8720	385	13	architecture	architecture	NOUN
ap-8720	385	14	allowed	allow	VERB
ap-8720	385	15	us	we	PRON
ap-8720	385	16	to	to	PART
ap-8720	385	17	obtain	obtain	VERB
ap-8720	385	18	remarkable	remarkable	ADJ
ap-8720	385	19	detection	detection	NOUN
ap-8720	385	20	rates	rate	NOUN
ap-8720	385	21	(	(	PUNCT
ap-8720	385	22	around	around	ADV
ap-8720	385	23	98	98	NUM
ap-8720	385	24	%	%	NOUN
ap-8720	385	25	)	)	PUNCT
ap-8720	385	26	for	for	ADP
ap-8720	385	27	the	the	DET
ap-8720	385	28	two	two	NUM
ap-8720	385	29	classes	class	NOUN
ap-8720	385	30	.	.	PUNCT
ap-8720	386	1	the	the	DET
ap-8720	386	2	results	result	NOUN
ap-8720	386	3	obtained	obtain	VERB
ap-8720	386	4	with	with	ADP
ap-8720	386	5	the	the	DET
ap-8720	386	6	proposed	propose	VERB
ap-8720	386	7	architecture	architecture	NOUN
ap-8720	386	8	(	(	PUNCT
ap-8720	386	9	mdd	mdd	PROPN
ap-8720	386	10	-	-	PUNCT
ap-8720	386	11	cnn	cnn	PROPN
ap-8720	386	12	)	)	PUNCT
ap-8720	386	13	outperform	outperform	VERB
ap-8720	386	14	the	the	DET
ap-8720	386	15	state	state	NOUN
ap-8720	386	16	of	of	ADP
ap-8720	386	17	the	the	DET
ap-8720	386	18	art	art	NOUN
ap-8720	386	19	by	by	ADP
ap-8720	386	20	0.8	0.8	NUM
ap-8720	386	21	%	%	NOUN
ap-8720	386	22	–	–	PUNCT
ap-8720	386	23	17	17	NUM
ap-8720	386	24	%	%	NOUN
ap-8720	386	25	.	.	PUNCT
ap-8720	387	1	these	these	DET
ap-8720	387	2	results	result	NOUN
ap-8720	387	3	remain	remain	VERB
ap-8720	387	4	preliminary	preliminary	ADJ
ap-8720	387	5	and	and	CCONJ
ap-8720	387	6	need	need	VERB
ap-8720	387	7	to	to	PART
ap-8720	387	8	be	be	AUX
ap-8720	387	9	deepened	deepen	VERB
ap-8720	387	10	by	by	ADP
ap-8720	387	11	testing	test	VERB
ap-8720	387	12	our	our	PRON
ap-8720	387	13	architecture	architecture	NOUN
ap-8720	387	14	on	on	ADP
ap-8720	387	15	other	other	ADJ
ap-8720	387	16	fake	fake	ADJ
ap-8720	387	17	datasets	dataset	NOUN
ap-8720	387	18	.	.	PUNCT
ap-8720	388	1	in	in	ADP
ap-8720	388	2	addition	addition	NOUN
ap-8720	388	3	to	to	ADP
ap-8720	388	4	this	this	DET
ap-8720	388	5	study	study	NOUN
ap-8720	388	6	,	,	PUNCT
ap-8720	388	7	in	in	ADP
ap-8720	388	8	the	the	DET
ap-8720	388	9	future	future	NOUN
ap-8720	388	10	we	we	PRON
ap-8720	388	11	plan	plan	VERB
ap-8720	388	12	to	to	PART
ap-8720	388	13	conduct	conduct	VERB
ap-8720	388	14	various	various	ADJ
ap-8720	388	15	tests	test	NOUN
ap-8720	388	16	on	on	ADP
ap-8720	388	17	more	more	ADJ
ap-8720	388	18	datasets	dataset	NOUN
ap-8720	388	19	of	of	ADP
ap-8720	388	20	images	image	NOUN
ap-8720	388	21	and	and	CCONJ
ap-8720	388	22	videos	video	NOUN
ap-8720	388	23	containing	contain	VERB
ap-8720	388	24	fake	fake	ADJ
ap-8720	388	25	faces	face	NOUN
ap-8720	388	26	in	in	ADP
ap-8720	388	27	order	order	NOUN
ap-8720	388	28	to	to	PART
ap-8720	388	29	implement	implement	VERB
ap-8720	388	30	a	a	DET
ap-8720	388	31	system	system	NOUN
ap-8720	388	32	for	for	ADP
ap-8720	388	33	detecting	detect	VERB
ap-8720	388	34	fakes	fake	NOUN
ap-8720	388	35	in	in	ADP
ap-8720	388	36	images	image	NOUN
ap-8720	388	37	and	and	CCONJ
ap-8720	388	38	videos	video	NOUN
ap-8720	388	39	manipulated	manipulate	VERB
ap-8720	388	40	in	in	ADP
ap-8720	388	41	social	social	ADJ
ap-8720	388	42	media	medium	NOUN
ap-8720	388	43	.	.	PUNCT
ap-8720	389	1	figure	figure	NOUN
ap-8720	389	2	17	17	NUM
ap-8720	389	3	.	.	PUNCT
ap-8720	389	4	obtained	obtain	VERB
ap-8720	389	5	results	result	NOUN
ap-8720	389	6	of	of	ADP
ap-8720	389	7	our	our	PRON
ap-8720	389	8	mdd	mdd	PROPN
ap-8720	389	9	-	-	PUNCT
ap-8720	389	10	cnn	cnn	PROPN
ap-8720	389	11	.	.	PUNCT
ap-8720	390	1	acknowledgements	acknowledgement	NOUN
ap-8720	390	2	this	this	DET
ap-8720	390	3	project	project	NOUN
ap-8720	390	4	is	be	AUX
ap-8720	390	5	supported	support	VERB
ap-8720	390	6	by	by	ADP
ap-8720	390	7	the	the	DET
ap-8720	390	8	laig	laig	PROPN
ap-8720	390	9	laboratory	laboratory	NOUN
ap-8720	390	10	and	and	CCONJ
ap-8720	390	11	the	the	DET
ap-8720	390	12	8	8	NUM
ap-8720	390	13	mai	mai	PROPN
ap-8720	390	14	1945	1945	NUM
ap-8720	390	15	-	-	PUNCT
ap-8720	390	16	university	university	NOUN
ap-8720	390	17	of	of	ADP
ap-8720	390	18	guelma	guelma	PROPN
ap-8720	390	19	–	–	PUNCT
ap-8720	390	20	http://www	http://www	PROPN
ap-8720	390	21	.	.	PUNCT
ap-8720	390	22	univ-guelma.dz	univ-guelma.dz	PROPN
ap-8720	390	23	.	.	PUNCT
ap-8720	391	1	references	reference	NOUN
ap-8720	391	2	[	[	X
ap-8720	391	3	1	1	NUM
ap-8720	391	4	]	]	PUNCT
ap-8720	391	5	c.	c.	NOUN
ap-8720	391	6	bregler	bregler	NOUN
ap-8720	391	7	,	,	PUNCT
ap-8720	391	8	m.	m.	NOUN
ap-8720	391	9	covell	covell	PROPN
ap-8720	391	10	,	,	PUNCT
ap-8720	391	11	m.	m.	NOUN
ap-8720	391	12	slaney	slaney	PROPN
ap-8720	391	13	.	.	PUNCT
ap-8720	392	1	video	video	PROPN
ap-8720	392	2	rewrite	rewrite	PROPN
ap-8720	392	3	:	:	PUNCT
ap-8720	392	4	driving	drive	VERB
ap-8720	392	5	visual	visual	ADJ
ap-8720	392	6	speech	speech	NOUN
ap-8720	392	7	with	with	ADP
ap-8720	392	8	audio	audio	NOUN
ap-8720	392	9	.	.	PUNCT
ap-8720	393	1	in	in	ADP
ap-8720	393	2	proceedings	proceeding	NOUN
ap-8720	393	3	of	of	ADP
ap-8720	393	4	the	the	DET
ap-8720	393	5	24th	24th	ADJ
ap-8720	393	6	annual	annual	ADJ
ap-8720	393	7	conference	conference	NOUN
ap-8720	393	8	on	on	ADP
ap-8720	393	9	computer	computer	NOUN
ap-8720	393	10	graphics	graphic	NOUN
ap-8720	393	11	and	and	CCONJ
ap-8720	393	12	interactive	interactive	ADJ
ap-8720	393	13	techniques	technique	NOUN
ap-8720	393	14	,	,	PUNCT
ap-8720	393	15	pp	pp	ADJ
ap-8720	393	16	.	.	PUNCT
ap-8720	394	1	353–360	353–360	NUM
ap-8720	394	2	.	.	PUNCT
ap-8720	395	1	1997	1997	NUM
ap-8720	395	2	.	.	PUNCT
ap-8720	396	1	https://doi.org/10.1145/258734.258880	https://doi.org/10.1145/258734.258880	PRON
ap-8720	396	2	[	[	X
ap-8720	396	3	2	2	X
ap-8720	396	4	]	]	PUNCT
ap-8720	396	5	t.	t.	PROPN
ap-8720	396	6	f.	f.	PROPN
ap-8720	396	7	cootes	cootes	PROPN
ap-8720	396	8	,	,	PUNCT
ap-8720	396	9	g.	g.	PROPN
ap-8720	396	10	j.	j.	PROPN
ap-8720	396	11	edwards	edwards	PROPN
ap-8720	396	12	,	,	PUNCT
ap-8720	396	13	c.	c.	PROPN
ap-8720	396	14	j.	j.	PROPN
ap-8720	396	15	taylor	taylor	PROPN
ap-8720	396	16	.	.	PUNCT
ap-8720	397	1	active	active	ADJ
ap-8720	397	2	appearance	appearance	NOUN
ap-8720	397	3	models	model	NOUN
ap-8720	397	4	.	.	PUNCT
ap-8720	398	1	ieee	ieee	NOUN
ap-8720	398	2	transactions	transaction	NOUN
ap-8720	398	3	on	on	ADP
ap-8720	398	4	pattern	pattern	NOUN
ap-8720	398	5	317	317	NUM
