id	sid	tid	token	lemma	pos
fcis-28893	1	1	frontiers	frontier	NOUN
fcis-28893	1	2	in	in	ADP
fcis-28893	1	3	computing	computing	NOUN
fcis-28893	1	4	and	and	CCONJ
fcis-28893	1	5	intelligent	intelligent	ADJ
fcis-28893	1	6	systems	system	NOUN
fcis-28893	1	7	issn	issn	VERB
fcis-28893	1	8	:	:	PUNCT
fcis-28893	1	9	2832	2832	NUM
fcis-28893	1	10	-	-	SYM
fcis-28893	1	11	6024	6024	NUM
fcis-28893	1	12	|	|	NOUN
fcis-28893	1	13	vol	vol	NOUN
fcis-28893	1	14	.	.	PROPN
fcis-28893	2	1	10	10	NUM
fcis-28893	2	2	,	,	PUNCT
fcis-28893	2	3	no	no	INTJ
fcis-28893	2	4	.	.	NOUN
fcis-28893	2	5	3	3	NUM
fcis-28893	2	6	,	,	PUNCT
fcis-28893	2	7	2024	2024	NUM
fcis-28893	2	8	118	118	NUM
fcis-28893	2	9	a	a	DET
fcis-28893	2	10	review	review	NOUN
fcis-28893	2	11	of	of	ADP
fcis-28893	2	12	text	text	NOUN
fcis-28893	2	13	-	-	PUNCT
fcis-28893	2	14	based	base	VERB
fcis-28893	2	15	pedestrian	pedestrian	NOUN
fcis-28893	2	16	retrieval	retrieval	NOUN
fcis-28893	2	17	methods	method	NOUN
fcis-28893	2	18	xiangmian	xiangmian	PROPN
fcis-28893	2	19	qiu	qiu	PROPN
fcis-28893	2	20	,	,	PUNCT
fcis-28893	2	21	yali	yali	PROPN
fcis-28893	2	22	zhang	zhang	PROPN
fcis-28893	3	1	*	*	PUNCT
fcis-28893	3	2	,	,	PUNCT
fcis-28893	3	3	yichen	yichen	ADV
fcis-28893	3	4	zhao	zhao	NOUN
fcis-28893	3	5	,	,	PUNCT
fcis-28893	3	6	jinzhao	jinzhao	PROPN
fcis-28893	3	7	li	li	PROPN
fcis-28893	3	8	school	school	PROPN
fcis-28893	3	9	of	of	ADP
fcis-28893	3	10	information	information	NOUN
fcis-28893	3	11	cyber	cyber	NOUN
fcis-28893	3	12	security	security	NOUN
fcis-28893	3	13	,	,	PUNCT
fcis-28893	3	14	people	people	NOUN
fcis-28893	3	15	’s	’s	PART
fcis-28893	3	16	public	public	ADJ
fcis-28893	3	17	security	security	NOUN
fcis-28893	3	18	university	university	PROPN
fcis-28893	3	19	of	of	ADP
fcis-28893	3	20	china	china	PROPN
fcis-28893	3	21	,	,	PUNCT
fcis-28893	3	22	beijing	beijing	PROPN
fcis-28893	3	23	,	,	PUNCT
fcis-28893	3	24	china	china	PROPN
fcis-28893	3	25	*	*	PUNCT
fcis-28893	3	26	corresponding	correspond	VERB
fcis-28893	3	27	author	author	NOUN
fcis-28893	3	28	:	:	PUNCT
fcis-28893	3	29	yali	yali	PROPN
fcis-28893	3	30	zhang	zhang	PROPN
fcis-28893	3	31	(	(	PUNCT
fcis-28893	3	32	email	email	NOUN
fcis-28893	3	33	:	:	PUNCT
fcis-28893	3	34	zhangyl_mail@163.com	zhangyl_mail@163.com	NUM
fcis-28893	3	35	)	)	PUNCT
fcis-28893	3	36	abstract	abstract	ADJ
fcis-28893	3	37	:	:	PUNCT
fcis-28893	3	38	text	text	NOUN
fcis-28893	3	39	-	-	PUNCT
fcis-28893	3	40	based	base	VERB
fcis-28893	3	41	pedestrian	pedestrian	NOUN
fcis-28893	3	42	retrieval	retrieval	NOUN
fcis-28893	3	43	task	task	NOUN
fcis-28893	3	44	uses	use	VERB
fcis-28893	3	45	textual	textual	ADJ
fcis-28893	3	46	descriptions	description	NOUN
fcis-28893	3	47	as	as	SCONJ
fcis-28893	3	48	query	query	NOUN
fcis-28893	3	49	inputs	input	NOUN
fcis-28893	3	50	to	to	PART
fcis-28893	3	51	retrieve	retrieve	VERB
fcis-28893	3	52	pedestrians	pedestrian	NOUN
fcis-28893	3	53	in	in	ADP
fcis-28893	3	54	image	image	NOUN
fcis-28893	3	55	gallery	gallery	NOUN
fcis-28893	3	56	,	,	PUNCT
fcis-28893	3	57	which	which	PRON
fcis-28893	3	58	is	be	AUX
fcis-28893	3	59	crucial	crucial	ADJ
fcis-28893	3	60	for	for	ADP
fcis-28893	3	61	social	social	ADJ
fcis-28893	3	62	security	security	NOUN
fcis-28893	3	63	and	and	CCONJ
fcis-28893	3	64	investigation	investigation	NOUN
fcis-28893	3	65	.	.	PUNCT
fcis-28893	4	1	by	by	ADP
fcis-28893	4	2	combing	comb	VERB
fcis-28893	4	3	the	the	DET
fcis-28893	4	4	relevant	relevant	ADJ
fcis-28893	4	5	working	working	NOUN
fcis-28893	4	6	literature	literature	NOUN
fcis-28893	4	7	,	,	PUNCT
fcis-28893	4	8	we	we	PRON
fcis-28893	4	9	summarize	summarize	VERB
fcis-28893	4	10	the	the	DET
fcis-28893	4	11	main	main	ADJ
fcis-28893	4	12	methods	method	NOUN
fcis-28893	4	13	in	in	ADP
fcis-28893	4	14	this	this	DET
fcis-28893	4	15	field	field	NOUN
fcis-28893	4	16	,	,	PUNCT
fcis-28893	4	17	which	which	PRON
fcis-28893	4	18	are	be	AUX
fcis-28893	4	19	grouped	group	VERB
fcis-28893	4	20	into	into	ADP
fcis-28893	4	21	five	five	NUM
fcis-28893	4	22	categories	category	NOUN
fcis-28893	4	23	of	of	ADP
fcis-28893	4	24	methods	method	NOUN
fcis-28893	4	25	based	base	VERB
fcis-28893	4	26	on	on	ADP
fcis-28893	4	27	feature	feature	NOUN
fcis-28893	4	28	matching	matching	NOUN
fcis-28893	4	29	,	,	PUNCT
fcis-28893	4	30	based	base	VERB
fcis-28893	4	31	on	on	ADP
fcis-28893	4	32	multi	multi	ADJ
fcis-28893	4	33	-	-	ADJ
fcis-28893	4	34	granularity	granularity	NOUN
fcis-28893	4	35	information	information	NOUN
fcis-28893	4	36	,	,	PUNCT
fcis-28893	4	37	based	base	VERB
fcis-28893	4	38	on	on	ADP
fcis-28893	4	39	adversarial	adversarial	ADJ
fcis-28893	4	40	ideal	ideal	NOUN
fcis-28893	4	41	,	,	PUNCT
fcis-28893	4	42	based	base	VERB
fcis-28893	4	43	on	on	ADP
fcis-28893	4	44	cross	cross	ADJ
fcis-28893	4	45	-	-	ADJ
fcis-28893	4	46	modal	modal	ADJ
fcis-28893	4	47	attention	attention	NOUN
fcis-28893	4	48	,	,	PUNCT
fcis-28893	4	49	and	and	CCONJ
fcis-28893	4	50	based	base	VERB
fcis-28893	4	51	on	on	ADP
fcis-28893	4	52	visual	visual	ADJ
fcis-28893	4	53	text	text	NOUN
fcis-28893	4	54	pre	pre	ADJ
fcis-28893	4	55	-	-	ADJ
fcis-28893	4	56	training	training	ADJ
fcis-28893	4	57	models	model	NOUN
fcis-28893	4	58	,	,	PUNCT
fcis-28893	4	59	and	and	CCONJ
fcis-28893	4	60	this	this	DET
fcis-28893	4	61	paper	paper	NOUN
fcis-28893	4	62	compare	compare	VERB
fcis-28893	4	63	and	and	CCONJ
fcis-28893	4	64	analyze	analyze	VERB
fcis-28893	4	65	the	the	DET
fcis-28893	4	66	design	design	NOUN
fcis-28893	4	67	ideas	idea	NOUN
fcis-28893	4	68	,	,	PUNCT
fcis-28893	4	69	method	method	NOUN
fcis-28893	4	70	features	feature	NOUN
fcis-28893	4	71	,	,	PUNCT
fcis-28893	4	72	advantages	advantage	NOUN
fcis-28893	4	73	and	and	CCONJ
fcis-28893	4	74	disadvantages	disadvantage	NOUN
fcis-28893	4	75	of	of	ADP
fcis-28893	4	76	classical	classical	ADJ
fcis-28893	4	77	model	model	NOUN
fcis-28893	4	78	of	of	ADP
fcis-28893	4	79	each	each	DET
fcis-28893	4	80	category	category	NOUN
fcis-28893	4	81	.	.	PUNCT
fcis-28893	5	1	the	the	DET
fcis-28893	5	2	performance	performance	NOUN
fcis-28893	5	3	of	of	ADP
fcis-28893	5	4	each	each	DET
fcis-28893	5	5	model	model	NOUN
fcis-28893	5	6	is	be	AUX
fcis-28893	5	7	compared	compare	VERB
fcis-28893	5	8	using	use	VERB
fcis-28893	5	9	the	the	DET
fcis-28893	5	10	commonly	commonly	ADV
fcis-28893	5	11	used	use	VERB
fcis-28893	5	12	datasets	dataset	NOUN
fcis-28893	5	13	and	and	CCONJ
fcis-28893	5	14	evaluation	evaluation	NOUN
fcis-28893	5	15	metrics	metric	NOUN
fcis-28893	5	16	(	(	PUNCT
fcis-28893	5	17	tpr	tpr	NOUN
fcis-28893	5	18	,	,	PUNCT
fcis-28893	5	19	map	map	NOUN
fcis-28893	5	20	)	)	PUNCT
fcis-28893	5	21	for	for	ADP
fcis-28893	5	22	this	this	DET
fcis-28893	5	23	task	task	NOUN
fcis-28893	5	24	,	,	PUNCT
fcis-28893	5	25	the	the	DET
fcis-28893	5	26	problems	problem	NOUN
fcis-28893	5	27	faced	face	VERB
fcis-28893	5	28	in	in	ADP
fcis-28893	5	29	field	field	NOUN
fcis-28893	5	30	are	be	AUX
fcis-28893	5	31	discussed	discuss	VERB
fcis-28893	5	32	and	and	CCONJ
fcis-28893	5	33	future	future	ADJ
fcis-28893	5	34	development	development	NOUN
fcis-28893	5	35	trends	trend	NOUN
fcis-28893	5	36	are	be	AUX
fcis-28893	5	37	envisioned	envision	VERB
fcis-28893	5	38	.	.	PUNCT
fcis-28893	6	1	recently	recently	ADV
fcis-28893	6	2	,	,	PUNCT
fcis-28893	6	3	the	the	DET
fcis-28893	6	4	method	method	NOUN
fcis-28893	6	5	based	base	VERB
fcis-28893	6	6	on	on	ADP
fcis-28893	6	7	vlp	vlp	PROPN
fcis-28893	6	8	has	have	AUX
fcis-28893	6	9	become	become	VERB
fcis-28893	6	10	a	a	DET
fcis-28893	6	11	mainstream	mainstream	NOUN
fcis-28893	6	12	,	,	PUNCT
fcis-28893	6	13	which	which	PRON
fcis-28893	6	14	can	can	AUX
fcis-28893	6	15	achieve	achieve	VERB
fcis-28893	6	16	higher	high	ADJ
fcis-28893	6	17	precision	precision	NOUN
fcis-28893	6	18	retrieval	retrieval	NOUN
fcis-28893	6	19	,	,	PUNCT
fcis-28893	6	20	but	but	CCONJ
fcis-28893	6	21	still	still	ADV
fcis-28893	6	22	faces	face	VERB
fcis-28893	6	23	the	the	DET
fcis-28893	6	24	problems	problem	NOUN
fcis-28893	6	25	of	of	ADP
fcis-28893	6	26	large	large	ADJ
fcis-28893	6	27	number	number	NOUN
fcis-28893	6	28	of	of	ADP
fcis-28893	6	29	model	model	NOUN
fcis-28893	6	30	parameters	parameter	NOUN
fcis-28893	6	31	and	and	CCONJ
fcis-28893	6	32	high	high	ADJ
fcis-28893	6	33	difficulty	difficulty	NOUN
fcis-28893	6	34	of	of	ADP
fcis-28893	6	35	training	training	NOUN
fcis-28893	6	36	,	,	PUNCT
fcis-28893	6	37	so	so	SCONJ
fcis-28893	6	38	it	it	PRON
fcis-28893	6	39	needs	need	VERB
fcis-28893	6	40	to	to	PART
fcis-28893	6	41	explore	explore	VERB
fcis-28893	6	42	the	the	DET
fcis-28893	6	43	lightweight	lightweight	ADJ
fcis-28893	6	44	solution	solution	NOUN
fcis-28893	6	45	in	in	ADP
fcis-28893	6	46	the	the	DET
fcis-28893	6	47	future	future	NOUN
fcis-28893	6	48	.	.	PUNCT
fcis-28893	7	1	keywords	keyword	NOUN
fcis-28893	7	2	:	:	PUNCT
fcis-28893	7	3	pedestrian	pedestrian	NOUN
fcis-28893	7	4	retrieval	retrieval	NOUN
fcis-28893	7	5	;	;	PUNCT
fcis-28893	7	6	cross	cross	ADJ
fcis-28893	7	7	-	-	ADJ
fcis-28893	7	8	modal	modal	ADJ
fcis-28893	7	9	retrieval	retrieval	NOUN
fcis-28893	7	10	;	;	PUNCT
fcis-28893	7	11	image	image	NOUN
fcis-28893	7	12	text	text	NOUN
fcis-28893	7	13	matching	matching	NOUN
fcis-28893	7	14	;	;	PUNCT
fcis-28893	7	15	visual	visual	ADJ
fcis-28893	7	16	language	language	NOUN
fcis-28893	7	17	pre	pre	ADJ
fcis-28893	7	18	-	-	ADJ