ap-8720	398	6	http://www.univ-guelma.dz	http://www.univ-guelma.dz	PROPN
ap-8720	398	7	http://www.univ-guelma.dz	http://www.univ-guelma.dz	PROPN
ap-8720	398	8	https://doi.org/10.1145/258734.258880	https://doi.org/10.1145/258734.258880	PROPN
ap-8720	398	9	ch	ch	PROPN
ap-8720	398	10	.	.	PUNCT
ap-8720	399	1	e.	e.	PROPN
ap-8720	399	2	bencheriet	bencheriet	PROPN
ap-8720	399	3	,	,	PUNCT
ap-8720	399	4	h.	h.	PROPN
ap-8720	399	5	abdelmoumène	abdelmoumène	PROPN
ap-8720	399	6	,	,	PUNCT
ap-8720	399	7	a.	a.	PROPN
ap-8720	399	8	sebbagh	sebbagh	PROPN
ap-8720	399	9	et	et	PROPN
ap-8720	400	1	al	al	PROPN
ap-8720	400	2	.	.	PROPN
ap-8720	400	3	acta	acta	PROPN
ap-8720	400	4	polytechnica	polytechnica	PROPN
ap-8720	400	5	analysis	analysis	NOUN
ap-8720	400	6	and	and	CCONJ
ap-8720	400	7	machine	machine	NOUN
ap-8720	400	8	intelligence	intelligence	NOUN
ap-8720	400	9	23(6):681–685	23(6):681–685	PROPN
ap-8720	400	10	,	,	PUNCT
ap-8720	400	11	2001	2001	NUM
ap-8720	400	12	.	.	PUNCT
ap-8720	401	1	https://doi.org/10.1109/34.927467	https://doi.org/10.1109/34.927467	PROPN
ap-8720	402	1	[	[	X
ap-8720	402	2	3	3	NUM
ap-8720	402	3	]	]	PUNCT
ap-8720	402	4	a.	a.	NOUN
ap-8720	402	5	santha	santha	PROPN
ap-8720	402	6	.	.	PUNCT
ap-8720	403	1	deepfakes	deepfake	VERB
ap-8720	403	2	generation	generation	NOUN
ap-8720	403	3	using	use	VERB
ap-8720	403	4	lstm	lstm	NOUN
ap-8720	403	5	based	base	VERB
ap-8720	403	6	generative	generative	ADJ
ap-8720	403	7	adversarial	adversarial	ADJ
ap-8720	403	8	networks	network	NOUN
ap-8720	403	9	.	.	PUNCT
ap-8720	404	1	master	master	NOUN
ap-8720	404	2	thesis	thesis	NOUN
ap-8720	404	3	,	,	PUNCT
ap-8720	404	4	rochester	rochester	PROPN
ap-8720	404	5	institute	institute	PROPN
ap-8720	404	6	of	of	ADP
ap-8720	404	7	technology	technology	PROPN
ap-8720	404	8	,	,	PUNCT
ap-8720	404	9	2020	2020	NUM
ap-8720	404	10	.	.	PUNCT
ap-8720	405	1	[	[	X
ap-8720	405	2	4	4	X
ap-8720	405	3	]	]	PUNCT
ap-8720	405	4	j.	j.	PROPN
ap-8720	405	5	thies	thies	PROPN
ap-8720	405	6	,	,	PUNCT
ap-8720	405	7	m.	m.	NOUN
ap-8720	405	8	zollhofer	zollhofer	NOUN
ap-8720	405	9	,	,	PUNCT
ap-8720	405	10	m.	m.	NOUN
ap-8720	405	11	stamminger	stamminger	PROPN
ap-8720	405	12	,	,	PUNCT
ap-8720	405	13	et	et	PROPN
ap-8720	405	14	al	al	PROPN
ap-8720	405	15	.	.	PROPN
ap-8720	405	16	face2face	face2face	PROPN
ap-8720	405	17	:	:	PUNCT
ap-8720	405	18	real	real	ADJ
ap-8720	405	19	-	-	PUNCT
ap-8720	405	20	time	time	NOUN
ap-8720	405	21	face	face	NOUN
ap-8720	405	22	capture	capture	NOUN
ap-8720	405	23	and	and	CCONJ
ap-8720	405	24	reenactment	reenactment	NOUN
ap-8720	405	25	of	of	ADP
ap-8720	405	26	rgb	rgb	PROPN
ap-8720	405	27	videos	video	NOUN
ap-8720	405	28	.	.	PUNCT
ap-8720	406	1	in	in	ADP
ap-8720	406	2	ieee	ieee	NOUN
ap-8720	406	3	conference	conference	NOUN
ap-8720	406	4	on	on	ADP
ap-8720	406	5	computer	computer	NOUN
ap-8720	406	6	vision	vision	NOUN
ap-8720	406	7	and	and	CCONJ
ap-8720	406	8	pattern	pattern	NOUN
ap-8720	406	9	recognition	recognition	NOUN
ap-8720	406	10	(	(	PUNCT
ap-8720	406	11	cvpr	cvpr	NOUN
ap-8720	406	12	)	)	PUNCT
ap-8720	406	13	,	,	PUNCT
ap-8720	406	14	pp	pp	ADP
ap-8720	406	15	.	.	PUNCT
ap-8720	407	1	2387–2395	2387–2395	NUM
ap-8720	407	2	.	.	PUNCT
ap-8720	408	1	2016	2016	NUM
ap-8720	408	2	.	.	PUNCT
ap-8720	409	1	https://doi.org/10.1109/cvpr.2016.262	https://doi.org/10.1109/cvpr.2016.262	NOUN
ap-8720	409	2	[	[	X
ap-8720	409	3	5	5	X
ap-8720	409	4	]	]	PUNCT
ap-8720	409	5	s.	s.	PROPN
ap-8720	409	6	suwajanakorn	suwajanakorn	VERB
ap-8720	409	7	,	,	PUNCT
ap-8720	409	8	s.	s.	PROPN
ap-8720	409	9	m.	m.	PROPN
ap-8720	409	10	seitz	seitz	PROPN
ap-8720	409	11	,	,	PUNCT
ap-8720	409	12	i.	i.	PROPN
ap-8720	409	13	kemelmachershlizerman	kemelmachershlizerman	PROPN
ap-8720	409	14	.	.	PUNCT
ap-8720	410	1	synthesizing	synthesize	VERB
ap-8720	410	2	obama	obama	NOUN
ap-8720	410	3	:	:	PUNCT
ap-8720	410	4	learning	learn	VERB
ap-8720	410	5	lip	lip	NOUN
ap-8720	410	6	sync	sync	NOUN
ap-8720	410	7	from	from	ADP
ap-8720	410	8	audio	audio	PROPN
ap-8720	410	9	.	.	PUNCT
ap-8720	410	10	acm	acm	PROPN
ap-8720	410	11	transactions	transaction	NOUN
ap-8720	410	12	on	on	ADP
ap-8720	410	13	graphics	graphic	NOUN
ap-8720	410	14	36(4):1–13	36(4):1–13	NUM
ap-8720	410	15	,	,	PUNCT
ap-8720	410	16	2017	2017	NUM
ap-8720	410	17	.	.	PUNCT
ap-8720	411	1	https://doi.org/10.1145/3072959.3073640	https://doi.org/10.1145/3072959.3073640	ADJ
ap-8720	411	2	[	[	X
ap-8720	411	3	6	6	NUM
ap-8720	411	4	]	]	PUNCT
ap-8720	411	5	m.	m.	NOUN
ap-8720	411	6	westerlund	westerlund	PROPN
ap-8720	411	7	.	.	PUNCT
ap-8720	412	1	the	the	DET
ap-8720	412	2	emergence	emergence	NOUN
ap-8720	412	3	of	of	ADP
ap-8720	412	4	deepfake	deepfake	NOUN
ap-8720	412	5	technology	technology	NOUN
ap-8720	412	6	:	:	PUNCT
ap-8720	412	7	a	a	DET
ap-8720	412	8	review	review	NOUN
ap-8720	412	9	.	.	PUNCT
ap-8720	413	1	technology	technology	NOUN
ap-8720	413	2	innovation	innovation	NOUN
ap-8720	413	3	management	management	PROPN
ap-8720	413	4	review	review	NOUN
ap-8720	413	5	9(11):39–52	9(11):39–52	NUM
ap-8720	413	6	,	,	PUNCT
ap-8720	413	7	2019	2019	NUM
ap-8720	413	8	.	.	PUNCT
ap-8720	414	1	https://doi.org/10.22215/timreview/1282	https://doi.org/10.22215/timreview/1282	PROPN
ap-8720	414	2	[	[	X
ap-8720	414	3	7	7	X
ap-8720	414	4	]	]	X
ap-8720	414	5	f.	f.	PROPN
ap-8720	414	6	marra	marra	PROPN
ap-8720	414	7	,	,	PUNCT
ap-8720	414	8	d.	d.	PROPN
ap-8720	414	9	gragnaniello	gragnaniello	PROPN
ap-8720	414	10	,	,	PUNCT
ap-8720	414	11	l.	l.	PROPN
ap-8720	414	12	verdoliva	verdoliva	PROPN
ap-8720	414	13	,	,	PUNCT
ap-8720	414	14	g.	g.	PROPN
ap-8720	414	15	poggi	poggi	PROPN
ap-8720	414	16	.	.	PUNCT
ap-8720	415	1	do	do	AUX
ap-8720	415	2	gans	gan	NOUN
ap-8720	415	3	leave	leave	VERB
ap-8720	415	4	artificial	artificial	ADJ
ap-8720	415	5	fingerprints	fingerprint	NOUN
ap-8720	415	6	?	?	PUNCT
ap-8720	416	1	in	in	ADP
ap-8720	416	2	ieee	ieee	NOUN
ap-8720	416	3	conference	conference	NOUN
ap-8720	416	4	on	on	ADP
ap-8720	416	5	multimedia	multimedia	NOUN
ap-8720	416	6	information	information	NOUN
ap-8720	416	7	processing	processing	NOUN
ap-8720	416	8	and	and	CCONJ
ap-8720	416	9	retrieval	retrieval	NOUN
ap-8720	416	10	(	(	PUNCT
ap-8720	416	11	mipr	mipr	PROPN
ap-8720	416	12	)	)	PUNCT
ap-8720	416	13	,	,	PUNCT
ap-8720	416	14	pp	pp	ADP
ap-8720	416	15	.	.	PUNCT
ap-8720	417	1	506–511	506–511	NUM
ap-8720	417	2	.	.	PUNCT
ap-8720	417	3	2019	2019	NUM
ap-8720	417	4	.	.	PUNCT
ap-8720	418	1	https://doi.org/10.1109/mipr.2019.00103	https://doi.org/10.1109/mipr.2019.00103	NOUN
ap-8720	418	2	[	[	X
ap-8720	418	3	8	8	NUM
ap-8720	418	4	]	]	PUNCT
ap-8720	418	5	a.	a.	NOUN
ap-8720	418	6	rössler	rössler	NOUN
ap-8720	418	7	,	,	PUNCT
ap-8720	418	8	d.	d.	PROPN
ap-8720	418	9	cozzolino	cozzolino	PROPN
ap-8720	418	10	,	,	PUNCT
ap-8720	418	11	l.	l.	PROPN
ap-8720	418	12	verdoliva	verdoliva	PROPN
ap-8720	418	13	,	,	PUNCT
ap-8720	418	14	et	et	PROPN
ap-8720	418	15	al	al	PROPN
ap-8720	418	16	.	.	PUNCT
ap-8720	419	1	faceforensics++	faceforensics++	ADV
ap-8720	419	2	:	:	PUNCT
ap-8720	419	3	learning	learn	VERB
ap-8720	419	4	to	to	PART
ap-8720	419	5	detect	detect	VERB
ap-8720	419	6	manipulated	manipulate	VERB
ap-8720	419	7	facial	facial	ADJ
ap-8720	419	8	images	image	NOUN
ap-8720	419	9	.	.	PUNCT
ap-8720	420	1	[	[	X
ap-8720	420	2	2023	2023	NUM
ap-8720	420	3	-	-	SYM
ap-8720	420	4	05	05	NUM
ap-8720	420	5	-	-	SYM
ap-8720	420	6	31	31	NUM
ap-8720	420	7	]	]	PUNCT
ap-8720	420	8	.	.	PUNCT
ap-8720	421	1	arxiv:1901.08971	arxiv:1901.08971	NOUN
ap-8720	421	2	[	[	X
ap-8720	421	3	9	9	NUM
ap-8720	421	4	]	]	X
ap-8720	421	5	y.	y.	PROPN
ap-8720	421	6	li	li	PROPN
ap-8720	421	7	,	,	PUNCT
ap-8720	421	8	x.	x.	PROPN
ap-8720	421	9	yang	yang	PROPN
ap-8720	421	10	,	,	PUNCT
ap-8720	421	11	p.	p.	PROPN
ap-8720	421	12	sun	sun	PROPN
ap-8720	421	13	,	,	PUNCT
ap-8720	421	14	et	et	PROPN
ap-8720	421	15	al	al	PROPN
ap-8720	421	16	.	.	PUNCT
ap-8720	422	1	celeb	celeb	PROPN
ap-8720	422	2	-	-	PUNCT
ap-8720	422	3	df	df	PROPN
ap-8720	422	4	:	:	PUNCT
ap-8720	422	5	a	a	DET
ap-8720	422	6	large	large	ADJ
ap-8720	422	7	-	-	PUNCT
ap-8720	422	8	scale	scale	NOUN
ap-8720	422	9	challenging	challenging	ADJ
ap-8720	422	10	dataset	dataset	NOUN
ap-8720	422	11	for	for	ADP
ap-8720	422	12	deepfake	deepfake	NOUN
ap-8720	422	13	forensics	forensic	NOUN
ap-8720	422	14	.	.	PUNCT
ap-8720	423	1	[	[	X
ap-8720	423	2	2023	2023	NUM
ap-8720	423	3	-	-	SYM
ap-8720	423	4	05	05	NUM
ap-8720	423	5	-	-	SYM
ap-8720	423	6	31	31	NUM
ap-8720	423	7	]	]	PUNCT
ap-8720	423	8	.	.	PUNCT
ap-8720	424	1	arxiv:1909.12962v4	arxiv:1909.12962v4	ADV
ap-8720	424	2	[	[	X
ap-8720	424	3	10	10	NUM
ap-8720	424	4	]	]	X
ap-8720	424	5	g.	g.	PROPN
ap-8720	424	6	mahfoudi	mahfoudi	PROPN
ap-8720	424	7	,	,	PUNCT
ap-8720	424	8	b.	b.	PROPN
ap-8720	424	9	tajini	tajini	PROPN
ap-8720	424	10	,	,	PUNCT
ap-8720	424	11	f.	f.	PROPN
ap-8720	424	12	retraint	retraint	PROPN
ap-8720	424	13	,	,	PUNCT
ap-8720	424	14	et	et	PROPN
ap-8720	424	15	al	al	PROPN
ap-8720	424	16	.	.	PUNCT
ap-8720	425	1	defacto	defacto	PROPN
ap-8720	425	2	:	:	PUNCT
ap-8720	425	3	image	image	NOUN
ap-8720	425	4	and	and	CCONJ
ap-8720	425	5	face	face	VERB
ap-8720	425	6	manipulation	manipulation	NOUN
ap-8720	425	7	dataset	dataset	NOUN
ap-8720	425	8	.	.	PUNCT
ap-8720	426	1	in	in	ADP
ap-8720	426	2	27th	27th	ADJ
ap-8720	426	3	european	european	PROPN
ap-8720	426	4	signal	signal	PROPN
ap-8720	426	5	processing	processing	NOUN
ap-8720	426	6	conference	conference	NOUN
ap-8720	426	7	(	(	PUNCT
ap-8720	426	8	eusipco	eusipco	PROPN
ap-8720	426	9	)	)	PUNCT
ap-8720	426	10	,	,	PUNCT
ap-8720	426	11	pp	pp	ADP
ap-8720	426	12	.	.	PUNCT
ap-8720	427	1	1–5	1–5	X
ap-8720	427	2	.	.	NOUN
ap-8720	427	3	2019	2019	NUM
ap-8720	427	4	.	.	PUNCT
ap-8720	428	1	https://doi.org/10.23919/eusipco.2019.8903181	https://doi.org/10.23919/eusipco.2019.8903181	PROPN
ap-8720	429	1	[	[	X
ap-8720	429	2	11	11	NUM
ap-8720	429	3	]	]	PUNCT
ap-8720	429	4	p.	p.	NOUN
ap-8720	429	5	zhou	zhou	PROPN
ap-8720	429	6	,	,	PUNCT
ap-8720	429	7	x.	x.	PROPN
ap-8720	429	8	han	han	PROPN
ap-8720	429	9	,	,	PUNCT
ap-8720	429	10	v.	v.	PROPN
ap-8720	429	11	i.	i.	PROPN
ap-8720	429	12	morariu	morariu	PROPN
ap-8720	429	13	,	,	PUNCT
ap-8720	429	14	l.	l.	PROPN
ap-8720	429	15	s.	s.	PROPN
ap-8720	429	16	davis	davis	PROPN
ap-8720	429	17	.	.	PUNCT
ap-8720	430	1	two	two	NUM
ap-8720	430	2	-	-	PUNCT
ap-8720	430	3	stream	stream	NOUN
ap-8720	430	4	neural	neural	ADJ
ap-8720	430	5	networks	network	NOUN
ap-8720	430	6	for	for	ADP
ap-8720	430	7	tampered	tamper	VERB
ap-8720	430	8	face	face	NOUN
ap-8720	430	9	detection	detection	NOUN
ap-8720	430	10	.	.	PUNCT
ap-8720	431	1	in	in	ADP
ap-8720	431	2	ieee	ieee	PROPN
ap-8720	431	3	conference	conference	NOUN
ap-8720	431	4	on	on	ADP
ap-8720	431	5	computer	computer	NOUN
ap-8720	431	6	vision	vision	NOUN
ap-8720	431	7	and	and	CCONJ
ap-8720	431	8	pattern	pattern	NOUN
ap-8720	431	9	recognition	recognition	NOUN
ap-8720	431	10	workshops	workshop	NOUN
ap-8720	431	11	(	(	PUNCT
ap-8720	431	12	cvprw	cvprw	ADJ
ap-8720	431	13	)	)	PUNCT
ap-8720	431	14	,	,	PUNCT
ap-8720	431	15	pp	pp	PROPN
ap-8720	431	16	.	.	PUNCT
ap-8720	431	17	1831–1839	1831–1839	NUM
ap-8720	431	18	.	.	NOUN
ap-8720	431	19	2017	2017	NUM
ap-8720	431	20	.	.	PUNCT
ap-8720	432	1	https://doi.org/10.1109/cvprw.2017.229	https://doi.org/10.1109/cvprw.2017.229	PROPN
ap-8720	433	1	[	[	X
ap-8720	433	2	12	12	NUM
ap-8720	433	3	]	]	PUNCT
ap-8720	433	4	z.	z.	PROPN
ap-8720	433	5	guo	guo	PROPN
ap-8720	433	6	,	,	PUNCT
ap-8720	433	7	g.	g.	PROPN
ap-8720	433	8	yang	yang	PROPN
ap-8720	433	9	,	,	PUNCT
ap-8720	433	10	j.	j.	PROPN
ap-8720	433	11	chen	chen	PROPN
ap-8720	433	12	,	,	PUNCT
ap-8720	433	13	x.	x.	PROPN
ap-8720	433	14	sun	sun	PROPN
ap-8720	433	15	.	.	PUNCT
ap-8720	434	1	fake	fake	ADJ
ap-8720	434	2	face	face	NOUN
ap-8720	434	3	detection	detection	NOUN
ap-8720	434	4	via	via	ADP
ap-8720	434	5	adaptive	adaptive	ADJ
ap-8720	434	6	manipulation	manipulation	NOUN
ap-8720	434	7	traces	trace	VERB
ap-8720	434	8	extraction	extraction	NOUN
ap-8720	434	9	network	network	NOUN
ap-8720	434	10	.	.	PUNCT
ap-8720	435	1	computer	computer	NOUN
ap-8720	435	2	vision	vision	NOUN
ap-8720	435	3	and	and	CCONJ
ap-8720	435	4	image	image	NOUN
ap-8720	435	5	understanding	understand	VERB
ap-8720	435	6	204:103170	204:103170	NUM
ap-8720	435	7	,	,	PUNCT
ap-8720	435	8	2021	2021	NUM
ap-8720	435	9	.	.	PUNCT
ap-8720	436	1	https://doi.org/10.1016/j.cviu.2021.103170	https://doi.org/10.1016/j.cviu.2021.103170	X
ap-8720	437	1	[	[	X
ap-8720	437	2	13	13	NUM
ap-8720	437	3	]	]	PUNCT
ap-8720	437	4	z.	z.	PROPN
ap-8720	437	5	alom	alom	PROPN
ap-8720	437	6	,	,	PUNCT
ap-8720	437	7	t.	t.	PROPN
ap-8720	437	8	m.	m.	PROPN