fcis-28893	7	19	training	training	ADJ
fcis-28893	7	20	models	model	NOUN
fcis-28893	7	21	.	.	PUNCT
fcis-28893	8	1	1	1	X
fcis-28893	8	2	.	.	X
fcis-28893	8	3	introduction	introduction	NOUN
fcis-28893	8	4	text	text	NOUN
fcis-28893	8	5	-	-	PUNCT
fcis-28893	8	6	based	base	VERB
fcis-28893	8	7	pedestrian	pedestrian	NOUN
fcis-28893	8	8	retrieval	retrieval	NOUN
fcis-28893	8	9	is	be	AUX
fcis-28893	8	10	closely	closely	ADV
fcis-28893	8	11	related	relate	VERB
fcis-28893	8	12	to	to	ADP
fcis-28893	8	13	the	the	DET
fcis-28893	8	14	tasks	task	NOUN
fcis-28893	8	15	of	of	ADP
fcis-28893	8	16	pedestrian	pedestrian	NOUN
fcis-28893	8	17	re	re	NOUN
fcis-28893	8	18	-	-	NOUN
fcis-28893	8	19	identification	identification	NOUN
fcis-28893	8	20	and	and	CCONJ
fcis-28893	8	21	image	image	NOUN
fcis-28893	8	22	retrieval	retrieval	NOUN
fcis-28893	8	23	.	.	PUNCT
fcis-28893	9	1	practically	practically	ADV
fcis-28893	9	2	,	,	PUNCT
fcis-28893	9	3	we	we	PRON
fcis-28893	9	4	face	face	VERB
fcis-28893	9	5	the	the	DET
fcis-28893	9	6	problem	problem	NOUN
fcis-28893	9	7	of	of	ADP
fcis-28893	9	8	lack	lack	NOUN
fcis-28893	9	9	of	of	ADP
fcis-28893	9	10	target	target	NOUN
fcis-28893	9	11	pedestrians	pedestrian	NOUN
fcis-28893	9	12	’	'	PUNCT
fcis-28893	9	13	image	image	NOUN
fcis-28893	9	14	or	or	CCONJ
fcis-28893	9	15	pedestrians	pedestrian	VERB
fcis-28893	9	16	real	real	ADJ
fcis-28893	9	17	-	-	PUNCT
fcis-28893	9	18	time	time	NOUN
fcis-28893	9	19	clothing	clothing	NOUN
fcis-28893	9	20	does	do	AUX
fcis-28893	9	21	not	not	PART
fcis-28893	9	22	match	match	VERB
fcis-28893	9	23	the	the	DET
fcis-28893	9	24	image	image	NOUN
fcis-28893	9	25	gallery	gallery	NOUN
fcis-28893	9	26	.	.	PUNCT
fcis-28893	10	1	in	in	ADP
fcis-28893	10	2	contrast	contrast	NOUN
fcis-28893	10	3	,	,	PUNCT
fcis-28893	10	4	text	text	NOUN
fcis-28893	10	5	-	-	PUNCT
fcis-28893	10	6	based	base	VERB
fcis-28893	10	7	retrieval	retrieval	NOUN
fcis-28893	10	8	methods	method	NOUN
fcis-28893	10	9	have	have	VERB
fcis-28893	10	10	a	a	DET
fcis-28893	10	11	greater	great	ADJ
fcis-28893	10	12	potential	potential	NOUN
fcis-28893	10	13	for	for	ADP
fcis-28893	10	14	application	application	NOUN
fcis-28893	10	15	because	because	SCONJ
fcis-28893	10	16	they	they	PRON
fcis-28893	10	17	use	use	VERB
fcis-28893	10	18	more	more	ADV
fcis-28893	10	19	accessible	accessible	ADJ
fcis-28893	10	20	text	text	NOUN
fcis-28893	10	21	as	as	ADP
fcis-28893	10	22	query	query	NOUN
fcis-28893	10	23	.	.	PUNCT
fcis-28893	11	1	to	to	PART
fcis-28893	11	2	better	well	ADV
fcis-28893	11	3	understand	understand	VERB
fcis-28893	11	4	the	the	DET
fcis-28893	11	5	content	content	NOUN
fcis-28893	11	6	of	of	ADP
fcis-28893	11	7	this	this	DET
fcis-28893	11	8	paper	paper	NOUN
fcis-28893	11	9	,	,	PUNCT
fcis-28893	11	10	we	we	PRON
fcis-28893	11	11	list	list	VERB
fcis-28893	11	12	the	the	DET
fcis-28893	11	13	definition	definition	NOUN
fcis-28893	11	14	of	of	ADP
fcis-28893	11	15	pedestrian	pedestrian	NOUN
fcis-28893	11	16	re	re	NOUN
fcis-28893	11	17	-	-	NOUN
fcis-28893	11	18	identification	identification	NOUN
fcis-28893	11	19	,	,	PUNCT
fcis-28893	11	20	person	person	NOUN
fcis-28893	11	21	search	search	NOUN
fcis-28893	11	22	and	and	CCONJ
fcis-28893	11	23	text	text	NOUN
fcis-28893	11	24	-	-	PUNCT
fcis-28893	11	25	based	base	VERB
fcis-28893	11	26	pedestrian	pedestrian	NOUN
fcis-28893	11	27	retrieval	retrieval	NOUN
fcis-28893	11	28	,	,	PUNCT
fcis-28893	11	29	and	and	CCONJ
fcis-28893	11	30	the	the	DET
fcis-28893	11	31	main	main	ADJ
fcis-28893	11	32	differences	difference	NOUN
fcis-28893	11	33	can	can	AUX
fcis-28893	11	34	be	be	AUX
fcis-28893	11	35	seen	see	VERB
fcis-28893	11	36	in	in	ADP
fcis-28893	11	37	table	table	NOUN
fcis-28893	11	38	1	1	NUM
fcis-28893	11	39	,	,	PUNCT
fcis-28893	11	40	the	the	DET
fcis-28893	11	41	process	process	NOUN
fcis-28893	11	42	shown	show	VERB
fcis-28893	11	43	in	in	ADP
fcis-28893	11	44	figure	figure	NOUN
fcis-28893	11	45	1	1	NUM
fcis-28893	11	46	.	.	PUNCT
fcis-28893	11	47	table	table	NOUN
fcis-28893	11	48	1	1	NUM
fcis-28893	11	49	.	.	PUNCT
fcis-28893	12	1	differences	difference	NOUN
fcis-28893	12	2	between	between	ADP
fcis-28893	12	3	pedestrian	pedestrian	NOUN
fcis-28893	12	4	re	re	NOUN
fcis-28893	12	5	-	-	NOUN
fcis-28893	12	6	identification	identification	NOUN
fcis-28893	12	7	,	,	PUNCT
fcis-28893	12	8	person	person	NOUN
fcis-28893	12	9	search	search	NOUN
fcis-28893	12	10	and	and	CCONJ
fcis-28893	12	11	text	text	NOUN
fcis-28893	12	12	-	-	PUNCT
fcis-28893	12	13	based	base	VERB
fcis-28893	12	14	pedestrian	pedestrian	NOUN
fcis-28893	12	15	retrieval	retrieval	NOUN
fcis-28893	12	16	tasks	task	NOUN
fcis-28893	12	17	task	task	NOUN
fcis-28893	12	18	query	query	NOUN
fcis-28893	12	19	input	input	NOUN
fcis-28893	12	20	application	application	NOUN
fcis-28893	12	21	scene	scene	NOUN
fcis-28893	12	22	goal	goal	NOUN
fcis-28893	12	23	pedestrian	pedestrian	PROPN
fcis-28893	12	24	re	re	NOUN
fcis-28893	12	25	-	-	NOUN
fcis-28893	12	26	identification	identification	NOUN
fcis-28893	12	27	imagery	imagery	NOUN
fcis-28893	12	28	cross	cross	NOUN
fcis-28893	12	29	-	-	NOUN
fcis-28893	12	30	device	device	ADJ
fcis-28893	12	31	,	,	PUNCT
fcis-28893	12	32	cross	cross	ADJ
fcis-28893	12	33	-	-	ADJ
fcis-28893	12	34	scene	scene	ADJ
fcis-28893	12	35	pedestrian	pedestrian	NOUN
fcis-28893	12	36	tracks	track	VERB
fcis-28893	12	37	pedestrian	pedestrian	NOUN
fcis-28893	12	38	search	search	NOUN
fcis-28893	12	39	imagery	imagery	NOUN
fcis-28893	12	40	single	single	ADJ
fcis-28893	12	41	device	device	NOUN
fcis-28893	12	42	,	,	PUNCT
fcis-28893	12	43	single	single	ADJ
fcis-28893	12	44	scene	scene	NOUN
fcis-28893	12	45	specific	specific	ADJ
fcis-28893	12	46	pedestrian	pedestrian	NOUN
fcis-28893	12	47	locations	location	NOUN
fcis-28893	12	48	text	text	NOUN
fcis-28893	12	49	-	-	PUNCT
fcis-28893	12	50	based	base	VERB
fcis-28893	12	51	pedestrian	pedestrian	NOUN
fcis-28893	12	52	retrieval	retrieval	NOUN
fcis-28893	12	53	copies	copy	NOUN
fcis-28893	12	54	cross	cross	NOUN
fcis-28893	12	55	-	-	NOUN
fcis-28893	12	56	device	device	ADJ
fcis-28893	12	57	,	,	PUNCT
fcis-28893	12	58	cross	cross	ADJ
fcis-28893	12	59	-	-	ADJ
fcis-28893	12	60	scene	scene	ADJ
fcis-28893	12	61	pedestrian	pedestrian	NOUN
fcis-28893	12	62	location	location	NOUN
fcis-28893	12	63	and	and	CCONJ
fcis-28893	12	64	tracking	track	VERB
fcis-28893	12	65	1	1	NUM
fcis-28893	12	66	)	)	PUNCT
fcis-28893	12	67	pedestrian	pedestrian	NOUN
fcis-28893	12	68	re	re	NOUN
fcis-28893	12	69	-	-	NOUN
fcis-28893	12	70	id	id	INTJ
fcis-28893	12	71	(	(	PUNCT
fcis-28893	12	72	pedestrian	pedestrian	NOUN
fcis-28893	12	73	re	re	NOUN
fcis-28893	12	74	-	-	NOUN
fcis-28893	12	75	id	id	NOUN
fcis-28893	12	76	):	):	PUNCT
fcis-28893	12	77	aims	aim	VERB
fcis-28893	12	78	at	at	ADP
fcis-28893	12	79	articulating	articulate	VERB
fcis-28893	12	80	the	the	DET
fcis-28893	12	81	tracking	tracking	NOUN
fcis-28893	12	82	of	of	ADP
fcis-28893	12	83	target	target	NOUN
fcis-28893	12	84	pedestrian	pedestrian	NOUN
fcis-28893	12	85	in	in	ADP
fcis-28893	12	86	different	different	ADJ
fcis-28893	12	87	monitoring	monitoring	NOUN
fcis-28893	12	88	areas	area	NOUN
fcis-28893	12	89	and	and	CCONJ
fcis-28893	12	90	reid	reid	PROPN
fcis-28893	12	91	pedestrian	pedestrian	PROPN
fcis-28893	12	92	tracking	track	VERB
fcis-28893	12	93	across	across	ADP
fcis-28893	12	94	time	time	NOUN
fcis-28893	12	95	,	,	PUNCT
fcis-28893	12	96	location	location	NOUN
fcis-28893	12	97	and	and	CCONJ
fcis-28893	12	98	devices	device	NOUN
fcis-28893	12	99	.	.	PUNCT
fcis-28893	13	1	its	its	PRON
fcis-28893	13	2	input	input	NOUN
fcis-28893	13	3	query	query	NOUN
fcis-28893	13	4	is	be	AUX
fcis-28893	13	5	generally	generally	ADV
fcis-28893	13	6	an	an	DET
fcis-28893	13	7	image	image	NOUN
fcis-28893	13	8	.	.	PUNCT
fcis-28893	14	1	2	2	X
fcis-28893	14	2	)	)	PUNCT
fcis-28893	14	3	pedestrian	pedestrian	NOUN
fcis-28893	14	4	search	search	NOUN
fcis-28893	14	5	(	(	PUNCT
fcis-28893	14	6	person	person	NOUN
fcis-28893	14	7	search	search	NOUN
fcis-28893	14	8	):	):	PUNCT
fcis-28893	14	9	it	it	PRON
fcis-28893	14	10	is	be	AUX
fcis-28893	14	11	to	to	PART
fcis-28893	14	12	retrieve	retrieve	VERB
fcis-28893	14	13	and	and	CCONJ
fcis-28893	14	14	match	match	VERB
fcis-28893	14	15	a	a	DET
fcis-28893	14	16	target	target	NOUN
fcis-28893	14	17	pedestrian	pedestrian	NOUN
fcis-28893	14	18	in	in	ADP
fcis-28893	14	19	a	a	DET
fcis-28893	14	20	complex	complex	ADJ
fcis-28893	14	21	scene	scene	NOUN
fcis-28893	14	22	given	give	VERB
fcis-28893	14	23	a	a	DET
fcis-28893	14	24	target	target	NOUN
fcis-28893	14	25	pedestrian	pedestrian	NOUN
fcis-28893	14	26	image	image	NOUN
fcis-28893	14	27	,	,	PUNCT
fcis-28893	14	28	using	use	VERB
fcis-28893	14	29	image	image	NOUN
fcis-28893	14	30	as	as	ADP
fcis-28893	14	31	a	a	DET
fcis-28893	14	32	query	query	NOUN
fcis-28893	14	33	.	.	PUNCT
fcis-28893	15	1	3	3	X
fcis-28893	15	2	)	)	PUNCT
fcis-28893	15	3	text	text	NOUN
fcis-28893	15	4	-	-	PUNCT
fcis-28893	15	5	based	base	VERB
fcis-28893	15	6	pedestrian	pedestrian	NOUN
fcis-28893	15	7	retrieval	retrieval	NOUN
fcis-28893	15	8	(	(	PUNCT
fcis-28893	15	9	text	text	NOUN
fcis-28893	15	10	-	-	PUNCT
fcis-28893	15	11	based	base	VERB
fcis-28893	15	12	pedestrian	pedestrian	NOUN
fcis-28893	15	13	retrieval	retrieval	NOUN
fcis-28893	15	14	):	):	PUNCT
fcis-28893	15	15	the	the	DET
fcis-28893	15	16	target	target	NOUN
fcis-28893	15	17	pedestrian	pedestrian	NOUN
fcis-28893	15	18	is	be	AUX
fcis-28893	15	19	retrieved	retrieve	VERB
fcis-28893	15	20	from	from	ADP
fcis-28893	15	21	the	the	DET
fcis-28893	15	22	image	image	NOUN
fcis-28893	15	23	gallery	gallery	NOUN
fcis-28893	15	24	by	by	ADP
fcis-28893	15	25	textual	textual	ADJ
fcis-28893	15	26	description	description	NOUN
fcis-28893	15	27	of	of	ADP
fcis-28893	15	28	the	the	DET
fcis-28893	15	29	pedestrian	pedestrian	NOUN
fcis-28893	15	30	.	.	PUNCT
fcis-28893	16	1	in	in	ADP
fcis-28893	16	2	this	this	DET
fcis-28893	16	3	paper	paper	NOUN
fcis-28893	16	4	,	,	PUNCT
fcis-28893	16	5	we	we	PRON
fcis-28893	16	6	have	have	AUX
fcis-28893	16	7	comprehensively	comprehensively	ADV
fcis-28893	16	8	sorted	sort	VERB
fcis-28893	16	9	out	out	ADP
fcis-28893	16	10	the	the	DET
fcis-28893	16	11	research	research	NOUN
fcis-28893	16	12	related	relate	VERB
fcis-28893	16	13	to	to	ADP
fcis-28893	16	14	text	text	NOUN
fcis-28893	16	15	-	-	PUNCT
fcis-28893	16	16	based	base	VERB
fcis-28893	16	17	pedestrian	pedestrian	NOUN
fcis-28893	16	18	retrieval	retrieval	NOUN
fcis-28893	16	19	technology	technology	NOUN
fcis-28893	16	20	since	since	SCONJ
fcis-28893	16	21	2017	2017	NUM
fcis-28893	16	22	,	,	PUNCT
fcis-28893	16	23	summarized	summarize	VERB
fcis-28893	16	24	the	the	DET
fcis-28893	16	25	current	current	ADJ
fcis-28893	16	26	research	research	NOUN
fcis-28893	16	27	status	status	NOUN
fcis-28893	16	28	at	at	ADP
fcis-28893	16	29	home	home	NOUN
fcis-28893	16	30	and	and	CCONJ
fcis-28893	16	31	abroad	abroad	ADV
fcis-28893	16	32	,	,	PUNCT
fcis-28893	16	33	explored	explore	VERB
fcis-28893	16	34	the	the	DET
fcis-28893	16	35	main	main	ADJ
fcis-28893	16	36	challenges	challenge	NOUN
fcis-28893	16	37	and	and	CCONJ
fcis-28893	16	38	difficulties	difficulty	NOUN
fcis-28893	16	39	faced	face	VERB
fcis-28893	16	40	by	by	ADP
fcis-28893	16	41	text	text	NOUN
fcis-28893	16	42	-	-	PUNCT
fcis-28893	16	43	based	base	VERB
fcis-28893	16	44	pedestrian	pedestrian	NOUN
fcis-28893	16	45	retrieval	retrieval	NOUN
fcis-28893	16	46	technology	technology	NOUN
fcis-28893	16	47	,	,	PUNCT
fcis-28893	16	48	analyzed	analyze	VERB
fcis-28893	16	49	the	the	DET
fcis-28893	16	50	advantages	advantage	NOUN
fcis-28893	16	51	and	and	CCONJ
fcis-28893	16	52	disadvantages	disadvantage	NOUN
fcis-28893	16	53	of	of	ADP
fcis-28893	16	54	various	various	ADJ
fcis-28893	16	55	methods	method	NOUN
fcis-28893	16	56	,	,	PUNCT
fcis-28893	16	57	and	and	CCONJ
fcis-28893	16	58	looked	look	VERB
fcis-28893	16	59	forward	forward	ADV
fcis-28893	16	60	to	to	ADP
fcis-28893	16	61	the	the	DET
fcis-28893	16	62	future	future	ADJ
fcis-28893	16	63	development	development	NOUN
fcis-28893	16	64	trend	trend	NOUN
fcis-28893	16	65	,	,	PUNCT
fcis-28893	16	66	and	and	CCONJ
fcis-28893	16	67	the	the	DET
fcis-28893	16	68	main	main	ADJ
fcis-28893	16	69	contributions	contribution	NOUN
fcis-28893	16	70	of	of	ADP
fcis-28893	16	71	this	this	DET
fcis-28893	16	72	paper	paper	NOUN
fcis-28893	16	73	:	:	PUNCT
fcis-28893	16	74	(	(	PUNCT
fcis-28893	16	75	1	1	X
fcis-28893	16	76	)	)	PUNCT
fcis-28893	16	77	the	the	DET
fcis-28893	16	78	text	text	NOUN
fcis-28893	16	79	-	-	PUNCT
fcis-28893	16	80	based	base	VERB
fcis-28893	16	81	pedestrian	pedestrian	NOUN
fcis-28893	16	82	retrieval	retrieval	NOUN
fcis-28893	16	83	methods	method	NOUN
fcis-28893	16	84	are	be	AUX
fcis-28893	16	85	systematically	systematically	ADV
fcis-28893	16	86	sorted	sort	VERB
fcis-28893	16	87	out	out	ADP
fcis-28893	16	88	and	and	CCONJ
fcis-28893	16	89	summarized	summarize	VERB
fcis-28893	16	90	into	into	ADP
fcis-28893	16	91	five	five	NUM
fcis-28893	16	92	categories	category	NOUN
fcis-28893	16	93	:	:	PUNCT
fcis-28893	16	94	global	global	ADJ
fcis-28893	16	95	matching	matching	NOUN
fcis-28893	16	96	,	,	PUNCT
fcis-28893	16	97	multi	multi	ADJ
fcis-28893	16	98	-	-	ADJ
fcis-28893	16	99	granularity	granularity	NOUN
fcis-28893	16	100	information	information	NOUN
fcis-28893	16	101	,	,	PUNCT
fcis-28893	16	102	adversarial	adversarial	ADJ
fcis-28893	16	103	learning	learning	NOUN
fcis-28893	16	104	,	,	PUNCT
fcis-28893	16	105	cross	cross	ADJ
fcis-28893	16	106	-	-	ADJ
fcis-28893	16	107	modal	modal	ADJ
fcis-28893	16	108	attention	attention	NOUN
fcis-28893	16	109	mechanism	mechanism	NOUN
fcis-28893	16	110	,	,	PUNCT
fcis-28893	16	111	visual	visual	ADJ
fcis-28893	16	112	-	-	PUNCT
fcis-28893	16	113	textual	textual	ADJ
fcis-28893	16	114	pre	pre	ADJ
fcis-28893	16	115	-	-	ADJ
fcis-28893	16	116	training	training	ADJ
fcis-28893	16	117	model	model	NOUN
fcis-28893	16	118	,	,	PUNCT
fcis-28893	16	119	and	and	CCONJ
fcis-28893	16	120	the	the	DET
fcis-28893	16	121	technical	technical	ADJ
fcis-28893	16	122	routes	route	NOUN
fcis-28893	16	123	,	,	PUNCT
fcis-28893	16	124	model	model	NOUN
fcis-28893	16	125	structures	structure	NOUN
fcis-28893	16	126	,	,	PUNCT
fcis-28893	16	127	advantages	advantage	NOUN
fcis-28893	16	128	and	and	CCONJ
fcis-28893	16	129	disadvantages	disadvantage	NOUN
fcis-28893	16	130	of	of	ADP
fcis-28893	16	131	each	each	DET
fcis-28893	16	132	type	type	NOUN
fcis-28893	16	133	of	of	ADP
fcis-28893	16	134	methods	method	NOUN
fcis-28893	16	135	are	be	AUX
fcis-28893	16	136	analyzed	analyze	VERB
fcis-28893	16	137	.	.	PUNCT
fcis-28893	17	1	(	(	PUNCT
fcis-28893	17	2	2	2	X
fcis-28893	17	3	)	)	PUNCT
fcis-28893	17	4	based	base	VERB
fcis-28893	17	5	on	on	ADP
fcis-28893	17	6	the	the	DET
fcis-28893	17	7	cuhk	cuhk	NOUN
fcis-28893	17	8	-	-	PUNCT
fcis-28893	17	9	pedes	pes	NOUN
fcis-28893	17	10	,	,	PUNCT
fcis-28893	17	11	icfg	icfg	NOUN
fcis-28893	17	12	-	-	ADJ
fcis-28893	17	13	pedes	pes	NOUN
fcis-28893	17	14	and	and	CCONJ
fcis-28893	17	15	rstpreid	rstpreid	NOUN
fcis-28893	17	16	datasets	dataset	NOUN
fcis-28893	17	17	,	,	PUNCT
fcis-28893	17	18	the	the	DET
fcis-28893	17	19	performance	performance	NOUN
fcis-28893	17	20	of	of	ADP
fcis-28893	17	21	the	the	DET
fcis-28893	17	22	classical	classical	ADJ
fcis-28893	17	23	models	model	NOUN
fcis-28893	17	24	was	be	AUX
fcis-28893	17	25	compared	compare	VERB
fcis-28893	17	26	using	use	VERB
fcis-28893	17	27	tpr	tpr	NOUN
fcis-28893	17	28	.	.	PUNCT
fcis-28893	18	1	(	(	PUNCT
fcis-28893	18	2	3	3	X
fcis-28893	18	3	)	)	PUNCT
fcis-28893	18	4	the	the	DET
fcis-28893	18	5	future	future	ADJ
fcis-28893	18	6	development	development	NOUN
fcis-28893	18	7	trend	trend	NOUN
fcis-28893	18	8	of	of	ADP
fcis-28893	18	9	text	text	NOUN
fcis-28893	18	10	-	-	PUNCT
fcis-28893	18	11	based	base	VERB
fcis-28893	18	12	pedestrian	pedestrian	NOUN
fcis-28893	18	13	retrieval	retrieval	NOUN
fcis-28893	18	14	technology	technology	NOUN
fcis-28893	18	15	is	be	AUX
fcis-28893	18	16	envisioned	envision	VERB
fcis-28893	18	17	to	to	PART
fcis-28893	18	18	provide	provide	VERB
fcis-28893	18	19	reference	reference	NOUN
fcis-28893	18	20	for	for	ADP
fcis-28893	18	21	subsequent	subsequent	ADJ
fcis-28893	18	22	research	research	NOUN
fcis-28893	18	23	.	.	PUNCT
fcis-28893	19	1	(	(	PUNCT
fcis-28893	19	2	a	a	X
fcis-28893	19	3	)	)	PUNCT
fcis-28893	19	4	pedestrian	pedestrian	NOUN
fcis-28893	19	5	re	re	NOUN
fcis-28893	19	6	-	-	NOUN
fcis-28893	19	7	identification	identification	NOUN
fcis-28893	19	8	(	(	PUNCT
fcis-28893	19	9	b)person	b)person	NOUN
fcis-28893	19	10	search	search	NOUN
fcis-28893	19	11	(	(	PUNCT
fcis-28893	19	12	c)text	c)text	PROPN
fcis-28893	19	13	based	base	VERB
fcis-28893	19	14	pedestrian	pedestrian	NOUN
fcis-28893	19	15	retrieval	retrieval	NOUN
fcis-28893	19	16	fig	fig	NOUN
fcis-28893	19	17	1	1	NUM
fcis-28893	19	18	.	.	PUNCT
fcis-28893	19	19	brief	brief	ADJ
fcis-28893	19	20	schematic	schematic	ADJ
fcis-28893	19	21	diagram	diagram	NOUN
fcis-28893	19	22	of	of	ADP
fcis-28893	19	23	pedestrian	pedestrian	NOUN
fcis-28893	19	24	re	re	NOUN
fcis-28893	19	25	-	-	NOUN
fcis-28893	19	26	identification	identification	NOUN
fcis-28893	19	27	,	,	PUNCT
fcis-28893	19	28	pedestrian	pedestrian	NOUN
fcis-28893	19	29	search	search	NOUN
fcis-28893	19	30	,	,	PUNCT
fcis-28893	19	31	and	and	CCONJ
fcis-28893	19	32	text	text	NOUN
fcis-28893	19	33	based	base	VERB
fcis-28893	19	34	pedestrian	pedestrian	NOUN
fcis-28893	19	35	retrieval	retrieval	NOUN
fcis-28893	19	36	process	process	NOUN
fcis-28893	19	37	119	119	NUM
fcis-28893	19	38	2	2	NUM
fcis-28893	19	39	.	.	PUNCT
fcis-28893	20	1	overview	overview	NOUN
fcis-28893	20	2	of	of	ADP
fcis-28893	20	3	the	the	DET
fcis-28893	20	4	development	development	NOUN
fcis-28893	20	5	of	of	ADP
fcis-28893	20	6	text	text	NOUN
fcis-28893	20	7	-	-	PUNCT
fcis-28893	20	8	based	base	VERB
fcis-28893	20	9	pedestrian	pedestrian	NOUN
fcis-28893	20	10	retrieval	retrieval	NOUN
fcis-28893	20	11	the	the	DET
fcis-28893	20	12	core	core	NOUN
fcis-28893	20	13	challenge	challenge	NOUN
fcis-28893	20	14	of	of	ADP
fcis-28893	20	15	text	text	NOUN
fcis-28893	20	16	-	-	PUNCT
fcis-28893	20	17	based	base	VERB
fcis-28893	20	18	pedestrian	pedestrian	NOUN
fcis-28893	20	19	retrieval	retrieval	NOUN
fcis-28893	20	20	is	be	AUX
fcis-28893	20	21	how	how	SCONJ
fcis-28893	20	22	to	to	PART
fcis-28893	20	23	effectively	effectively	ADV
fcis-28893	20	24	bridge	bridge	VERB
fcis-28893	20	25	the	the	DET
fcis-28893	20	26	modal	modal	ADJ
fcis-28893	20	27	differences	difference	NOUN
fcis-28893	20	28	between	between	ADP
fcis-28893	20	29	images	image	NOUN
fcis-28893	20	30	and	and	CCONJ