ap-8720	437	9	taha	taha	PROPN
ap-8720	437	10	,	,	PUNCT
ap-8720	437	11	c.	c.	PROPN
ap-8720	437	12	yakopcic	yakopcic	PROPN
ap-8720	437	13	,	,	PUNCT
ap-8720	437	14	et	et	PROPN
ap-8720	437	15	al	al	PROPN
ap-8720	437	16	.	.	PUNCT
ap-8720	438	1	a	a	DET
ap-8720	438	2	state	state	NOUN
ap-8720	438	3	-	-	PUNCT
ap-8720	438	4	of	of	ADP
ap-8720	438	5	-	-	PUNCT
ap-8720	438	6	the	the	DET
ap-8720	438	7	-	-	PUNCT
ap-8720	438	8	art	art	NOUN
ap-8720	438	9	survey	survey	NOUN
ap-8720	438	10	on	on	ADP
ap-8720	438	11	deep	deep	ADJ
ap-8720	438	12	learning	learning	NOUN
ap-8720	438	13	theory	theory	NOUN
ap-8720	438	14	and	and	CCONJ
ap-8720	438	15	architectures	architecture	NOUN
ap-8720	438	16	.	.	PUNCT
ap-8720	439	1	electronics	electronic	NOUN
ap-8720	439	2	8(3):292	8(3):292	NUM
ap-8720	439	3	,	,	PUNCT
ap-8720	439	4	2019	2019	NUM
ap-8720	439	5	.	.	PUNCT
ap-8720	440	1	https://doi.org/10.3390/electronics8030292	https://doi.org/10.3390/electronics8030292	NOUN
ap-8720	441	1	[	[	X
ap-8720	441	2	14	14	NUM
ap-8720	441	3	]	]	PUNCT
ap-8720	441	4	k.	k.	PROPN
ap-8720	442	1	g.	g.	PROPN
ap-8720	442	2	kim	kim	PROPN
ap-8720	442	3	.	.	PUNCT
ap-8720	443	1	book	book	PROPN
ap-8720	443	2	review	review	PROPN
ap-8720	443	3	:	:	PUNCT
ap-8720	443	4	deep	deep	ADJ
ap-8720	443	5	learning	learning	NOUN
ap-8720	443	6	.	.	PUNCT
ap-8720	444	1	healthcare	healthcare	NOUN
ap-8720	444	2	informatics	informatics	PROPN
ap-8720	444	3	research	research	PROPN
ap-8720	444	4	22(4):351–354	22(4):351–354	PROPN
ap-8720	444	5	,	,	PUNCT
ap-8720	444	6	2016	2016	NUM
ap-8720	444	7	.	.	PUNCT
ap-8720	445	1	https://doi.org/10.4258/hir.2016.22.4.351	https://doi.org/10.4258/hir.2016.22.4.351	NOUN
ap-8720	446	1	[	[	X
ap-8720	446	2	15	15	NUM
ap-8720	446	3	]	]	X
ap-8720	446	4	d.	d.	PROPN
ap-8720	446	5	güera	güera	PROPN
ap-8720	446	6	,	,	PUNCT
ap-8720	446	7	e.	e.	PROPN
ap-8720	446	8	j.	j.	PROPN
ap-8720	446	9	delp	delp	PROPN
ap-8720	446	10	.	.	PUNCT
ap-8720	447	1	deepfake	deepfake	NOUN
ap-8720	447	2	video	video	NOUN
ap-8720	447	3	detection	detection	NOUN
ap-8720	447	4	using	use	VERB
ap-8720	447	5	recurrent	recurrent	ADJ
ap-8720	447	6	neural	neural	ADJ
ap-8720	447	7	networks	network	NOUN
ap-8720	447	8	.	.	PUNCT
ap-8720	448	1	in	in	ADP
ap-8720	448	2	15th	15th	ADJ
ap-8720	448	3	ieee	ieee	PROPN
ap-8720	448	4	international	international	ADJ
ap-8720	448	5	conference	conference	NOUN
ap-8720	448	6	on	on	ADP
ap-8720	448	7	advanced	advanced	ADJ
ap-8720	448	8	video	video	NOUN
ap-8720	448	9	and	and	CCONJ
ap-8720	448	10	signal	signal	NOUN
ap-8720	448	11	based	base	VERB
ap-8720	448	12	surveillance	surveillance	NOUN
ap-8720	448	13	(	(	PUNCT
ap-8720	448	14	avss	avss	NOUN
ap-8720	448	15	)	)	PUNCT
ap-8720	448	16	,	,	PUNCT
ap-8720	448	17	pp	pp	ADP
ap-8720	448	18	.	.	PUNCT
ap-8720	449	1	1–6	1–6	NUM
ap-8720	449	2	.	.	NOUN
ap-8720	449	3	2018	2018	NUM
ap-8720	449	4	.	.	PUNCT
ap-8720	450	1	https://doi.org/10.1109/avss.2018.8639163	https://doi.org/10.1109/avss.2018.8639163	PROPN
ap-8720	451	1	[	[	X
ap-8720	451	2	16	16	NUM
ap-8720	451	3	]	]	X
ap-8720	451	4	i.	i.	PROPN
ap-8720	451	5	laptev	laptev	PROPN
ap-8720	451	6	,	,	PUNCT
ap-8720	451	7	m.	m.	NOUN
ap-8720	451	8	marszalek	marszalek	PROPN
ap-8720	451	9	,	,	PUNCT
ap-8720	451	10	c.	c.	PROPN
ap-8720	451	11	schmid	schmid	PROPN
ap-8720	451	12	,	,	PUNCT
ap-8720	451	13	b.	b.	PROPN
ap-8720	451	14	rozenfeld	rozenfeld	PROPN
ap-8720	451	15	.	.	PUNCT
ap-8720	452	1	learning	learn	VERB
ap-8720	452	2	realistic	realistic	ADJ
ap-8720	452	3	human	human	ADJ
ap-8720	452	4	actions	action	NOUN
ap-8720	452	5	from	from	ADP
ap-8720	452	6	movies	movie	NOUN
ap-8720	452	7	.	.	PUNCT
ap-8720	453	1	in	in	ADP
ap-8720	453	2	ieee	ieee	PROPN
ap-8720	453	3	conference	conference	NOUN
ap-8720	453	4	on	on	ADP
ap-8720	453	5	computer	computer	NOUN
ap-8720	453	6	vision	vision	NOUN
ap-8720	453	7	and	and	CCONJ
ap-8720	453	8	pattern	pattern	NOUN
ap-8720	453	9	recognition	recognition	NOUN
ap-8720	453	10	,	,	PUNCT
ap-8720	453	11	pp	pp	PROPN
ap-8720	453	12	.	.	PUNCT
ap-8720	454	1	1–8	1–8	X
ap-8720	454	2	.	.	NOUN
ap-8720	454	3	2008	2008	NUM
ap-8720	454	4	.	.	PUNCT
ap-8720	455	1	https://doi.org/10.1109/cvpr.2008.4587756	https://doi.org/10.1109/cvpr.2008.4587756	PROPN
ap-8720	456	1	[	[	X
ap-8720	456	2	17	17	NUM
ap-8720	456	3	]	]	PUNCT
ap-8720	456	4	z.	z.	PROPN
ap-8720	456	5	liu	liu	PROPN
ap-8720	456	6	,	,	PUNCT
ap-8720	456	7	x.	x.	PROPN
ap-8720	456	8	qi	qi	PROPN
ap-8720	456	9	,	,	PUNCT
ap-8720	456	10	p.	p.	PROPN
ap-8720	456	11	h.	h.	PROPN
ap-8720	456	12	s.	s.	PROPN
ap-8720	456	13	torr	torr	PROPN
ap-8720	456	14	.	.	PUNCT
ap-8720	457	1	global	global	ADJ
ap-8720	457	2	texture	texture	NOUN
ap-8720	457	3	enhancement	enhancement	NOUN
ap-8720	457	4	for	for	ADP
ap-8720	457	5	fake	fake	ADJ
ap-8720	457	6	face	face	NOUN
ap-8720	457	7	detection	detection	NOUN
ap-8720	457	8	in	in	ADP
ap-8720	457	9	the	the	DET
ap-8720	457	10	wild	wild	NOUN
ap-8720	457	11	.	.	PUNCT
ap-8720	458	1	in	in	ADP
ap-8720	458	2	ieee	ieee	NOUN
ap-8720	458	3	/	/	SYM
ap-8720	458	4	cvf	cvf	NOUN
ap-8720	458	5	conference	conference	NOUN
ap-8720	458	6	on	on	ADP
ap-8720	458	7	computer	computer	NOUN
ap-8720	458	8	vision	vision	NOUN
ap-8720	458	9	and	and	CCONJ
ap-8720	458	10	pattern	pattern	NOUN
ap-8720	458	11	recognition	recognition	NOUN
ap-8720	458	12	(	(	PUNCT
ap-8720	458	13	cvpr	cvpr	NOUN
ap-8720	458	14	)	)	PUNCT
ap-8720	458	15	,	,	PUNCT
ap-8720	458	16	pp	pp	PROPN
ap-8720	458	17	.	.	PUNCT
ap-8720	458	18	8057–8066	8057–8066	NUM
ap-8720	458	19	.	.	PUNCT
ap-8720	458	20	2020	2020	NUM
ap-8720	458	21	.	.	PUNCT
ap-8720	459	1	https://doi.org/10.1109/cvpr42600.2020.00808	https://doi.org/10.1109/cvpr42600.2020.00808	PROPN
ap-8720	460	1	[	[	X
ap-8720	460	2	18	18	NUM
ap-8720	460	3	]	]	PUNCT
ap-8720	460	4	t.	t.	PROPN
ap-8720	460	5	karras	karras	PROPN
ap-8720	460	6	,	,	PUNCT
ap-8720	460	7	s.	s.	PROPN
ap-8720	460	8	laine	laine	PROPN
ap-8720	460	9	,	,	PUNCT
ap-8720	460	10	t.	t.	PROPN
ap-8720	460	11	aila	aila	PROPN
ap-8720	460	12	.	.	PUNCT
ap-8720	461	1	a	a	DET
ap-8720	461	2	style	style	NOUN
ap-8720	461	3	-	-	PUNCT
ap-8720	461	4	based	base	VERB
ap-8720	461	5	generator	generator	NOUN
ap-8720	461	6	architecture	architecture	NOUN
ap-8720	461	7	for	for	ADP
ap-8720	461	8	generative	generative	ADJ
ap-8720	461	9	adversarial	adversarial	ADJ
ap-8720	461	10	networks	network	NOUN