fcis-28893	20	31	texts	text	NOUN
fcis-28893	20	32	.	.	PUNCT
fcis-28893	21	1	in	in	ADP
fcis-28893	21	2	this	this	DET
fcis-28893	21	3	paper	paper	NOUN
fcis-28893	21	4	,	,	PUNCT
fcis-28893	21	5	we	we	PRON
fcis-28893	21	6	categorize	categorize	VERB
fcis-28893	21	7	the	the	DET
fcis-28893	21	8	existing	exist	VERB
fcis-28893	21	9	methods	method	NOUN
fcis-28893	21	10	into	into	ADP
fcis-28893	21	11	the	the	DET
fcis-28893	21	12	following	follow	VERB
fcis-28893	21	13	types	type	NOUN
fcis-28893	21	14	:	:	PUNCT
fcis-28893	21	15	based	base	VERB
fcis-28893	21	16	on	on	ADP
fcis-28893	21	17	global	global	ADJ
fcis-28893	21	18	feature	feature	NOUN
fcis-28893	21	19	matching	matching	NOUN
fcis-28893	21	20	,	,	PUNCT
fcis-28893	21	21	based	base	VERB
fcis-28893	21	22	on	on	ADP
fcis-28893	21	23	multi	multi	ADJ
fcis-28893	21	24	-	-	ADJ
fcis-28893	21	25	granularity	granularity	NOUN
fcis-28893	21	26	information	information	NOUN
fcis-28893	21	27	,	,	PUNCT
fcis-28893	21	28	based	base	VERB
fcis-28893	21	29	on	on	ADP
fcis-28893	21	30	adversarial	adversarial	ADJ
fcis-28893	21	31	learning	learning	NOUN
fcis-28893	21	32	,	,	PUNCT
fcis-28893	21	33	based	base	VERB
fcis-28893	21	34	on	on	ADP
fcis-28893	21	35	cross	cross	ADJ
fcis-28893	21	36	-	-	ADJ
fcis-28893	21	37	modal	modal	ADJ
fcis-28893	21	38	attention	attention	NOUN
fcis-28893	21	39	mechanism	mechanism	NOUN
fcis-28893	21	40	,	,	PUNCT
fcis-28893	21	41	and	and	CCONJ
fcis-28893	21	42	based	base	VERB
fcis-28893	21	43	on	on	ADP
fcis-28893	21	44	visual	visual	ADJ
fcis-28893	21	45	-	-	PUNCT
fcis-28893	21	46	text	text	NOUN
fcis-28893	21	47	pre	pre	ADJ
fcis-28893	21	48	-	-	ADJ
fcis-28893	21	49	training	training	ADJ
fcis-28893	21	50	model	model	NOUN
fcis-28893	21	51	.	.	PUNCT
fcis-28893	22	1	1	1	NUM
fcis-28893	22	2	)	)	PUNCT
fcis-28893	22	3	global	global	ADJ
fcis-28893	22	4	feature	feature	NOUN
fcis-28893	22	5	matching	matching	NOUN
fcis-28893	22	6	-	-	PUNCT
fcis-28893	22	7	based	base	VERB
fcis-28893	22	8	approach	approach	NOUN
fcis-28893	22	9	the	the	DET
fcis-28893	22	10	core	core	NOUN
fcis-28893	22	11	idea	idea	NOUN
fcis-28893	22	12	of	of	ADP
fcis-28893	22	13	this	this	DET
fcis-28893	22	14	type	type	NOUN
fcis-28893	22	15	of	of	ADP
fcis-28893	22	16	method	method	NOUN
fcis-28893	22	17	is	be	AUX
fcis-28893	22	18	to	to	PART
fcis-28893	22	19	use	use	VERB
fcis-28893	22	20	cnn	cnn	PROPN
fcis-28893	22	21	and	and	CCONJ
fcis-28893	22	22	rnn	rnn	VERB
fcis-28893	22	23	to	to	PART
fcis-28893	22	24	extract	extract	VERB
fcis-28893	22	25	image	image	NOUN
fcis-28893	22	26	and	and	CCONJ
fcis-28893	22	27	text	text	NOUN
fcis-28893	22	28	features	feature	VERB
fcis-28893	22	29	respectively	respectively	ADV
fcis-28893	22	30	,	,	PUNCT
fcis-28893	22	31	map	map	VERB
fcis-28893	22	32	the	the	DET
fcis-28893	22	33	extracted	extract	VERB
fcis-28893	22	34	global	global	ADJ
fcis-28893	22	35	features	feature	NOUN
fcis-28893	22	36	into	into	ADP
fcis-28893	22	37	a	a	DET
fcis-28893	22	38	common	common	ADJ
fcis-28893	22	39	feature	feature	NOUN
fcis-28893	22	40	space	space	NOUN
fcis-28893	22	41	,	,	PUNCT
fcis-28893	22	42	bring	bring	VERB
fcis-28893	22	43	similar	similar	ADJ
fcis-28893	22	44	features	feature	NOUN
fcis-28893	22	45	closer	close	ADJ
fcis-28893	22	46	,	,	PUNCT
fcis-28893	22	47	and	and	CCONJ
fcis-28893	22	48	push	push	VERB
fcis-28893	22	49	away	away	ADV
fcis-28893	22	50	the	the	DET
fcis-28893	22	51	features	feature	NOUN
fcis-28893	22	52	with	with	ADP
fcis-28893	22	53	larger	large	ADJ
fcis-28893	22	54	differences	difference	NOUN
fcis-28893	22	55	.	.	PUNCT
fcis-28893	23	1	cmpc+cmpm	cmpc+cmpm	PROPN
fcis-28893	23	2	uses	use	VERB
fcis-28893	23	3	resnet-50	resnet-50	PROPN
fcis-28893	23	4	and	and	CCONJ
fcis-28893	23	5	lstm	lstm	NOUN
fcis-28893	23	6	network	network	NOUN
fcis-28893	23	7	to	to	PART
fcis-28893	23	8	extract	extract	VERB
fcis-28893	23	9	image	image	NOUN
fcis-28893	23	10	features	feature	NOUN
fcis-28893	23	11	,	,	PUNCT
fcis-28893	23	12	and	and	CCONJ
fcis-28893	23	13	designs	design	VERB
fcis-28893	23	14	the	the	DET
fcis-28893	23	15	cmpm	cmpm	PROPN
fcis-28893	23	16	loss	loss	NOUN
fcis-28893	23	17	function	function	NOUN
fcis-28893	23	18	to	to	PART
fcis-28893	23	19	map	map	VERB
fcis-28893	23	20	the	the	DET
fcis-28893	23	21	cross	cross	ADJ
fcis-28893	23	22	-	-	ADJ
fcis-28893	23	23	modal	modal	ADJ
fcis-28893	23	24	features	feature	NOUN
fcis-28893	23	25	into	into	ADP
fcis-28893	23	26	the	the	DET
fcis-28893	23	27	kl	kl	PROPN
fcis-28893	23	28	scatter	scatter	NOUN
fcis-28893	23	29	chain	chain	NOUN
fcis-28893	23	30	to	to	PART
fcis-28893	23	31	bridge	bridge	VERB
fcis-28893	23	32	the	the	DET
fcis-28893	23	33	modal	modal	ADJ
fcis-28893	23	34	gap	gap	NOUN
fcis-28893	23	35	;	;	PUNCT
fcis-28893	23	36	in	in	ADP
fcis-28893	23	37	addition	addition	NOUN
fcis-28893	23	38	,	,	PUNCT
fcis-28893	23	39	cmpc	cmpc	PROPN
fcis-28893	23	40	projects	project	VERB
fcis-28893	23	41	the	the	DET
fcis-28893	23	42	features	feature	NOUN
fcis-28893	23	43	onto	onto	ADP
fcis-28893	23	44	the	the	DET
fcis-28893	23	45	other	other	ADJ
fcis-28893	23	46	modality	modality	NOUN
fcis-28893	23	47	to	to	PART
fcis-28893	23	48	strengthen	strengthen	VERB
fcis-28893	23	49	the	the	DET
fcis-28893	23	50	image	image	NOUN
fcis-28893	23	51	-	-	PUNCT
fcis-28893	23	52	text	text	NOUN
fcis-28893	23	53	pairs	pair	NOUN
fcis-28893	23	54	and	and	CCONJ
fcis-28893	23	55	the	the	DET
fcis-28893	23	56	cmpm	cmpm	PROPN
fcis-28893	23	57	loss	loss	NOUN
fcis-28893	23	58	function	function	NOUN
fcis-28893	23	59	map	map	VERB
fcis-28893	23	60	the	the	DET
fcis-28893	23	61	crossmodal	crossmodal	ADJ
fcis-28893	23	62	features	feature	NOUN
fcis-28893	23	63	into	into	ADP
fcis-28893	23	64	the	the	DET
fcis-28893	23	65	kl	kl	PROPN
fcis-28893	23	66	scatter	scatter	NOUN
fcis-28893	23	67	chain	chain	NOUN
fcis-28893	23	68	to	to	PART
fcis-28893	23	69	bridge	bridge	VERB
fcis-28893	23	70	the	the	DET
fcis-28893	23	71	modal	modal	ADJ
fcis-28893	23	72	gap	gap	NOUN
fcis-28893	23	73	;	;	PUNCT
fcis-28893	23	74	in	in	ADP
fcis-28893	23	75	addition	addition	NOUN
fcis-28893	23	76	,	,	PUNCT
fcis-28893	23	77	cmpc	cmpc	PROPN
fcis-28893	23	78	projects	project	VERB
fcis-28893	23	79	the	the	DET
fcis-28893	23	80	features	feature	NOUN
fcis-28893	23	81	onto	onto	ADP
fcis-28893	23	82	another	another	DET
fcis-28893	23	83	modality	modality	NOUN
fcis-28893	23	84	to	to	PART
fcis-28893	23	85	strengthen	strengthen	VERB
fcis-28893	23	86	the	the	DET
fcis-28893	23	87	intrinsic	intrinsic	ADJ
fcis-28893	23	88	connection	connection	NOUN
fcis-28893	23	89	between	between	ADP
fcis-28893	23	90	the	the	DET
fcis-28893	23	91	image	image	NOUN
fcis-28893	23	92	and	and	CCONJ
fcis-28893	23	93	text	text	NOUN
fcis-28893	23	94	pairs	pair	NOUN
fcis-28893	23	95	.	.	PUNCT
fcis-28893	24	1	2	2	X
fcis-28893	24	2	)	)	PUNCT
fcis-28893	24	3	methods	method	NOUN
fcis-28893	24	4	based	base	VERB
fcis-28893	24	5	on	on	ADP
fcis-28893	24	6	multi	multi	ADJ
fcis-28893	24	7	-	-	ADJ
fcis-28893	24	8	granularity	granularity	NOUN
fcis-28893	24	9	information	information	NOUN
fcis-28893	24	10	the	the	DET
fcis-28893	24	11	main	main	ADJ
fcis-28893	24	12	idea	idea	NOUN
fcis-28893	24	13	is	be	AUX
fcis-28893	24	14	to	to	PART
fcis-28893	24	15	slice	slice	VERB
fcis-28893	24	16	the	the	DET
fcis-28893	24	17	image	image	NOUN
fcis-28893	24	18	horizontally	horizontally	ADV
fcis-28893	24	19	divide	divide	VERB
fcis-28893	24	20	into	into	ADP
fcis-28893	24	21	fine	fine	ADV
fcis-28893	24	22	-	-	PUNCT
fcis-28893	24	23	grained	grain	VERB
fcis-28893	24	24	regions	region	NOUN
fcis-28893	24	25	such	such	ADJ
fcis-28893	24	26	as	as	ADP
fcis-28893	24	27	head	head	NOUN
fcis-28893	24	28	,	,	PUNCT
fcis-28893	24	29	upper	upper	ADJ
fcis-28893	24	30	body	body	NOUN
fcis-28893	24	31	,	,	PUNCT
fcis-28893	24	32	lower	low	ADJ
fcis-28893	24	33	body	body	NOUN
fcis-28893	24	34	,	,	PUNCT
fcis-28893	24	35	foot	foot	NOUN
fcis-28893	24	36	,	,	PUNCT
fcis-28893	24	37	hand	hand	NOUN
fcis-28893	24	38	,	,	PUNCT
fcis-28893	24	39	and	and	CCONJ
fcis-28893	24	40	so	so	ADV
fcis-28893	24	41	on	on	ADV
fcis-28893	24	42	,	,	PUNCT
fcis-28893	24	43	and	and	CCONJ
fcis-28893	24	44	then	then	ADV
fcis-28893	24	45	realize	realize	VERB
fcis-28893	24	46	the	the	DET
fcis-28893	24	47	level	level	NOUN
fcis-28893	24	48	-	-	PUNCT
fcis-28893	24	49	by	by	ADP
fcis-28893	24	50	-	-	PUNCT
fcis-28893	24	51	level	level	NOUN
fcis-28893	24	52	finegrained	finegraine	VERB
fcis-28893	24	53	alignment	alignment	NOUN
fcis-28893	24	54	of	of	ADP
fcis-28893	24	55	the	the	DET
fcis-28893	24	56	image	image	NOUN
fcis-28893	24	57	patch	patch	NOUN
fcis-28893	24	58	with	with	ADP
fcis-28893	24	59	words	word	NOUN
fcis-28893	24	60	,	,	PUNCT
fcis-28893	24	61	image	image	NOUN
fcis-28893	24	62	regions	region	NOUN
fcis-28893	24	63	with	with	ADP
fcis-28893	24	64	phrases	phrase	NOUN
fcis-28893	24	65	,	,	PUNCT
fcis-28893	24	66	and	and	CCONJ
fcis-28893	24	67	the	the	DET
fcis-28893	24	68	whole	whole	ADJ
fcis-28893	24	69	image	image	NOUN
fcis-28893	24	70	with	with	ADP
fcis-28893	24	71	text	text	NOUN
fcis-28893	24	72	sentences	sentence	NOUN
fcis-28893	24	73	,	,	PUNCT
fcis-28893	24	74	this	this	DET
fcis-28893	24	75	type	type	NOUN
fcis-28893	24	76	of	of	ADP
fcis-28893	24	77	method	method	NOUN
fcis-28893	24	78	can	can	AUX
fcis-28893	24	79	effectively	effectively	ADV
fcis-28893	24	80	improve	improve	VERB
fcis-28893	24	81	the	the	DET
fcis-28893	24	82	retrieval	retrieval	NOUN
fcis-28893	24	83	accuracy	accuracy	NOUN
fcis-28893	24	84	.	.	PUNCT
fcis-28893	25	1	for	for	ADP
fcis-28893	25	2	example	example	NOUN
fcis-28893	25	3	,	,	PUNCT
fcis-28893	25	4	vitaa	vitaa	PRON
fcis-28893	25	5	segments	segment	VERB
fcis-28893	25	6	the	the	DET
fcis-28893	25	7	image	image	NOUN
fcis-28893	25	8	according	accord	VERB
fcis-28893	25	9	to	to	ADP
fcis-28893	25	10	specific	specific	ADJ
fcis-28893	25	11	pedestrian	pedestrian	NOUN
fcis-28893	25	12	attributes	attribute	VERB
fcis-28893	25	13	through	through	ADP
fcis-28893	25	14	the	the	DET
fcis-28893	25	15	semantic	semantic	ADJ
fcis-28893	25	16	segmentation	segmentation	NOUN
fcis-28893	25	17	layer	layer	NOUN
fcis-28893	25	18	of	of	ADP
fcis-28893	25	19	the	the	DET
fcis-28893	25	20	lightweight	lightweight	ADJ
fcis-28893	25	21	maskhead	maskhead	ADJ
fcis-28893	25	22	architecture	architecture	NOUN
fcis-28893	25	23	to	to	PART
fcis-28893	25	24	form	form	VERB
fcis-28893	25	25	a	a	DET
fcis-28893	25	26	fine	fine	ADV
fcis-28893	25	27	-	-	PUNCT
fcis-28893	25	28	grained	grain	VERB
fcis-28893	25	29	segmentation	segmentation	NOUN
fcis-28893	25	30	map	map	NOUN
fcis-28893	25	31	;	;	PUNCT
fcis-28893	25	32	the	the	DET
fcis-28893	25	33	text	text	NOUN
fcis-28893	25	34	part	part	NOUN
fcis-28893	25	35	is	be	AUX
fcis-28893	25	36	parsed	parse	VERB
fcis-28893	25	37	and	and	CCONJ
fcis-28893	25	38	extracted	extract	VERB
fcis-28893	25	39	by	by	ADP
fcis-28893	25	40	stanford	stanford	PROPN
fcis-28893	25	41	pos	pos	PROPN
fcis-28893	25	42	tagger	tagger	NOUN
fcis-28893	25	43	to	to	PART
fcis-28893	25	44	extract	extract	VERB
fcis-28893	25	45	words	word	NOUN
fcis-28893	25	46	and	and	CCONJ
fcis-28893	25	47	phrases	phrase	NOUN
fcis-28893	25	48	related	relate	VERB
fcis-28893	25	49	to	to	ADP
fcis-28893	25	50	the	the	DET
fcis-28893	25	51	attributes	attribute	NOUN
fcis-28893	25	52	,	,	PUNCT
fcis-28893	25	53	and	and	CCONJ
fcis-28893	25	54	then	then	ADV
fcis-28893	25	55	lstm	lstm	NOUN
fcis-28893	25	56	is	be	AUX
fcis-28893	25	57	used	use	VERB
fcis-28893	25	58	to	to	PART
fcis-28893	25	59	extract	extract	VERB
fcis-28893	25	60	the	the	DET
fcis-28893	25	61	global	global	ADJ
fcis-28893	25	62	and	and	CCONJ
fcis-28893	25	63	local	local	ADJ
fcis-28893	25	64	textual	textual	ADJ
fcis-28893	25	65	features	feature	NOUN
fcis-28893	25	66	,	,	PUNCT
fcis-28893	25	67	and	and	CCONJ
fcis-28893	25	68	a	a	DET
fcis-28893	25	69	finegrained	finegraine	VERB
fcis-28893	25	70	segmentation	segmentation	NOUN
fcis-28893	25	71	map	map	NOUN
fcis-28893	25	72	is	be	AUX
fcis-28893	25	73	made	make	VERB
fcis-28893	25	74	by	by	ADP
fcis-28893	25	75	establishing	establish	VERB
fcis-28893	25	76	correspondences	correspondence	NOUN
fcis-28893	25	77	between	between	ADP
fcis-28893	25	78	words	word	NOUN
fcis-28893	25	79	,	,	PUNCT
fcis-28893	25	80	phrases	phrase	NOUN
fcis-28893	25	81	and	and	CCONJ
fcis-28893	25	82	segmentation	segmentation	NOUN
fcis-28893	25	83	maps	map	NOUN
fcis-28893	25	84	.	.	PUNCT
fcis-28893	26	1	the	the	DET
fcis-28893	26	2	fine	fine	ADV
fcis-28893	26	3	-	-	PUNCT
fcis-28893	26	4	grained	grain	VERB
fcis-28893	26	5	image	image	NOUN
fcis-28893	26	6	feature	feature	NOUN
fcis-28893	26	7	alignment	alignment	NOUN
fcis-28893	26	8	is	be	AUX
fcis-28893	26	9	performed	perform	VERB
fcis-28893	26	10	by	by	ADP
fcis-28893	26	11	establishing	establish	VERB
fcis-28893	26	12	the	the	DET
fcis-28893	26	13	correspondence	correspondence	NOUN
fcis-28893	26	14	between	between	ADP
fcis-28893	26	15	words	word	NOUN
fcis-28893	26	16	,	,	PUNCT
fcis-28893	26	17	phrases	phrase	NOUN
fcis-28893	26	18	and	and	CCONJ
fcis-28893	26	19	segmentation	segmentation	NOUN
fcis-28893	26	20	graphs	graph	NOUN
fcis-28893	26	21	.	.	PUNCT
fcis-28893	27	1	3	3	X
fcis-28893	27	2	)	)	PUNCT
fcis-28893	27	3	methods	method	NOUN
fcis-28893	27	4	based	base	VERB
fcis-28893	27	5	on	on	ADP
fcis-28893	27	6	adversarial	adversarial	ADJ
fcis-28893	27	7	learning	learning	NOUN
fcis-28893	27	8	since	since	SCONJ
fcis-28893	27	9	gan	gan	PROPN
fcis-28893	27	10	(	(	PUNCT
fcis-28893	27	11	generative	generative	ADJ
fcis-28893	27	12	adversarial	adversarial	ADJ
fcis-28893	27	13	network	network	NOUN
fcis-28893	27	14	)	)	PUNCT
fcis-28893	27	15	was	be	AUX
fcis-28893	27	16	proposed	propose	VERB
fcis-28893	27	17	,	,	PUNCT
fcis-28893	27	18	it	it	PRON
fcis-28893	27	19	has	have	AUX
fcis-28893	27	20	received	receive	VERB
fcis-28893	27	21	a	a	DET
fcis-28893	27	22	lot	lot	NOUN
fcis-28893	27	23	of	of	ADP
fcis-28893	27	24	attention	attention	NOUN
fcis-28893	27	25	.	.	PUNCT
fcis-28893	28	1	researchers	researcher	NOUN
fcis-28893	28	2	in	in	ADP
fcis-28893	28	3	the	the	DET
fcis-28893	28	4	field	field	NOUN
fcis-28893	28	5	of	of	ADP
fcis-28893	28	6	text	text	NOUN
fcis-28893	28	7	-	-	PUNCT
fcis-28893	28	8	based	base	VERB
fcis-28893	28	9	pedestrian	pedestrian	NOUN
fcis-28893	28	10	retrieval	retrieval	NOUN
fcis-28893	28	11	have	have	AUX
fcis-28893	28	12	also	also	ADV
fcis-28893	28	13	borrowed	borrow	VERB
fcis-28893	28	14	the	the	DET
fcis-28893	28	15	adversarial	adversarial	ADJ
fcis-28893	28	16	idea	idea	NOUN
fcis-28893	28	17	of	of	ADP
fcis-28893	28	18	gan	gan	PROPN
fcis-28893	28	19	and	and	CCONJ
fcis-28893	28	20	designed	design	VERB
fcis-28893	28	21	models	model	NOUN
fcis-28893	28	22	such	such	ADJ
fcis-28893	28	23	as	as	ADP
fcis-28893	28	24	aganet	aganet	NOUN
fcis-28893	28	25	and	and	CCONJ
fcis-28893	28	26	timam	timam	NOUN
fcis-28893	28	27	.	.	PUNCT
fcis-28893	29	1	the	the	DET
fcis-28893	29	2	core	core	ADJ
fcis-28893	29	3	idea	idea	NOUN
fcis-28893	29	4	of	of	ADP
fcis-28893	29	5	this	this	DET
fcis-28893	29	6	class	class	NOUN
fcis-28893	29	7	of	of	ADP
fcis-28893	29	8	methods	method	NOUN
fcis-28893	29	9	is	be	AUX
fcis-28893	29	10	to	to	PART
fcis-28893	29	11	use	use	VERB
fcis-28893	29	12	generators	generator	NOUN
fcis-28893	29	13	to	to	PART
fcis-28893	29	14	generate	generate	VERB
fcis-28893	29	15	features	feature	NOUN
fcis-28893	29	16	that	that	PRON
fcis-28893	29	17	are	be	AUX
fcis-28893	29	18	as	as	ADV
fcis-28893	29	19	simulated	simulate	VERB
fcis-28893	29	20	as	as	ADP
fcis-28893	29	21	possible	possible	ADJ
fcis-28893	29	22	,	,	PUNCT
fcis-28893	29	23	while	while	SCONJ
fcis-28893	29	24	the	the	DET
fcis-28893	29	25	discriminator	discriminator	NOUN
fcis-28893	29	26	determines	determine	VERB
fcis-28893	29	27	the	the	DET
fcis-28893	29	28	modal	modal	ADJ
fcis-28893	29	29	class	class	NOUN
fcis-28893	29	30	of	of	ADP
fcis-28893	29	31	the	the	DET
fcis-28893	29	32	input	input	NOUN
fcis-28893	29	33	features	feature	NOUN
fcis-28893	29	34	and	and	CCONJ
fcis-28893	29	35	identifies	identify	VERB
fcis-28893	29	36	whether	whether	SCONJ
fcis-28893	29	37	they	they	PRON
fcis-28893	29	38	are	be	AUX
fcis-28893	29	39	generated	generate	VERB
fcis-28893	29	40	by	by	ADP
fcis-28893	29	41	the	the	DET
fcis-28893	29	42	generator	generator	NOUN
fcis-28893	29	43	,	,	PUNCT
fcis-28893	29	44	with	with	ADP
fcis-28893	29	45	the	the	DET
fcis-28893	29	46	main	main	ADJ
fcis-28893	29	47	advantage	advantage	NOUN
fcis-28893	29	48	that	that	PRON
fcis-28893	29	49	the	the	DET
fcis-28893	29	50	model	model	NOUN
fcis-28893	29	51	is	be	AUX
fcis-28893	29	52	simple	simple	ADJ
fcis-28893	29	53	in	in	ADP
fcis-28893	29	54	structure	structure	NOUN
fcis-28893	29	55	and	and	CCONJ
fcis-28893	29	56	does	do	AUX
fcis-28893	29	57	not	not	PART
fcis-28893	29	58	require	require	VERB
fcis-28893	29	59	additional	additional	ADJ
fcis-28893	29	60	modules	module	NOUN
fcis-28893	29	61	or	or	CCONJ
fcis-28893	29	62	plug	plug	NOUN
fcis-28893	29	63	-	-	PUNCT
fcis-28893	29	64	ins	in	NOUN
fcis-28893	29	65	.	.	NOUN
fcis-28893	29	66	4	4	X
fcis-28893	29	67	)	)	PUNCT
fcis-28893	29	68	methods	method	NOUN
fcis-28893	29	69	based	base	VERB
fcis-28893	29	70	on	on	ADP
fcis-28893	29	71	cross	cross	ADJ
fcis-28893	29	72	-	-	ADJ
fcis-28893	29	73	modal	modal	ADJ
fcis-28893	29	74	attention	attention	NOUN
fcis-28893	29	75	mechanisms	mechanism	NOUN
fcis-28893	29	76	representative	representative	NOUN
fcis-28893	29	77	models	model	NOUN
fcis-28893	29	78	include	include	VERB
fcis-28893	29	79	dssl	dssl	PROPN
fcis-28893	29	80	,	,	PUNCT
fcis-28893	29	81	nafs	nafs	PROPN
fcis-28893	29	82	,	,	PUNCT
fcis-28893	29	83	saf	saf	PROPN
fcis-28893	29	84	,	,	PUNCT
fcis-28893	29	85	acsa	acsa	NOUN
fcis-28893	29	86	,	,	PUNCT
fcis-28893	29	87	etc	etc	X
fcis-28893	29	88	.	.	X
fcis-28893	30	1	the	the	DET
fcis-28893	30	2	core	core	ADJ
fcis-28893	30	3	idea	idea	NOUN
fcis-28893	30	4	of	of	ADP
fcis-28893	30	5	such	such	ADJ
fcis-28893	30	6	methods	method	NOUN
fcis-28893	30	7	is	be	AUX
fcis-28893	30	8	to	to	PART
fcis-28893	30	9	pay	pay	VERB
fcis-28893	30	10	attention	attention	NOUN
fcis-28893	30	11	to	to	ADP
fcis-28893	30	12	each	each	DET
fcis-28893	30	13	pixel	pixel	NOUN
fcis-28893	30	14	in	in	ADP
fcis-28893	30	15	the	the	DET
fcis-28893	30	16	feature	feature	NOUN
fcis-28893	30	17	image	image	NOUN
fcis-28893	30	18	through	through	ADP
fcis-28893	30	19	the	the	DET
fcis-28893	30	20	multi	multi	ADJ
fcis-28893	30	21	-	-	ADJ
fcis-28893	30	22	head	head	ADJ
fcis-28893	30	23	selfattention	selfattention	NOUN
fcis-28893	30	24	mechanism	mechanism	NOUN
fcis-28893	30	25	,	,	PUNCT
fcis-28893	30	26	and	and	CCONJ
fcis-28893	30	27	focus	focus	VERB