ap-8720	461	11	.	.	PUNCT
ap-8720	462	1	[	[	X
ap-8720	462	2	2023	2023	NUM
ap-8720	462	3	-	-	SYM
ap-8720	462	4	05	05	NUM
ap-8720	462	5	-	-	SYM
ap-8720	462	6	31	31	NUM
ap-8720	462	7	]	]	PUNCT
ap-8720	462	8	.	.	PUNCT
ap-8720	463	1	arxiv:1812.04948	arxiv:1812.04948	PROPN
ap-8720	464	1	[	[	X
ap-8720	464	2	19	19	NUM
ap-8720	464	3	]	]	PUNCT
ap-8720	464	4	t.	t.	PROPN
ap-8720	464	5	karras	karras	PROPN
ap-8720	464	6	,	,	PUNCT
ap-8720	464	7	t.	t.	PROPN
ap-8720	464	8	aila	aila	PROPN
ap-8720	464	9	,	,	PUNCT
ap-8720	464	10	s.	s.	PROPN
ap-8720	464	11	laine	laine	PROPN
ap-8720	464	12	,	,	PUNCT
ap-8720	464	13	j.	j.	PROPN
ap-8720	464	14	lehtinen	lehtinen	PROPN
ap-8720	464	15	.	.	PUNCT
ap-8720	465	1	progressive	progressive	ADJ
ap-8720	465	2	growing	grow	VERB
ap-8720	465	3	of	of	ADP
ap-8720	465	4	gans	gan	NOUN
ap-8720	465	5	for	for	ADP
ap-8720	465	6	improved	improved	ADJ
ap-8720	465	7	quality	quality	NOUN
ap-8720	465	8	,	,	PUNCT
ap-8720	465	9	stability	stability	NOUN
ap-8720	465	10	,	,	PUNCT
ap-8720	465	11	and	and	CCONJ
ap-8720	465	12	variation	variation	NOUN
ap-8720	465	13	.	.	PUNCT
ap-8720	466	1	[	[	X
ap-8720	466	2	2023	2023	NUM
ap-8720	466	3	-	-	SYM
ap-8720	466	4	05	05	NUM
ap-8720	466	5	-	-	SYM
ap-8720	466	6	31	31	NUM
ap-8720	466	7	]	]	PUNCT
ap-8720	466	8	.	.	PUNCT
ap-8720	467	1	arxiv:1710.10196	arxiv:1710.10196	NUM
ap-8720	467	2	[	[	X
ap-8720	467	3	20	20	NUM
ap-8720	467	4	]	]	PUNCT
ap-8720	467	5	a.	a.	NOUN
ap-8720	467	6	deshmukh	deshmukh	PROPN
ap-8720	467	7	,	,	PUNCT
ap-8720	467	8	s.	s.	PROPN
ap-8720	467	9	b.	b.	PROPN
ap-8720	467	10	wankhade	wankhade	PROPN
ap-8720	467	11	.	.	PUNCT
ap-8720	468	1	deepfake	deepfake	NOUN
ap-8720	468	2	detection	detection	NOUN
ap-8720	468	3	approaches	approach	NOUN
ap-8720	468	4	using	use	VERB
ap-8720	468	5	deep	deep	ADJ
ap-8720	468	6	learning	learning	NOUN
ap-8720	468	7	:	:	PUNCT
ap-8720	468	8	a	a	DET
ap-8720	468	9	systematic	systematic	ADJ
ap-8720	468	10	review	review	NOUN
ap-8720	468	11	.	.	PUNCT
ap-8720	469	1	in	in	ADP
ap-8720	469	2	intelligent	intelligent	ADJ
ap-8720	469	3	computing	computing	NOUN
ap-8720	469	4	and	and	CCONJ
ap-8720	469	5	networking	networking	NOUN
ap-8720	469	6	.	.	PUNCT
ap-8720	470	1	lecture	lecture	NOUN
ap-8720	470	2	notes	note	NOUN
ap-8720	470	3	in	in	ADP
ap-8720	470	4	networks	network	NOUN
ap-8720	470	5	and	and	CCONJ
ap-8720	470	6	systems	system	NOUN
ap-8720	470	7	,	,	PUNCT
ap-8720	470	8	vol	vol	NOUN
ap-8720	470	9	.	.	PROPN
ap-8720	471	1	146	146	NUM
ap-8720	471	2	.	.	NOUN
ap-8720	472	1	2021	2021	NUM
ap-8720	472	2	.	.	PUNCT
ap-8720	473	1	https://doi.org/10.1007/978-981-15-7421-4_27	https://doi.org/10.1007/978-981-15-7421-4_27	PUNCT
ap-8720	474	1	[	[	X
ap-8720	474	2	21	21	NUM
ap-8720	474	3	]	]	X
ap-8720	474	4	s.	s.	PROPN
ap-8720	474	5	albawi	albawi	PROPN
ap-8720	474	6	,	,	PUNCT
ap-8720	474	7	t.	t.	PROPN
ap-8720	474	8	a.	a.	PROPN
ap-8720	474	9	mohammed	mohammed	PROPN
ap-8720	474	10	,	,	PUNCT
ap-8720	474	11	s.	s.	PROPN
ap-8720	474	12	al	al	PROPN
ap-8720	474	13	-	-	PROPN
ap-8720	474	14	zawi	zawi	PROPN
ap-8720	474	15	.	.	PUNCT
ap-8720	475	1	understanding	understanding	NOUN
ap-8720	475	2	of	of	ADP
ap-8720	475	3	a	a	DET
ap-8720	475	4	convolutional	convolutional	ADJ
ap-8720	475	5	neural	neural	ADJ
ap-8720	475	6	network	network	NOUN
ap-8720	475	7	.	.	PUNCT
ap-8720	476	1	in	in	ADP
ap-8720	476	2	international	international	ADJ
ap-8720	476	3	conference	conference	NOUN
ap-8720	476	4	on	on	ADP
ap-8720	476	5	engineering	engineering	NOUN
ap-8720	476	6	and	and	CCONJ
ap-8720	476	7	technology	technology	NOUN
ap-8720	476	8	(	(	PUNCT
ap-8720	476	9	icet	icet	PROPN
ap-8720	476	10	)	)	PUNCT
ap-8720	476	11	,	,	PUNCT
ap-8720	476	12	pp	pp	ADP
ap-8720	476	13	.	.	PUNCT
ap-8720	477	1	1–6	1–6	NUM
ap-8720	477	2	.	.	NOUN
ap-8720	477	3	2017	2017	NUM
ap-8720	477	4	.	.	PUNCT
ap-8720	478	1	https	https	NOUN
ap-8720	478	2	:	:	PUNCT
ap-8720	478	3	//doi.org/10.1109	//doi.org/10.1109	VERB
ap-8720	478	4	/	/	SYM
ap-8720	478	5	icengtechnol.2017.8308186	icengtechnol.2017.8308186	ADJ
ap-8720	479	1	[	[	X
ap-8720	479	2	22	22	NUM
ap-8720	479	3	]	]	PUNCT
ap-8720	479	4	t.	t.	PROPN
ap-8720	479	5	jung	jung	PROPN
ap-8720	479	6	,	,	PUNCT
ap-8720	479	7	s.	s.	PROPN
ap-8720	479	8	kim	kim	PROPN
ap-8720	479	9	,	,	PUNCT
ap-8720	479	10	k.	k.	PROPN
ap-8720	479	11	kim	kim	PROPN
ap-8720	479	12	.	.	PUNCT
ap-8720	480	1	deepvision	deepvision	PROPN
ap-8720	480	2	:	:	PUNCT
ap-8720	480	3	deepfakes	deepfake	VERB
ap-8720	480	4	detection	detection	NOUN
ap-8720	480	5	using	use	VERB
ap-8720	480	6	human	human	ADJ
ap-8720	480	7	eye	eye	NOUN
ap-8720	480	8	blinking	blink	VERB
ap-8720	480	9	pattern	pattern	NOUN
ap-8720	480	10	.	.	PUNCT
ap-8720	481	1	ieee	ieee	NOUN
ap-8720	481	2	access	access	NOUN
ap-8720	481	3	8:83144–83154	8:83144–83154	NUM
ap-8720	481	4	,	,	PUNCT
ap-8720	481	5	2020	2020	NUM
ap-8720	481	6	.	.	PUNCT
ap-8720	482	1	https://doi.org/10.1109/access.2020.2988660	https://doi.org/10.1109/access.2020.2988660	PROPN
ap-8720	483	1	[	[	X
ap-8720	483	2	23	23	NUM
ap-8720	483	3	]	]	PUNCT
ap-8720	483	4	p.	p.	PROPN
ap-8720	483	5	korshunov	korshunov	PROPN
ap-8720	483	6	,	,	PUNCT
ap-8720	483	7	s.	s.	PROPN
ap-8720	483	8	marcel	marcel	PROPN
ap-8720	483	9	.	.	PUNCT
ap-8720	483	10	vulnerability	vulnerability	NOUN
ap-8720	483	11	assessment	assessment	NOUN
ap-8720	483	12	and	and	CCONJ
ap-8720	483	13	detection	detection	NOUN
ap-8720	483	14	of	of	ADP
ap-8720	483	15	deepfake	deepfake	NOUN
ap-8720	483	16	videos	video	NOUN
ap-8720	483	17	.	.	PUNCT
ap-8720	484	1	in	in	ADP
ap-8720	484	2	international	international	ADJ
ap-8720	484	3	conference	conference	NOUN
ap-8720	484	4	on	on	ADP
ap-8720	484	5	biometrics	biometric	NOUN
ap-8720	484	6	(	(	PUNCT
ap-8720	484	7	icb	icb	PROPN
ap-8720	484	8	)	)	PUNCT
ap-8720	484	9	,	,	PUNCT
ap-8720	484	10	pp	pp	ADP
ap-8720	484	11	.	.	PUNCT
ap-8720	485	1	1–6	1–6	NUM
ap-8720	485	2	.	.	NOUN
ap-8720	486	1	2019	2019	NUM
ap-8720	486	2	.	.	PUNCT
ap-8720	486	3	https://doi.org/10.1109/icb45273.2019.8987375	https://doi.org/10.1109/icb45273.2019.8987375	X
ap-8720	487	1	[	[	X
ap-8720	487	2	24	24	NUM
ap-8720	487	3	]	]	X
ap-8720	487	4	o.	o.	PROPN
ap-8720	487	5	parkhi	parkhi	PROPN
ap-8720	487	6	,	,	PUNCT
ap-8720	487	7	a.	a.	NOUN
ap-8720	487	8	vedaldi	vedaldi	PROPN
ap-8720	487	9	,	,	PUNCT
ap-8720	487	10	a.	a.	NOUN
ap-8720	487	11	zisserman	zisserman	PROPN
ap-8720	487	12	.	.	PUNCT
ap-8720	488	1	deep	deep	ADJ
ap-8720	488	2	face	face	NOUN
ap-8720	488	3	recognition	recognition	NOUN
ap-8720	488	4	.	.	PUNCT
ap-8720	489	1	in	in	ADP
ap-8720	489	2	british	british	ADJ
ap-8720	489	3	machine	machine	PROPN
ap-8720	489	4	vision	vision	PROPN
ap-8720	489	5	conference	conference	PROPN
ap-8720	489	6	,	,	PUNCT
ap-8720	489	7	pp	pp	PROPN
ap-8720	489	8	.	.	PUNCT
ap-8720	489	9	1–12	1–12	NOUN
ap-8720	489	10	.	.	PUNCT
ap-8720	490	1	2015	2015	NUM
ap-8720	490	2	.	.	PUNCT
ap-8720	490	3	https://doi.org/10.1109/icb45273.2019.8987375	https://doi.org/10.1109/icb45273.2019.8987375	X
ap-8720	491	1	[	[	X
ap-8720	491	2	25	25	NUM
ap-8720	491	3	]	]	PUNCT
ap-8720	491	4	f.	f.	PROPN
ap-8720	491	5	schroff	schroff	PROPN
ap-8720	491	6	,	,	PUNCT
ap-8720	491	7	d.	d.	PROPN
ap-8720	491	8	kalenichenko	kalenichenko	PROPN
ap-8720	491	9	,	,	PUNCT
ap-8720	491	10	j.	j.	PROPN
ap-8720	491	11	philbin	philbin	PROPN
ap-8720	491	12	.	.	PUNCT
ap-8720	492	1	facenet	facenet	PROPN
ap-8720	492	2	:	:	PUNCT
ap-8720	492	3	a	a	DET
ap-8720	492	4	unified	unified	ADJ
ap-8720	492	5	embedding	embedding	NOUN
ap-8720	492	6	for	for	ADP
ap-8720	492	7	face	face	NOUN
ap-8720	492	8	recognition	recognition	NOUN
ap-8720	492	9	and	and	CCONJ
ap-8720	492	10	clustering	clustering	NOUN
ap-8720	492	11	.	.	PUNCT
ap-8720	493	1	in	in	ADP
ap-8720	493	2	ieee	ieee	PROPN
ap-8720	493	3	conference	conference	NOUN
ap-8720	493	4	on	on	ADP
ap-8720	493	5	computer	computer	NOUN
ap-8720	493	6	vision	vision	NOUN
ap-8720	493	7	and	and	CCONJ
ap-8720	493	8	pattern	pattern	NOUN
ap-8720	493	9	recognition	recognition	NOUN
ap-8720	493	10	(	(	PUNCT
ap-8720	493	11	cvpr	cvpr	NOUN
ap-8720	493	12	)	)	PUNCT
ap-8720	493	13	,	,	PUNCT
ap-8720	493	14	pp	pp	ADP
ap-8720	493	15	.	.	PUNCT
ap-8720	494	1	815–823	815–823	NUM
ap-8720	494	2	.	.	NOUN
ap-8720	494	3	2015	2015	NUM
ap-8720	494	4	.	.	PUNCT
ap-8720	495	1	https://doi.org/10.1109/cvpr.2015.7298682	https://doi.org/10.1109/cvpr.2015.7298682	X
ap-8720	496	1	[	[	X
ap-8720	496	2	26	26	NUM
ap-8720	496	3	]	]	X
ap-8720	496	4	y.	y.	PROPN
ap-8720	496	5	li	li	PROPN
ap-8720	496	6	,	,	PUNCT
ap-8720	496	7	s.	s.	PROPN
ap-8720	496	8	lyu	lyu	PROPN
ap-8720	496	9	.	.	PUNCT
ap-8720	497	1	exposing	expose	VERB
ap-8720	497	2	deepfake	deepfake	NOUN
ap-8720	497	3	videos	video	NOUN
ap-8720	497	4	by	by	ADP
ap-8720	497	5	detecting	detect	VERB
ap-8720	497	6	face	face	NOUN
ap-8720	497	7	warping	warp	VERB
ap-8720	497	8	artifacts	artifact	NOUN
ap-8720	497	9	.	.	PUNCT
ap-8720	498	1	[	[	X
ap-8720	498	2	2023	2023	NUM
ap-8720	498	3	-	-	SYM
ap-8720	498	4	05	05	NUM
ap-8720	498	5	-	-	SYM
ap-8720	498	6	31	31	NUM
ap-8720	498	7	]	]	PUNCT
ap-8720	498	8	.	.	PUNCT
ap-8720	499	1	arxiv:1811.00656	arxiv:1811.00656	PROPN
ap-8720	499	2	[	[	X
ap-8720	499	3	27	27	NUM
ap-8720	499	4	]	]	PUNCT
ap-8720	499	5	x.	x.	PROPN
ap-8720	499	6	yang	yang	PROPN
ap-8720	499	7	,	,	PUNCT
ap-8720	499	8	y.	y.	PROPN
ap-8720	499	9	li	li	PROPN
ap-8720	499	10	,	,	PUNCT
ap-8720	499	11	s.	s.	PROPN
ap-8720	499	12	lyu	lyu	PROPN
ap-8720	499	13	.	.	PUNCT
ap-8720	500	1	exposing	expose	VERB
ap-8720	500	2	deep	deep	ADJ
ap-8720	500	3	fakes	fake	NOUN
ap-8720	500	4	using	use	VERB
ap-8720	500	5	inconsistent	inconsistent	ADJ
ap-8720	500	6	head	head	NOUN
ap-8720	500	7	poses	pose	NOUN
ap-8720	500	8	.	.	PUNCT
ap-8720	501	1	in	in	ADP
ap-8720	501	2	ieee	ieee	PROPN
ap-8720	501	3	international	international	ADJ
ap-8720	501	4	conference	conference	NOUN
ap-8720	501	5	on	on	ADP
ap-8720	501	6	acoustics	acoustic	NOUN
ap-8720	501	7	,	,	PUNCT
ap-8720	501	8	speech	speech	NOUN
ap-8720	501	9	and	and	CCONJ
ap-8720	501	10	signal	signal	NOUN
ap-8720	501	11	processing	processing	NOUN
ap-8720	501	12	(	(	PUNCT
ap-8720	501	13	icassp	icassp	PROPN
ap-8720	501	14	)	)	PUNCT
ap-8720	501	15	,	,	PUNCT
ap-8720	501	16	pp	pp	PROPN
ap-8720	501	17	.	.	PUNCT
ap-8720	502	1	8261–8265	8261–8265	NUM
ap-8720	502	2	.	.	NOUN
ap-8720	502	3	2019	2019	NUM
ap-8720	502	4	.	.	PUNCT
ap-8720	503	1	https://doi.org/10.1109/icassp.2019.8683164	https://doi.org/10.1109/icassp.2019.8683164	PROPN
ap-8720	504	1	[	[	X
ap-8720	504	2	28	28	NUM
ap-8720	504	3	]	]	X
ap-8720	504	4	a.	a.	NOUN
ap-8720	504	5	chintha	chintha	PROPN
ap-8720	504	6	,	,	PUNCT
ap-8720	504	7	b.	b.	PROPN
ap-8720	504	8	thai	thai	PROPN
ap-8720	504	9	,	,	PUNCT
ap-8720	504	10	s.	s.	PROPN
ap-8720	504	11	j.	j.	PROPN
ap-8720	504	12	sohrawardi	sohrawardi	PROPN
ap-8720	504	13	,	,	PUNCT
ap-8720	504	14	et	et	PROPN
ap-8720	504	15	al	al	PROPN
ap-8720	504	16	.	.	PROPN
ap-8720	505	1	recurrent	recurrent	ADJ
ap-8720	505	2	convolutional	convolutional	ADJ
ap-8720	505	3	structures	structure	NOUN
ap-8720	505	4	for	for	ADP
ap-8720	505	5	audio	audio	ADJ
ap-8720	505	6	spoof	spoof	NOUN
ap-8720	505	7	and	and	CCONJ
ap-8720	505	8	video	video	NOUN
ap-8720	505	9	deepfake	deepfake	NOUN
ap-8720	505	10	detection	detection	NOUN
ap-8720	505	11	.	.	PUNCT
ap-8720	506	1	ieee	ieee	PROPN
ap-8720	506	2	journal	journal	PROPN
ap-8720	506	3	of	of	ADP
ap-8720	506	4	selected	select	VERB
ap-8720	506	5	topics	topic	NOUN
ap-8720	506	6	in	in	ADP
ap-8720	506	7	signal	signal	ADJ
ap-8720	506	8	processing	processing	NOUN
ap-8720	506	9	14(5):1024–1037	14(5):1024–1037	NUM
ap-8720	506	10	,	,	PUNCT
ap-8720	506	11	2020	2020	NUM
ap-8720	506	12	.	.	PUNCT
ap-8720	507	1	https://doi.org/10.1109/jstsp.2020.2999185	https://doi.org/10.1109/jstsp.2020.2999185	PROPN
ap-8720	507	2	[	[	X
ap-8720	507	3	29	29	NUM
ap-8720	507	4	]	]	PUNCT
ap-8720	507	5	f.	f.	PROPN
ap-8720	507	6	chollet	chollet	PROPN
ap-8720	507	7	.	.	PUNCT
ap-8720	508	1	xception	xception	NOUN
ap-8720	508	2	:	:	PUNCT
ap-8720	508	3	deep	deep	ADJ
ap-8720	508	4	learning	learn	VERB
ap-8720	508	5	with	with	ADP
ap-8720	508	6	depthwise	depthwise	NOUN
ap-8720	508	7	separable	separable	ADJ
ap-8720	508	8	convolutions	convolution	NOUN
ap-8720	508	9	.	.	PUNCT
ap-8720	509	1	in	in	ADP
ap-8720	509	2	ieee	ieee	NOUN
ap-8720	509	3	conference	conference	NOUN
ap-8720	509	4	on	on	ADP
ap-8720	509	5	computer	computer	NOUN
ap-8720	509	6	vision	vision	NOUN
ap-8720	509	7	and	and	CCONJ
ap-8720	509	8	pattern	pattern	NOUN
ap-8720	509	9	recognition	recognition	NOUN
ap-8720	509	10	(	(	PUNCT
ap-8720	509	11	cvpr	cvpr	NOUN
ap-8720	509	12	)	)	PUNCT
ap-8720	509	13	,	,	PUNCT