fcis-28893	30	28	the	the	DET
fcis-28893	30	29	attention	attention	NOUN
fcis-28893	30	30	on	on	ADP
fcis-28893	30	31	more	more	ADV
fcis-28893	30	32	important	important	ADJ
fcis-28893	30	33	regions	region	NOUN
fcis-28893	30	34	,	,	PUNCT
fcis-28893	30	35	extract	extract	VERB
fcis-28893	30	36	more	more	ADJ
fcis-28893	30	37	fine	fine	ADV
fcis-28893	30	38	-	-	PUNCT
fcis-28893	30	39	grained	grain	VERB
fcis-28893	30	40	features	feature	NOUN
fcis-28893	30	41	,	,	PUNCT
fcis-28893	30	42	and	and	CCONJ
fcis-28893	30	43	calculate	calculate	VERB
fcis-28893	30	44	the	the	DET
fcis-28893	30	45	similarity	similarity	NOUN
fcis-28893	30	46	between	between	ADP
fcis-28893	30	47	each	each	DET
fcis-28893	30	48	part	part	NOUN
fcis-28893	30	49	of	of	ADP
fcis-28893	30	50	the	the	DET
fcis-28893	30	51	image	image	NOUN
fcis-28893	30	52	and	and	CCONJ
fcis-28893	30	53	the	the	DET
fcis-28893	30	54	text	text	NOUN
fcis-28893	30	55	.	.	PUNCT
fcis-28893	31	1	therefore	therefore	ADV
fcis-28893	31	2	,	,	PUNCT
fcis-28893	31	3	the	the	DET
fcis-28893	31	4	attention	attention	NOUN
fcis-28893	31	5	mechanism	mechanism	NOUN
fcis-28893	31	6	approach	approach	NOUN
fcis-28893	31	7	can	can	AUX
fcis-28893	31	8	further	far	ADV
fcis-28893	31	9	improve	improve	VERB
fcis-28893	31	10	the	the	DET
fcis-28893	31	11	retrieval	retrieval	ADJ
fcis-28893	31	12	accuracy	accuracy	NOUN
fcis-28893	31	13	.	.	PUNCT
fcis-28893	32	1	just	just	ADV
fcis-28893	32	2	as	as	SCONJ
fcis-28893	32	3	the	the	DET
fcis-28893	32	4	acsa	acsa	NOUN
fcis-28893	32	5	approach	approach	NOUN
fcis-28893	32	6	uses	use	VERB
fcis-28893	32	7	swin	swin	PROPN
fcis-28893	32	8	transformer	transformer	PROPN
fcis-28893	32	9	and	and	CCONJ
fcis-28893	32	10	bert	bert	NOUN
fcis-28893	32	11	to	to	PART
fcis-28893	32	12	extract	extract	VERB
fcis-28893	32	13	image	image	NOUN
fcis-28893	32	14	features	feature	NOUN
fcis-28893	32	15	separately	separately	ADV
fcis-28893	32	16	,	,	PUNCT
fcis-28893	32	17	swin	swin	PROPN
fcis-28893	32	18	transformer	transformer	PROPN
fcis-28893	32	19	hierarchically	hierarchically	ADV
fcis-28893	32	20	focuses	focus	VERB
fcis-28893	32	21	on	on	ADP
fcis-28893	32	22	each	each	DET
fcis-28893	32	23	pixel	pixel	NOUN
fcis-28893	32	24	point	point	NOUN
fcis-28893	32	25	in	in	ADP
fcis-28893	32	26	the	the	DET
fcis-28893	32	27	image	image	NOUN
fcis-28893	32	28	and	and	CCONJ
fcis-28893	32	29	understands	understand	VERB
fcis-28893	32	30	the	the	DET
fcis-28893	32	31	textual	textual	ADJ
fcis-28893	32	32	contextual	contextual	ADJ
fcis-28893	32	33	semantic	semantic	ADJ
fcis-28893	32	34	information	information	NOUN
fcis-28893	32	35	through	through	ADP
fcis-28893	32	36	bert	bert	PROPN
fcis-28893	32	37	to	to	PART
fcis-28893	32	38	perform	perform	VERB
fcis-28893	32	39	deeply	deeply	ADV
fcis-28893	32	40	alignment	alignment	NOUN
fcis-28893	32	41	.	.	PUNCT
fcis-28893	33	1	however	however	ADV
fcis-28893	33	2	,	,	PUNCT
fcis-28893	33	3	since	since	SCONJ
fcis-28893	33	4	each	each	DET
fcis-28893	33	5	pixel	pixel	PROPN
fcis-28893	33	6	point	point	NOUN
fcis-28893	33	7	is	be	AUX
fcis-28893	33	8	involved	involve	VERB
fcis-28893	33	9	in	in	ADP
fcis-28893	33	10	the	the	DET
fcis-28893	33	11	computation	computation	NOUN
fcis-28893	33	12	,	,	PUNCT
fcis-28893	33	13	such	such	ADJ
fcis-28893	33	14	methods	method	NOUN
fcis-28893	33	15	have	have	VERB
fcis-28893	33	16	higher	high	ADJ
fcis-28893	33	17	computational	computational	ADJ
fcis-28893	33	18	and	and	CCONJ
fcis-28893	33	19	slower	slow	ADJ
fcis-28893	33	20	retrieval	retrieval	NOUN
fcis-28893	33	21	speed	speed	NOUN
fcis-28893	33	22	.	.	PUNCT
fcis-28893	34	1	5	5	NUM
fcis-28893	34	2	)	)	PUNCT
fcis-28893	34	3	methods	method	NOUN
fcis-28893	34	4	based	base	VERB
fcis-28893	34	5	on	on	ADP
fcis-28893	34	6	visual	visual	ADJ
fcis-28893	34	7	text	text	NOUN
fcis-28893	34	8	pre	pre	ADJ
fcis-28893	34	9	-	-	ADJ
fcis-28893	34	10	training	training	ADJ
fcis-28893	34	11	models	model	NOUN
fcis-28893	34	12	the	the	DET
fcis-28893	34	13	main	main	ADJ
fcis-28893	34	14	idea	idea	NOUN
fcis-28893	34	15	of	of	ADP
fcis-28893	34	16	such	such	ADJ
fcis-28893	34	17	methods	method	NOUN
fcis-28893	34	18	is	be	AUX
fcis-28893	34	19	to	to	PART
fcis-28893	34	20	utilize	utilize	VERB
fcis-28893	34	21	pre	pre	ADJ
fcis-28893	34	22	-	-	ADJ
fcis-28893	34	23	trained	train	VERB
fcis-28893	34	24	vlp	vlp	NOUN
fcis-28893	34	25	models	model	NOUN
fcis-28893	34	26	,	,	PUNCT
fcis-28893	34	27	such	such	ADJ
fcis-28893	34	28	as	as	ADP
fcis-28893	34	29	the	the	DET
fcis-28893	34	30	irra	irra	ADJ
fcis-28893	34	31	model	model	NOUN
fcis-28893	34	32	using	use	VERB
fcis-28893	34	33	clip	clip	NOUN
fcis-28893	34	34	as	as	ADP
fcis-28893	34	35	the	the	DET
fcis-28893	34	36	backbone	backbone	NOUN
fcis-28893	34	37	network	network	NOUN
fcis-28893	34	38	,	,	PUNCT
fcis-28893	34	39	to	to	PART
fcis-28893	34	40	achieve	achieve	VERB
fcis-28893	34	41	efficient	efficient	ADJ
fcis-28893	34	42	alignment	alignment	NOUN
fcis-28893	34	43	of	of	ADP
fcis-28893	34	44	image	image	NOUN
fcis-28893	34	45	features	feature	VERB
fcis-28893	34	46	with	with	ADP
fcis-28893	34	47	the	the	DET
fcis-28893	34	48	help	help	NOUN
fcis-28893	34	49	of	of	ADP
fcis-28893	34	50	the	the	DET
fcis-28893	34	51	vlp	vlp	PROPN
fcis-28893	34	52	model	model	NOUN
fcis-28893	34	53	's	's	PART
fcis-28893	34	54	strong	strong	ADJ
fcis-28893	34	55	cross	cross	ADJ
fcis-28893	34	56	-	-	ADJ
fcis-28893	34	57	modal	modal	ADJ
fcis-28893	34	58	learning	learning	NOUN
fcis-28893	34	59	ability	ability	NOUN
fcis-28893	34	60	,	,	PUNCT
fcis-28893	34	61	so	so	SCONJ
fcis-28893	34	62	that	that	SCONJ
fcis-28893	34	63	the	the	DET
fcis-28893	34	64	model	model	NOUN
fcis-28893	34	65	understands	understand	VERB
fcis-28893	34	66	in	in	ADP
fcis-28893	34	67	depth	depth	NOUN
fcis-28893	34	68	that	that	SCONJ
fcis-28893	34	69	the	the	DET
fcis-28893	34	70	image	image	NOUN
fcis-28893	34	71	and	and	CCONJ
fcis-28893	34	72	text	text	NOUN
fcis-28893	34	73	pairs	pair	NOUN
fcis-28893	34	74	contain	contain	VERB
fcis-28893	34	75	semantic	semantic	ADJ
fcis-28893	34	76	information	information	NOUN
fcis-28893	34	77	,	,	PUNCT
fcis-28893	34	78	which	which	PRON
fcis-28893	34	79	makes	make	VERB
fcis-28893	34	80	it	it	PRON
fcis-28893	34	81	easy	easy	ADJ
fcis-28893	34	82	for	for	SCONJ
fcis-28893	34	83	the	the	DET
fcis-28893	34	84	model	model	NOUN
fcis-28893	34	85	to	to	PART
fcis-28893	34	86	make	make	VERB
fcis-28893	34	87	the	the	DET
fcis-28893	34	88	correct	correct	ADJ
fcis-28893	34	89	prediction	prediction	NOUN
fcis-28893	34	90	in	in	ADP
fcis-28893	34	91	the	the	DET
fcis-28893	34	92	retrieval	retrieval	NOUN
fcis-28893	34	93	.	.	PUNCT
fcis-28893	35	1	carrying	carry	VERB
fcis-28893	35	2	out	out	ADP
fcis-28893	35	3	such	such	ADJ
fcis-28893	35	4	methods	method	NOUN
fcis-28893	35	5	to	to	PART
fcis-28893	35	6	become	become	VERB
fcis-28893	35	7	the	the	DET
fcis-28893	35	8	mainstream	mainstream	NOUN
fcis-28893	35	9	methods	method	NOUN
fcis-28893	35	10	in	in	ADP
fcis-28893	35	11	the	the	DET
fcis-28893	35	12	field	field	NOUN
fcis-28893	35	13	,	,	PUNCT
fcis-28893	35	14	significantly	significantly	ADV
fcis-28893	35	15	improving	improve	VERB
fcis-28893	35	16	the	the	DET
fcis-28893	35	17	retrieval	retrieval	NOUN
fcis-28893	35	18	accuracy	accuracy	NOUN
fcis-28893	35	19	in	in	ADP
fcis-28893	35	20	the	the	DET
fcis-28893	35	21	field	field	NOUN
fcis-28893	35	22	.	.	PUNCT
fcis-28893	36	1	3	3	X
fcis-28893	36	2	.	.	X
fcis-28893	36	3	dataset	dataset	NOUN
fcis-28893	36	4	and	and	CCONJ
fcis-28893	36	5	evaluation	evaluation	NOUN
fcis-28893	36	6	metrics	metric	NOUN
fcis-28893	36	7	3.1	3.1	NUM
fcis-28893	36	8	.	.	PUNCT
fcis-28893	37	1	dataset	dataset	NOUN
fcis-28893	37	2	text	text	NOUN
fcis-28893	37	3	-	-	PUNCT
fcis-28893	37	4	based	base	VERB
fcis-28893	37	5	pedestrian	pedestrian	NOUN
fcis-28893	37	6	retrieval	retrieval	NOUN
fcis-28893	37	7	is	be	AUX
fcis-28893	37	8	an	an	DET
fcis-28893	37	9	emerging	emerge	VERB
fcis-28893	37	10	research	research	NOUN
fcis-28893	37	11	problem	problem	NOUN
fcis-28893	37	12	,	,	PUNCT
fcis-28893	37	13	and	and	CCONJ
fcis-28893	37	14	the	the	DET
fcis-28893	37	15	commonly	commonly	ADV
fcis-28893	37	16	used	use	VERB
fcis-28893	37	17	datasets	dataset	NOUN
fcis-28893	37	18	in	in	ADP
fcis-28893	37	19	this	this	DET
fcis-28893	37	20	area	area	NOUN
fcis-28893	37	21	are	be	AUX
fcis-28893	37	22	relatively	relatively	ADV
fcis-28893	37	23	limited	limited	ADJ
fcis-28893	37	24	,	,	PUNCT
fcis-28893	37	25	mainly	mainly	ADV
fcis-28893	37	26	including	include	VERB
fcis-28893	37	27	cuhk	cuhk	NOUN
fcis-28893	37	28	-	-	PUNCT
fcis-28893	37	29	pedes	pes	NOUN
fcis-28893	37	30	dataset[1	dataset[1	NUM
fcis-28893	37	31	]	]	PUNCT
fcis-28893	37	32	,	,	PUNCT
fcis-28893	37	33	icfg	icfg	NOUN
fcis-28893	37	34	-	-	ADJ
fcis-28893	37	35	pedes	pes	NOUN
fcis-28893	37	36	dataset[2	dataset[2	NOUN
fcis-28893	37	37	]	]	PUNCT
fcis-28893	37	38	and	and	CCONJ
fcis-28893	37	39	rstpreid	rstpreid	PROPN
fcis-28893	37	40	dataset	dataset	VERB
fcis-28893	37	41	[	[	X
fcis-28893	37	42	3	3	NUM
fcis-28893	37	43	]	]	PUNCT
fcis-28893	37	44	.	.	PUNCT
fcis-28893	38	1	table	table	NOUN
fcis-28893	38	2	2	2	NUM
fcis-28893	38	3	details	detail	NOUN
fcis-28893	38	4	the	the	DET
fcis-28893	38	5	information	information	NOUN
fcis-28893	38	6	about	about	ADP
fcis-28893	38	7	these	these	DET
fcis-28893	38	8	three	three	NUM
fcis-28893	38	9	datasets	dataset	NOUN
fcis-28893	38	10	.	.	PUNCT
fcis-28893	39	1	table	table	NOUN
fcis-28893	39	2	2	2	NUM
fcis-28893	39	3	.	.	PUNCT
fcis-28893	39	4	common	common	ADJ
fcis-28893	39	5	datasets	dataset	NOUN
fcis-28893	39	6	for	for	ADP
fcis-28893	39	7	text	text	NOUN
fcis-28893	39	8	-	-	PUNCT
fcis-28893	39	9	based	base	VERB
fcis-28893	39	10	pedestrian	pedestrian	NOUN
fcis-28893	39	11	retrieval	retrieval	NOUN
fcis-28893	39	12	dataset	dataset	NOUN
fcis-28893	39	13	number	number	NOUN
fcis-28893	39	14	of	of	ADP
fcis-28893	39	15	pedestrians	pedestrian	NOUN
fcis-28893	39	16	number	number	NOUN
fcis-28893	39	17	of	of	ADP
fcis-28893	39	18	images	image	NOUN
fcis-28893	39	19	number	number	NOUN
fcis-28893	39	20	of	of	ADP
fcis-28893	39	21	texts	text	NOUN
fcis-28893	39	22	average	average	ADJ
fcis-28893	39	23	length	length	NOUN
fcis-28893	39	24	of	of	ADP
fcis-28893	39	25	text	text	NOUN
fcis-28893	39	26	chuk	chuk	PROPN
fcis-28893	39	27	-	-	PUNCT
fcis-28893	39	28	pedes	pes	NOUN
fcis-28893	39	29	13003	13003	NUM
fcis-28893	39	30	40206	40206	NUM
fcis-28893	39	31	80440	80440	NUM
fcis-28893	39	32	23.5	23.5	NUM
fcis-28893	39	33	igfg	igfg	NOUN
fcis-28893	39	34	-	-	PUNCT
fcis-28893	39	35	pedes	pes	NOUN
fcis-28893	39	36	4102	4102	NUM
fcis-28893	39	37	54522	54522	NUM
fcis-28893	39	38	54522	54522	NUM
fcis-28893	39	39	37.2	37.2	NUM
fcis-28893	39	40	rstpreid	rstpreid	NOUN
fcis-28893	39	41	4101	4101	NUM
fcis-28893	39	42	20505	20505	NUM
fcis-28893	39	43	41010	41010	NUM
fcis-28893	39	44	23	23	NUM
fcis-28893	39	45	3.2	3.2	NUM
fcis-28893	39	46	.	.	PUNCT
fcis-28893	40	1	evaluation	evaluation	NOUN
fcis-28893	40	2	metrics	metric	NOUN
fcis-28893	40	3	3.2.1	3.2.1	NUM
fcis-28893	40	4	.	.	PUNCT
fcis-28893	41	1	tpr	tpr	PROPN
fcis-28893	41	2	true	true	ADJ
fcis-28893	41	3	positive	positive	ADJ
fcis-28893	41	4	rate	rate	NOUN
fcis-28893	41	5	(	(	PUNCT
fcis-28893	41	6	tpr	tpr	NOUN
fcis-28893	41	7	)	)	PUNCT
fcis-28893	41	8	or	or	CCONJ
fcis-28893	41	9	recall	recall	NOUN
fcis-28893	41	10	indicates	indicate	VERB
fcis-28893	41	11	the	the	DET
fcis-28893	41	12	proportion	proportion	NOUN
fcis-28893	41	13	of	of	ADP
fcis-28893	41	14	positive	positive	ADJ
fcis-28893	41	15	samples	sample	NOUN
fcis-28893	41	16	to	to	ADP
fcis-28893	41	17	the	the	DET
fcis-28893	41	18	total	total	ADJ
fcis-28893	41	19	positive	positive	ADJ
fcis-28893	41	20	samples	sample	NOUN
fcis-28893	41	21	in	in	ADP
fcis-28893	41	22	the	the	DET
fcis-28893	41	23	retrieval	retrieval	NOUN
fcis-28893	41	24	results	result	NOUN
fcis-28893	41	25	,	,	PUNCT
fcis-28893	41	26	reflecting	reflect	VERB
fcis-28893	41	27	the	the	DET
fcis-28893	41	28	number	number	NOUN
fcis-28893	41	29	of	of	ADP
fcis-28893	41	30	samples	sample	NOUN
fcis-28893	41	31	that	that	PRON
fcis-28893	41	32	can	can	AUX
fcis-28893	41	33	be	be	AUX
fcis-28893	41	34	successfully	successfully	ADV
fcis-28893	41	35	retrieved	retrieve	VERB
fcis-28893	41	36	by	by	ADP
fcis-28893	41	37	the	the	DET
fcis-28893	41	38	model	model	NOUN
fcis-28893	41	39	among	among	ADP
fcis-28893	41	40	all	all	DET
fcis-28893	41	41	the	the	DET
fcis-28893	41	42	positive	positive	ADJ
fcis-28893	41	43	samples	sample	NOUN
fcis-28893	41	44	.	.	PUNCT
fcis-28893	42	1	the	the	DET
fcis-28893	42	2	trp	trp	PROPN
fcis-28893	42	3	value	value	NOUN
fcis-28893	42	4	can	can	AUX
fcis-28893	42	5	be	be	AUX
fcis-28893	42	6	calculated	calculate	VERB
fcis-28893	42	7	by	by	ADP
fcis-28893	42	8	using	use	VERB
fcis-28893	42	9	equation	equation	NOUN
fcis-28893	42	10	(	(	PUNCT
fcis-28893	42	11	1	1	NUM
fcis-28893	42	12	)	)	PUNCT
fcis-28893	42	13	.	.	PUNCT
fcis-28893	43	1	tp	tp	ADP
fcis-28893	43	2	tpr	tpr	PROPN
fcis-28893	43	3	tp	tp	PROPN
fcis-28893	43	4	fn	fn	PROPN
fcis-28893	44	1			PROPN
fcis-28893	44	2			PUNCT
fcis-28893	44	3	(	(	PUNCT
fcis-28893	44	4	1	1	X
fcis-28893	44	5	)	)	PUNCT
fcis-28893	44	6	where	where	SCONJ
fcis-28893	44	7	tp	tp	NOUN
fcis-28893	44	8	is	be	AUX
fcis-28893	44	9	the	the	DET
fcis-28893	44	10	sample	sample	NOUN
fcis-28893	44	11	that	that	PRON
fcis-28893	44	12	is	be	AUX
fcis-28893	44	13	positive	positive	ADJ
fcis-28893	44	14	and	and	CCONJ
fcis-28893	44	15	is	be	AUX
fcis-28893	44	16	predicted	predict	VERB
fcis-28893	44	17	to	to	PART
fcis-28893	44	18	be	be	AUX
fcis-28893	44	19	positive	positive	ADJ
fcis-28893	44	20	,	,	PUNCT
fcis-28893	44	21	and	and	CCONJ
fcis-28893	44	22	fn	fn	NOUN
fcis-28893	44	23	represents	represent	VERB
fcis-28893	44	24	the	the	DET
fcis-28893	44	25	sample	sample	NOUN
fcis-28893	44	26	that	that	PRON
fcis-28893	44	27	is	be	AUX
fcis-28893	44	28	negative	negative	ADJ
fcis-28893	44	29	but	but	CCONJ
fcis-28893	44	30	is	be	AUX
fcis-28893	44	31	incorrectly	incorrectly	ADV
fcis-28893	44	32	predicted	predict	VERB
fcis-28893	44	33	to	to	PART
fcis-28893	44	34	be	be	AUX
fcis-28893	44	35	negative	negative	ADJ
fcis-28893	44	36	by	by	ADP
fcis-28893	44	37	the	the	DET
fcis-28893	44	38	model	model	NOUN
fcis-28893	44	39	.	.	PUNCT
fcis-28893	45	1	3.2.2	3.2.2	X
fcis-28893	45	2	.	.	X
fcis-28893	45	3	map	map	VERB
fcis-28893	45	4	the	the	DET
fcis-28893	45	5	mean	mean	ADJ
fcis-28893	45	6	accuracy	accuracy	NOUN
fcis-28893	45	7	value	value	NOUN
fcis-28893	45	8	map	map	NOUN
fcis-28893	45	9	is	be	AUX
fcis-28893	45	10	the	the	DET
fcis-28893	45	11	average	average	NOUN
fcis-28893	45	12	of	of	ADP
fcis-28893	45	13	the	the	DET
fcis-28893	45	14	maximum	maximum	ADJ
fcis-28893	45	15	accuracy	accuracy	NOUN
fcis-28893	45	16	at	at	ADP
fcis-28893	45	17	different	different	ADJ
fcis-28893	45	18	recall	recall	NOUN
fcis-28893	45	19	rates	rate	NOUN
fcis-28893	45	20	and	and	CCONJ
fcis-28893	45	21	is	be	AUX
fcis-28893	45	22	commonly	commonly	ADV
fcis-28893	45	23	120	120	NUM
fcis-28893	45	24	used	use	VERB
fcis-28893	45	25	to	to	PART
fcis-28893	45	26	measure	measure	VERB
fcis-28893	45	27	the	the	DET
fcis-28893	45	28	performance	performance	NOUN
fcis-28893	45	29	of	of	ADP
fcis-28893	45	30	the	the	DET
fcis-28893	45	31	target	target	NOUN
fcis-28893	45	32	detector	detector	NOUN
fcis-28893	45	33	.	.	PUNCT
fcis-28893	46	1	map	map	NOUN
fcis-28893	46	2	value	value	NOUN
fcis-28893	46	3	can	can	AUX
fcis-28893	46	4	be	be	AUX
fcis-28893	46	5	calculated	calculate	VERB
fcis-28893	46	6	by	by	ADP
fcis-28893	46	7	using	use	VERB
fcis-28893	46	8	equation	equation	NOUN
fcis-28893	46	9	(	(	PUNCT
fcis-28893	46	10	2	2	NUM
fcis-28893	46	11	)	)	PUNCT
fcis-28893	46	12	.	.	PUNCT
fcis-28893	47	1	1	1	NUM
fcis-28893	47	2	1	1	NUM
fcis-28893	47	3	s	s	NOUN
fcis-28893	47	4	s	s	X
fcis-28893	47	5	i	i	PRON
fcis-28893	47	6	map	map	VERB
fcis-28893	47	7	ap	ap	PROPN
fcis-28893	47	8	s	s	PROPN
fcis-28893	47	9			PROPN
fcis-28893	47	10			ADJ
fcis-28893	47	11			X
fcis-28893	47	12	(	(	PUNCT
fcis-28893	47	13	2	2	X
fcis-28893	47	14	)	)	PUNCT
fcis-28893	47	15	where	where	SCONJ
fcis-28893	47	16	s	s	NOUN
fcis-28893	47	17	represents	represent	VERB
fcis-28893	47	18	the	the	DET
fcis-28893	47	19	total	total	ADJ
fcis-28893	47	20	number	number	NOUN
fcis-28893	47	21	of	of	ADP
fcis-28893	47	22	samples	sample	NOUN
fcis-28893	47	23	and	and	CCONJ
fcis-28893	47	24	aps	ap	NOUN
fcis-28893	47	25	represents	represent	VERB
fcis-28893	47	26	the	the	DET
fcis-28893	47	27	average	average	ADJ
fcis-28893	47	28	accuracy	accuracy	NOUN
fcis-28893	47	29	of	of	ADP
fcis-28893	47	30	the	the	DET
fcis-28893	47	31	prediction	prediction	NOUN
fcis-28893	47	32	for	for	ADP
fcis-28893	47	33	the	the	DET
fcis-28893	47	34	sample	sample	NOUN
fcis-28893	47	35	set	set	VERB
fcis-28893	47	36	s.	s.	PROPN
fcis-28893	47	37	4	4	NUM
fcis-28893	47	38	.	.	PUNCT
fcis-28893	48	1	methods	method	NOUN
fcis-28893	48	2	of	of	ADP
fcis-28893	48	3	text	text	NOUN
fcis-28893	48	4	-	-	PUNCT
fcis-28893	48	5	based	base	VERB
fcis-28893	48	6	pedestrian	pedestrian	NOUN
fcis-28893	48	7	retrieval	retrieval	NOUN
fcis-28893	48	8	text	text	NOUN
fcis-28893	48	9	-	-	PUNCT
fcis-28893	48	10	based	base	VERB
fcis-28893	48	11	pedestrian	pedestrian	NOUN
fcis-28893	48	12	retrieval	retrieval	NOUN
fcis-28893	48	13	aims	aim	VERB
fcis-28893	48	14	to	to	PART
fcis-28893	48	15	retrieve	retrieve	VERB
fcis-28893	48	16	images	image	NOUN
fcis-28893	48	17	of	of	ADP
fcis-28893	48	18	specific	specific	ADJ
fcis-28893	48	19	pedestrians	pedestrian	NOUN
fcis-28893	48	20	through	through	ADP
fcis-28893	48	21	textual	textual	ADJ
fcis-28893	48	22	descriptions	description	NOUN
fcis-28893	48	23	,	,	PUNCT
fcis-28893	48	24	and	and	CCONJ
fcis-28893	48	25	its	its	PRON
fcis-28893	48	26	core	core	NOUN
fcis-28893	48	27	challenge	challenge	NOUN
fcis-28893	48	28	lies	lie	VERB
fcis-28893	48	29	in	in	ADP
fcis-28893	48	30	bridging	bridge	VERB
fcis-28893	48	31	the	the	DET
fcis-28893	48	32	inherent	inherent	ADJ
fcis-28893	48	33	gap	gap	NOUN
fcis-28893	48	34	between	between	ADP
fcis-28893	48	35	the	the	DET
fcis-28893	48	36	two	two	NUM
fcis-28893	48	37	modalities	modality	NOUN
fcis-28893	48	38	and	and	CCONJ
fcis-28893	48	39	realizing	realize	VERB
fcis-28893	48	40	cross	cross	ADJ
fcis-28893	48	41	-	-	ADJ
fcis-28893	48	42	modal	modal	ADJ
fcis-28893	48	43	alignment	alignment	NOUN
fcis-28893	48	44	between	between	ADP
fcis-28893	48	45	image	image	NOUN
fcis-28893	48	46	features	feature	NOUN
fcis-28893	48	47	and	and	CCONJ
fcis-28893	48	48	text	text	NOUN
fcis-28893	48	49	features	feature	NOUN
fcis-28893	48	50	.	.	PUNCT
fcis-28893	49	1	in	in	ADP