ap-8720	509	14	pp	pp	ADJ
ap-8720	509	15	.	.	PUNCT
ap-8720	510	1	1800–1807	1800–1807	NUM
ap-8720	510	2	.	.	NOUN
ap-8720	510	3	2017	2017	NUM
ap-8720	510	4	.	.	PUNCT
ap-8720	511	1	https://doi.org/10.1109/cvpr.2017.195	https://doi.org/10.1109/cvpr.2017.195	PROPN
ap-8720	511	2	[	[	NOUN
ap-8720	511	3	30	30	NUM
ap-8720	511	4	]	]	X
ap-8720	511	5	s.	s.	PROPN
ap-8720	511	6	lee	lee	PROPN
ap-8720	511	7	,	,	PUNCT
ap-8720	511	8	s.	s.	PROPN
ap-8720	511	9	tariq	tariq	PROPN
ap-8720	511	10	,	,	PUNCT
ap-8720	511	11	y.	y.	PROPN
ap-8720	511	12	shin	shin	PROPN
ap-8720	511	13	,	,	PUNCT
ap-8720	511	14	s.	s.	PROPN
ap-8720	511	15	s.	s.	PROPN
ap-8720	511	16	woo	woo	PROPN
ap-8720	511	17	.	.	PUNCT
ap-8720	512	1	detecting	detect	VERB
ap-8720	512	2	handcrafted	handcraft	VERB
ap-8720	512	3	facial	facial	ADJ
ap-8720	512	4	image	image	NOUN
ap-8720	512	5	manipulations	manipulation	NOUN
ap-8720	512	6	and	and	CCONJ
ap-8720	512	7	gan	gin	VERB
ap-8720	512	8	-	-	PUNCT
ap-8720	512	9	generated	generate	VERB
ap-8720	512	10	facial	facial	ADJ
ap-8720	512	11	images	image	NOUN
ap-8720	512	12	using	use	VERB
ap-8720	512	13	shallow	shallow	NOUN
ap-8720	512	14	-	-	PUNCT
ap-8720	512	15	fakefacenet	fakefacenet	NOUN
ap-8720	512	16	.	.	PUNCT
ap-8720	513	1	applied	apply	VERB
ap-8720	513	2	soft	soft	ADJ
ap-8720	513	3	computing	computing	NOUN
ap-8720	513	4	105:107256	105:107256	NUM
ap-8720	513	5	,	,	PUNCT
ap-8720	513	6	2021	2021	NUM
ap-8720	513	7	.	.	PUNCT
ap-8720	514	1	https://doi.org/10.1016/j.asoc.2021.107256	https://doi.org/10.1016/j.asoc.2021.107256	ADJ
ap-8720	514	2	318	318	NUM
ap-8720	514	3	https://doi.org/10.1109/34.927467	https://doi.org/10.1109/34.927467	ADJ
ap-8720	514	4	https://doi.org/10.1109/cvpr.2016.262	https://doi.org/10.1109/cvpr.2016.262	NOUN
ap-8720	514	5	https://doi.org/10.1145/3072959.3073640	https://doi.org/10.1145/3072959.3073640	ADJ
ap-8720	514	6	https://doi.org/10.22215/timreview/1282	https://doi.org/10.22215/timreview/1282	PROPN
ap-8720	514	7	https://doi.org/10.1109/mipr.2019.00103	https://doi.org/10.1109/mipr.2019.00103	PROPN
ap-8720	514	8	https://arxiv.org/pdf/1901.08971.pdf	https://arxiv.org/pdf/1901.08971.pdf	PROPN
ap-8720	514	9	https://arxiv.org/pdf/1909.12962.pdf	https://arxiv.org/pdf/1909.12962.pdf	PROPN
ap-8720	514	10	https://doi.org/10.23919/eusipco.2019.8903181	https://doi.org/10.23919/eusipco.2019.8903181	PROPN
ap-8720	514	11	https://doi.org/10.1109/cvprw.2017.229	https://doi.org/10.1109/cvprw.2017.229	PROPN
ap-8720	514	12	https://doi.org/10.1016/j.cviu.2021.103170	https://doi.org/10.1016/j.cviu.2021.103170	PROPN
ap-8720	514	13	https://doi.org/10.3390/electronics8030292	https://doi.org/10.3390/electronics8030292	NOUN
ap-8720	514	14	https://doi.org/10.4258/hir.2016.22.4.351	https://doi.org/10.4258/hir.2016.22.4.351	ADP
ap-8720	514	15	https://doi.org/10.1109/avss.2018.8639163	https://doi.org/10.1109/avss.2018.8639163	PROPN
ap-8720	514	16	https://doi.org/10.1109/cvpr.2008.4587756	https://doi.org/10.1109/cvpr.2008.4587756	PROPN
ap-8720	514	17	https://doi.org/10.1109/cvpr42600.2020.00808	https://doi.org/10.1109/cvpr42600.2020.00808	PROPN
ap-8720	514	18	https://arxiv.org/pdf/1812.04948.pdf	https://arxiv.org/pdf/1812.04948.pdf	PROPN
ap-8720	514	19	https://arxiv.org/pdf/1710.10196.pdf	https://arxiv.org/pdf/1710.10196.pdf	PROPN
ap-8720	514	20	https://doi.org/10.1007/978-981-15-7421-4_27	https://doi.org/10.1007/978-981-15-7421-4_27	PRON
ap-8720	514	21	https://doi.org/10.1109/icengtechnol.2017.8308186	https://doi.org/10.1109/icengtechnol.2017.8308186	PROPN
ap-8720	514	22	https://doi.org/10.1109/icengtechnol.2017.8308186	https://doi.org/10.1109/icengtechnol.2017.8308186	PROPN
ap-8720	514	23	https://doi.org/10.1109/access.2020.2988660	https://doi.org/10.1109/access.2020.2988660	PROPN
ap-8720	514	24	https://doi.org/10.1109/icb45273.2019.8987375	https://doi.org/10.1109/icb45273.2019.8987375	NUM
ap-8720	514	25	https://doi.org/10.1109/icb45273.2019.8987375	https://doi.org/10.1109/icb45273.2019.8987375	NOUN
ap-8720	515	1	https://doi.org/10.1109/cvpr.2015.7298682	https://doi.org/10.1109/cvpr.2015.7298682	PROPN
ap-8720	515	2	https://arxiv.org/pdf/1811.00656.pdf	https://arxiv.org/pdf/1811.00656.pdf	X
ap-8720	515	3	https://doi.org/10.1109/icassp.2019.8683164	https://doi.org/10.1109/icassp.2019.8683164	PROPN
ap-8720	515	4	https://doi.org/10.1109/jstsp.2020.2999185	https://doi.org/10.1109/jstsp.2020.2999185	PROPN
ap-8720	515	5	https://doi.org/10.1109/cvpr.2017.195	https://doi.org/10.1109/cvpr.2017.195	ADJ
ap-8720	515	6	https://doi.org/10.1016/j.asoc.2021.107256	https://doi.org/10.1016/j.asoc.2021.107256	ADJ
ap-8720	515	7	vol	vol	NOUN
ap-8720	515	8	.	.	PUNCT
ap-8720	516	1	63	63	NUM
ap-8720	516	2	no	no	NOUN
ap-8720	516	3	.	.	PUNCT
ap-8720	517	1	5/2023	5/2023	NUM
ap-8720	517	2	fake	fake	ADJ
ap-8720	517	3	face	face	NOUN
ap-8720	517	4	detection	detection	NOUN
ap-8720	517	5	based	base	VERB
ap-8720	517	6	on	on	ADP
ap-8720	517	7	a	a	DET
ap-8720	517	8	multi	multi	ADJ
ap-8720	517	9	discriminator	discriminator	NOUN
ap-8720	517	10	deep	deep	ADJ
ap-8720	517	11	cnn	cnn	PROPN
ap-8720	517	12	architecture	architecture	NOUN
ap-8720	517	13	[	[	X
ap-8720	517	14	31	31	NUM
ap-8720	517	15	]	]	PUNCT
ap-8720	517	16	i.	i.	PROPN
ap-8720	517	17	j.	j.	PROPN
ap-8720	517	18	goodfellow	goodfellow	PROPN
ap-8720	517	19	,	,	PUNCT
ap-8720	517	20	j.	j.	PROPN
ap-8720	517	21	shlens	shlens	PROPN
ap-8720	517	22	,	,	PUNCT
ap-8720	517	23	c.	c.	PROPN
ap-8720	517	24	szegedy	szegedy	PROPN
ap-8720	517	25	.	.	PUNCT
ap-8720	518	1	explaining	explain	VERB
ap-8720	518	2	and	and	CCONJ
ap-8720	518	3	harnessing	harness	VERB
ap-8720	518	4	adversarial	adversarial	ADJ
ap-8720	518	5	examples	example	NOUN
ap-8720	518	6	.	.	PUNCT
ap-8720	519	1	[	[	X
ap-8720	519	2	2023	2023	NUM
ap-8720	519	3	-	-	SYM
ap-8720	519	4	05	05	NUM
ap-8720	519	5	-	-	SYM
ap-8720	519	6	31	31	NUM
ap-8720	519	7	]	]	PUNCT
ap-8720	519	8	.	.	PUNCT
ap-8720	519	9	arxiv:1412.6572	arxiv:1412.6572	NUM
ap-8720	520	1	[	[	X
ap-8720	520	2	32	32	NUM
ap-8720	520	3	]	]	PUNCT
ap-8720	520	4	a.	a.	PROPN
ap-8720	520	5	radford	radford	PROPN
ap-8720	520	6	,	,	PUNCT
ap-8720	520	7	l.	l.	PROPN
ap-8720	520	8	metz	metz	PROPN
ap-8720	520	9	,	,	PUNCT
ap-8720	520	10	s.	s.	PROPN
ap-8720	520	11	chintala	chintala	PROPN
ap-8720	520	12	.	.	PUNCT
ap-8720	521	1	unsupervised	unsupervised	ADJ
ap-8720	521	2	representation	representation	NOUN
ap-8720	521	3	learning	learn	VERB
ap-8720	521	4	with	with	ADP
ap-8720	521	5	deep	deep	ADJ
ap-8720	521	6	convolutional	convolutional	ADJ
ap-8720	521	7	generative	generative	ADJ
ap-8720	521	8	adversarial	adversarial	ADJ
ap-8720	521	9	networks	network	NOUN
ap-8720	521	10	.	.	PUNCT
ap-8720	522	1	[	[	X
ap-8720	522	2	2023	2023	NUM
ap-8720	522	3	-	-	SYM
ap-8720	522	4	05	05	NUM
ap-8720	522	5	-	-	SYM
ap-8720	522	6	31	31	NUM
ap-8720	522	7	]	]	PUNCT
ap-8720	522	8	.	.	PUNCT
ap-8720	523	1	arxiv:1511.06434	arxiv:1511.06434	NOUN
ap-8720	523	2	[	[	X
ap-8720	523	3	33	33	NUM
ap-8720	523	4	]	]	PUNCT
ap-8720	523	5	a.	a.	NOUN
ap-8720	523	6	mangal	mangal	PROPN
ap-8720	523	7	,	,	PUNCT
ap-8720	523	8	n.	n.	PROPN
ap-8720	523	9	kumar	kumar	PROPN
ap-8720	523	10	.	.	PUNCT
ap-8720	524	1	using	use	VERB
ap-8720	524	2	big	big	ADJ
ap-8720	524	3	data	datum	NOUN
ap-8720	524	4	to	to	PART
ap-8720	524	5	enhance	enhance	VERB
ap-8720	524	6	the	the	DET
ap-8720	524	7	bosch	bosch	PROPN
ap-8720	524	8	production	production	NOUN
ap-8720	524	9	line	line	NOUN
ap-8720	524	10	performance	performance	NOUN
ap-8720	524	11	:	:	PUNCT
ap-8720	524	12	a	a	DET
ap-8720	524	13	kaggle	kaggle	ADJ
ap-8720	524	14	challenge	challenge	NOUN
ap-8720	524	15	.	.	PUNCT
ap-8720	525	1	in	in	ADP
ap-8720	525	2	ieee	ieee	PROPN
ap-8720	525	3	international	international	PROPN
ap-8720	525	4	conference	conference	NOUN
ap-8720	525	5	on	on	ADP
ap-8720	525	6	big	big	ADJ
ap-8720	525	7	data	datum	NOUN
ap-8720	525	8	(	(	PUNCT
ap-8720	525	9	big	big	ADJ
ap-8720	525	10	data	datum	NOUN
ap-8720	525	11	)	)	PUNCT
ap-8720	525	12	,	,	PUNCT
ap-8720	525	13	pp	pp	ADJ
ap-8720	525	14	.	.	PUNCT
ap-8720	526	1	2029–2035	2029–2035	NUM
ap-8720	526	2	.	.	PUNCT
ap-8720	527	1	2016	2016	NUM
ap-8720	527	2	.	.	PUNCT
ap-8720	528	1	https://doi.org/10.1109/bigdata.2016.7840826	https://doi.org/10.1109/bigdata.2016.7840826	PROPN
ap-8720	529	1	[	[	X
ap-8720	529	2	34	34	NUM
ap-8720	529	3	]	]	PUNCT
ap-8720	529	4	t.	t.	PROPN
ap-8720	529	5	carneiro	carneiro	PROPN
ap-8720	529	6	,	,	PUNCT
ap-8720	529	7	r.	r.	PROPN
ap-8720	529	8	v.	v.	PROPN
ap-8720	529	9	medeiros	medeiros	PROPN
ap-8720	529	10	da	da	PROPN
ap-8720	529	11	nóbrega	nóbrega	PROPN
ap-8720	529	12	,	,	PUNCT
ap-8720	529	13	t.	t.	PROPN
ap-8720	529	14	nepomuceno	nepomuceno	PROPN
ap-8720	529	15	,	,	PUNCT
ap-8720	529	16	et	et	PROPN
ap-8720	529	17	al	al	PROPN
ap-8720	529	18	.	.	PROPN
ap-8720	529	19	performance	performance	NOUN
ap-8720	529	20	analysis	analysis	NOUN
ap-8720	529	21	of	of	ADP
ap-8720	529	22	google	google	PROPN
ap-8720	529	23	colaboratory	colaboratory	PROPN
ap-8720	529	24	as	as	ADP
ap-8720	529	25	a	a	DET
ap-8720	529	26	tool	tool	NOUN
ap-8720	529	27	for	for	ADP
ap-8720	529	28	accelerating	accelerate	VERB
ap-8720	529	29	deep	deep	ADJ
ap-8720	529	30	learning	learning	NOUN
ap-8720	529	31	applications	application	NOUN
ap-8720	529	32	.	.	PUNCT
ap-8720	530	1	ieee	ieee	NOUN
ap-8720	530	2	access	access	NOUN
ap-8720	530	3	6:61677–61685	6:61677–61685	NUM
ap-8720	530	4	,	,	PUNCT
ap-8720	530	5	2018	2018	NUM
ap-8720	530	6	.	.	PUNCT
ap-8720	531	1	https://doi.org/10.1109/access.2018.2874767	https://doi.org/10.1109/access.2018.2874767	PROPN
ap-8720	531	2	319	319	NUM
ap-8720	531	3	https://arxiv.org/pdf/1412.6572.pdf	https://arxiv.org/pdf/1412.6572.pdf	PROPN
ap-8720	531	4	https://arxiv.org/pdf/1511.06434.pdf	https://arxiv.org/pdf/1511.06434.pdf	PROPN
ap-8720	531	5	https://doi.org/10.1109/bigdata.2016.7840826	https://doi.org/10.1109/bigdata.2016.7840826	PROPN
ap-8720	531	6	https://doi.org/10.1109/access.2018.2874767	https://doi.org/10.1109/access.2018.2874767	PROPN
ap-8720	531	7	acta	acta	PROPN
ap-8720	531	8	polytechnica	polytechnica	PROPN
ap-8720	531	9	63(5):305–319	63(5):305–319	PROPN
ap-8720	531	10	,	,	PUNCT
ap-8720	531	11	2023	2023	NUM
ap-8720	531	12	1	1	NUM
ap-8720	531	13	introduction	introduction	NOUN
ap-8720	531	14	2	2	NUM
ap-8720	531	15	related	relate	VERB
ap-8720	531	16	work	work	NOUN
ap-8720	531	17	3	3	NUM
ap-8720	531	18	proposed	propose	VERB
ap-8720	531	19	system	system	NOUN
ap-8720	531	20	architecture	architecture	NOUN
ap-8720	531	21	3.1	3.1	NUM
ap-8720	531	22	deep	deep	ADJ
ap-8720	531	23	convolutional	convolutional	ADJ
ap-8720	531	24	generative	generative	ADJ
ap-8720	531	25	adversarial	adversarial	ADJ
ap-8720	531	26	network	network	NOUN
ap-8720	531	27	(	(	PUNCT
ap-8720	531	28	dcgan	dcgan	NOUN
ap-8720	531	29	)	)	PUNCT
ap-8720	531	30	3.2	3.2	NUM
ap-8720	531	31	multi	multi	NOUN
ap-8720	531	32	discriminator	discriminator	NOUN
ap-8720	531	33	deep	deep	PROPN
ap-8720	531	34	cnn	cnn	PROPN
ap-8720	531	35	architecture	architecture	NOUN
ap-8720	531	36	(	(	PUNCT
ap-8720	531	37	mdd	mdd	PROPN
ap-8720	531	38	-	-	PUNCT
ap-8720	531	39	cnn	cnn	PROPN
ap-8720	531	40	)	)	PUNCT
ap-8720	531	41	3.2.1	3.2.1	NUM
ap-8720	531	42	new	new	ADJ
ap-8720	531	43	dcgan	dcgan	NOUN
ap-8720	531	44	architecture	architecture	NOUN
ap-8720	531	45	3.2.2	3.2.2	NUM
ap-8720	531	46	stability	stability	NOUN
ap-8720	531	47	conditions	condition	NOUN
ap-8720	531	48	4	4	NUM
ap-8720	531	49	experimental	experimental	ADJ
ap-8720	531	50	results	result	NOUN
ap-8720	531	51	4.1	4.1	NUM
ap-8720	531	52	dataset	dataset	VERB
ap-8720	531	53	4.2	4.2	NUM
ap-8720	531	54	evaluation	evaluation	NOUN
ap-8720	531	55	metrics	metric	NOUN
ap-8720	531	56	4.3	4.3	NUM
ap-8720	531	57	training	training	NOUN
ap-8720	531	58	and	and	CCONJ
ap-8720	531	59	validation	validation	NOUN
ap-8720	531	60	4.3.1	4.3.1	NUM
ap-8720	531	61	training	training	NOUN
ap-8720	531	62	of	of	ADP
ap-8720	531	63	discriminator	discriminator	NOUN
ap-8720	531	64	1	1	NUM
ap-8720	531	65	4.3.2	4.3.2	NUM
ap-8720	531	66	training	training	NOUN
ap-8720	531	67	of	of	ADP
ap-8720	531	68	discriminator	discriminator	NOUN
ap-8720	531	69	2	2	NUM
ap-8720	531	70	4.3.3	4.3.3	NOUN
ap-8720	531	71	training	training	NOUN
ap-8720	531	72	of	of	ADP
ap-8720	531	73	discriminator	discriminator	NOUN
ap-8720	531	74	3	3	NUM
ap-8720	531	75	4.4	4.4	NUM
ap-8720	531	76	test	test	NOUN
ap-8720	531	77	and	and	CCONJ
ap-8720	531	78	analysis	analysis	NOUN
ap-8720	531	79	5	5	NUM
ap-8720	531	80	conclusion	conclusion	NOUN
ap-8720	531	81	acknowledgements	acknowledgement	NOUN
ap-8720	531	82	references	reference	NOUN