fcis-28893	49	2	this	this	DET
fcis-28893	49	3	paper	paper	NOUN
fcis-28893	49	4	,	,	PUNCT
fcis-28893	49	5	text	text	NOUN
fcis-28893	49	6	-	-	PUNCT
fcis-28893	49	7	based	base	VERB
fcis-28893	49	8	pedestrian	pedestrian	NOUN
fcis-28893	49	9	retrieval	retrieval	NOUN
fcis-28893	49	10	methods	method	NOUN
fcis-28893	49	11	are	be	AUX
fcis-28893	49	12	categorized	categorize	VERB
fcis-28893	49	13	as	as	SCONJ
fcis-28893	49	14	follows	follow	VERB
fcis-28893	49	15	.	.	PUNCT
fcis-28893	50	1	4.1	4.1	NUM
fcis-28893	50	2	.	.	PUNCT
fcis-28893	50	3	method	method	NOUN
fcis-28893	50	4	based	base	VERB
fcis-28893	50	5	on	on	ADP
fcis-28893	50	6	global	global	ADJ
fcis-28893	50	7	feature	feature	NOUN
fcis-28893	50	8	matching	match	VERB
fcis-28893	50	9	li	li	PROPN
fcis-28893	50	10	et	et	PROPN
fcis-28893	50	11	al.[1	al.[1	PROPN
fcis-28893	50	12	]	]	PUNCT
fcis-28893	50	13	first	first	ADV
fcis-28893	50	14	proposed	propose	VERB
fcis-28893	50	15	text	text	NOUN
fcis-28893	50	16	-	-	PUNCT
fcis-28893	50	17	based	base	VERB
fcis-28893	50	18	pedestrian	pedestrian	NOUN
fcis-28893	50	19	retrieval	retrieval	NOUN
fcis-28893	50	20	as	as	ADP
fcis-28893	50	21	a	a	DET
fcis-28893	50	22	cross	cross	ADJ
fcis-28893	50	23	-	-	ADJ
fcis-28893	50	24	modal	modal	ADJ
fcis-28893	50	25	task	task	NOUN
fcis-28893	50	26	,	,	PUNCT
fcis-28893	50	27	and	and	CCONJ
fcis-28893	50	28	designed	design	VERB
fcis-28893	50	29	gna	gna	PROPN
fcis-28893	50	30	-	-	PUNCT
fcis-28893	50	31	rnn	rnn	PROPN
fcis-28893	50	32	,	,	PUNCT
fcis-28893	50	33	an	an	DET
fcis-28893	50	34	rnn	rnn	NOUN
fcis-28893	50	35	network	network	NOUN
fcis-28893	50	36	framework	framework	NOUN
fcis-28893	50	37	with	with	ADP
fcis-28893	50	38	a	a	DET
fcis-28893	50	39	gate	gate	ADJ
fcis-28893	50	40	-	-	PUNCT
fcis-28893	50	41	attention	attention	NOUN
fcis-28893	50	42	mechanism	mechanism	NOUN
fcis-28893	50	43	,	,	PUNCT
fcis-28893	50	44	which	which	PRON
fcis-28893	50	45	is	be	AUX
fcis-28893	50	46	shown	show	VERB
fcis-28893	50	47	in	in	ADP
fcis-28893	50	48	fig	fig	NOUN
fcis-28893	50	49	.	.	PUNCT
fcis-28893	51	1	2	2	X
fcis-28893	51	2	.	.	X
fcis-28893	51	3	li	li	PROPN
fcis-28893	51	4	et	et	PROPN
fcis-28893	51	5	al	al	PROPN
fcis-28893	51	6	.	.	PROPN
fcis-28893	51	7	designed	design	VERB
fcis-28893	51	8	gna	gna	PROPN
fcis-28893	51	9	-	-	PUNCT
fcis-28893	51	10	rnn	rnn	VERB
fcis-28893	51	11	by	by	ADP
fcis-28893	51	12	dividing	divide	VERB
fcis-28893	51	13	the	the	DET
fcis-28893	51	14	network	network	NOUN
fcis-28893	51	15	into	into	ADP
fcis-28893	51	16	a	a	DET
fcis-28893	51	17	visual	visual	ADJ
fcis-28893	51	18	sub	sub	NOUN
fcis-28893	51	19	-	-	NOUN
fcis-28893	51	20	network	network	NOUN
fcis-28893	51	21	based	base	VERB
fcis-28893	51	22	on	on	ADP
fcis-28893	51	23	the	the	DET
fcis-28893	51	24	vgg-16	vgg-16	NOUN
fcis-28893	51	25	and	and	CCONJ
fcis-28893	51	26	a	a	DET
fcis-28893	51	27	textual	textual	ADJ
fcis-28893	51	28	sub	sub	NOUN
fcis-28893	51	29	-	-	NOUN
fcis-28893	51	30	network	network	NOUN
fcis-28893	51	31	based	base	VERB
fcis-28893	51	32	on	on	ADP
fcis-28893	51	33	the	the	DET
fcis-28893	51	34	lstm	lstm	NOUN
fcis-28893	51	35	network[4	network[4	PROPN
fcis-28893	51	36	]	]	PUNCT
fcis-28893	51	37	.	.	PUNCT
fcis-28893	52	1	in	in	ADP
fcis-28893	52	2	the	the	DET
fcis-28893	52	3	visual	visual	ADJ
fcis-28893	52	4	subnet	subnet	NOUN
fcis-28893	52	5	two	two	NUM
fcis-28893	52	6	512	512	NUM
fcis-28893	52	7	-	-	PUNCT
fcis-28893	52	8	dimensional	dimensional	ADJ
fcis-28893	52	9	fully	fully	ADV
fcis-28893	52	10	connected	connected	ADJ
fcis-28893	52	11	layers	layer	NOUN
fcis-28893	52	12	have	have	AUX
fcis-28893	52	13	added	add	VERB
fcis-28893	52	14	for	for	ADP
fcis-28893	52	15	generating	generate	VERB
fcis-28893	52	16	visual	visual	ADJ
fcis-28893	52	17	units	unit	NOUN
fcis-28893	52	18	on	on	ADP
fcis-28893	52	19	top	top	NOUN
fcis-28893	52	20	of	of	ADP
fcis-28893	52	21	the	the	DET
fcis-28893	52	22	drop7	drop7	PROPN
fcis-28893	52	23	layer	layer	NOUN
fcis-28893	52	24	of	of	ADP
fcis-28893	52	25	vgg-16	vgg-16	NOUN
fcis-28893	52	26	to	to	PART
fcis-28893	52	27	recognize	recognize	VERB
fcis-28893	52	28	the	the	DET
fcis-28893	52	29	presence	presence	NOUN
fcis-28893	52	30	or	or	CCONJ
fcis-28893	52	31	absence	absence	NOUN
fcis-28893	52	32	of	of	ADP
fcis-28893	52	33	a	a	DET
fcis-28893	52	34	specific	specific	ADJ
fcis-28893	52	35	pedestrian	pedestrian	NOUN
fcis-28893	52	36	appearance	appearance	NOUN
fcis-28893	52	37	pattern	pattern	NOUN
fcis-28893	52	38	in	in	ADP
fcis-28893	52	39	the	the	DET
fcis-28893	52	40	pedestrian	pedestrian	NOUN
fcis-28893	52	41	images	image	NOUN
fcis-28893	52	42	by	by	ADP
fcis-28893	52	43	visual	visual	ADJ
fcis-28893	52	44	units	unit	NOUN
fcis-28893	52	45	.	.	PUNCT
fcis-28893	53	1	the	the	DET
fcis-28893	53	2	visual	visual	ADJ
fcis-28893	53	3	subnet	subnet	NOUN
fcis-28893	53	4	is	be	AUX
fcis-28893	53	5	first	first	ADJ
fcis-28893	53	6	pre	pre	ADJ
fcis-28893	53	7	-	-	VERB
fcis-28893	53	8	trained	train	VERB
fcis-28893	53	9	on	on	ADP
fcis-28893	53	10	the	the	DET
fcis-28893	53	11	chuk	chuk	PROPN
fcis-28893	53	12	-	-	PUNCT
fcis-28893	53	13	pedes	pes	NOUN
fcis-28893	53	14	dataset	dataset	NOUN
fcis-28893	53	15	assembled	assemble	VERB
fcis-28893	53	16	by	by	ADP
fcis-28893	53	17	the	the	DET
fcis-28893	53	18	authors	author	NOUN
fcis-28893	53	19	,	,	PUNCT
fcis-28893	53	20	then	then	ADV
fcis-28893	53	21	only	only	ADV
fcis-28893	53	22	the	the	DET
fcis-28893	53	23	parameters	parameter	NOUN
fcis-28893	53	24	of	of	ADP
fcis-28893	53	25	the	the	DET
fcis-28893	53	26	added	add	VERB
fcis-28893	53	27	fully	fully	ADV
fcis-28893	53	28	connected	connected	ADJ
fcis-28893	53	29	layer	layer	NOUN
fcis-28893	53	30	are	be	AUX
fcis-28893	53	31	updated	update	VERB
fcis-28893	53	32	during	during	ADP
fcis-28893	53	33	the	the	DET
fcis-28893	53	34	joint	joint	ADJ
fcis-28893	53	35	training	training	NOUN
fcis-28893	53	36	with	with	ADP
fcis-28893	53	37	the	the	DET
fcis-28893	53	38	text	text	NOUN
fcis-28893	53	39	subnet	subnet	NOUN
fcis-28893	53	40	,	,	PUNCT
fcis-28893	53	41	and	and	CCONJ
fcis-28893	53	42	regions	region	NOUN
fcis-28893	53	43	with	with	ADP
fcis-28893	53	44	higher	high	ADJ
fcis-28893	53	45	similarity	similarity	NOUN
fcis-28893	53	46	to	to	ADP
fcis-28893	53	47	the	the	DET
fcis-28893	53	48	text	text	NOUN
fcis-28893	53	49	are	be	AUX
fcis-28893	53	50	given	give	VERB
fcis-28893	53	51	higher	high	ADJ
fcis-28893	53	52	weights	weight	NOUN
fcis-28893	53	53	;	;	PUNCT
fcis-28893	53	54	meanwhile	meanwhile	ADV
fcis-28893	53	55	,	,	PUNCT
fcis-28893	53	56	a	a	DET
fcis-28893	53	57	scalar	scalar	ADJ
fcis-28893	53	58	gate	gate	NOUN
fcis-28893	53	59	mechanism	mechanism	NOUN
fcis-28893	53	60	is	be	AUX
fcis-28893	53	61	designed	design	VERB
fcis-28893	53	62	according	accord	VERB
fcis-28893	53	63	to	to	ADP
fcis-28893	53	64	the	the	DET
fcis-28893	53	65	trait	trait	NOUN
fcis-28893	53	66	that	that	PRON
fcis-28893	53	67	different	different	ADJ
fcis-28893	53	68	words	word	NOUN
fcis-28893	53	69	have	have	VERB
fcis-28893	53	70	different	different	ADJ
fcis-28893	53	71	contribution	contribution	NOUN
fcis-28893	53	72	for	for	ADP
fcis-28893	53	73	image	image	NOUN
fcis-28893	53	74	-	-	PUNCT
fcis-28893	53	75	text	text	NOUN
fcis-28893	53	76	matching	matching	NOUN
fcis-28893	53	77	,	,	PUNCT
fcis-28893	53	78	so	so	SCONJ
fcis-28893	53	79	that	that	SCONJ
fcis-28893	53	80	the	the	DET
fcis-28893	53	81	text	text	NOUN
fcis-28893	53	82	subnet	subnet	NOUN
fcis-28893	53	83	learns	learn	VERB
fcis-28893	53	84	the	the	DET
fcis-28893	53	85	weights	weight	NOUN
fcis-28893	53	86	in	in	ADP
fcis-28893	53	87	the	the	DET
fcis-28893	53	88	word	word	NOUN
fcis-28893	53	89	dimensions	dimension	NOUN
fcis-28893	53	90	,	,	PUNCT
fcis-28893	53	91	measures	measure	VERB
fcis-28893	53	92	the	the	DET
fcis-28893	53	93	contribution	contribution	NOUN
fcis-28893	53	94	of	of	ADP
fcis-28893	53	95	different	different	ADJ
fcis-28893	53	96	words	word	NOUN
fcis-28893	53	97	'	'	PART
fcis-28893	53	98	contribution	contribution	NOUN
fcis-28893	53	99	and	and	CCONJ
fcis-28893	53	100	focus	focus	VERB
fcis-28893	53	101	on	on	ADP
fcis-28893	53	102	words	word	NOUN
fcis-28893	53	103	with	with	ADP
fcis-28893	53	104	high	high	ADJ
fcis-28893	53	105	contribution	contribution	NOUN
fcis-28893	53	106	.	.	PUNCT
fcis-28893	54	1	however	however	ADV
fcis-28893	54	2	,	,	PUNCT
fcis-28893	54	3	gna	gna	PROPN
fcis-28893	54	4	-	-	PUNCT
fcis-28893	54	5	rnn	rnn	PROPN
fcis-28893	54	6	also	also	ADV
fcis-28893	54	7	has	have	VERB
fcis-28893	54	8	many	many	ADJ
fcis-28893	54	9	limitations	limitation	NOUN
fcis-28893	54	10	.	.	PUNCT
fcis-28893	55	1	first	first	ADV
fcis-28893	55	2	,	,	PUNCT
fcis-28893	55	3	gna	gna	PROPN
fcis-28893	55	4	-	-	PUNCT
fcis-28893	55	5	rnn	rnn	PROPN
fcis-28893	55	6	processes	process	NOUN
fcis-28893	55	7	text	text	NOUN
fcis-28893	55	8	word	word	NOUN
fcis-28893	55	9	by	by	ADP
fcis-28893	55	10	word	word	NOUN
fcis-28893	55	11	via	via	ADP
fcis-28893	55	12	lstm	lstm	NOUN
fcis-28893	55	13	,	,	PUNCT
fcis-28893	55	14	ignoring	ignore	VERB
fcis-28893	55	15	the	the	DET
fcis-28893	55	16	continuity	continuity	NOUN
fcis-28893	55	17	between	between	ADP
fcis-28893	55	18	words	word	NOUN
fcis-28893	55	19	,	,	PUNCT
fcis-28893	55	20	and	and	CCONJ
fcis-28893	55	21	is	be	AUX
fcis-28893	55	22	insensitive	insensitive	ADJ
fcis-28893	55	23	to	to	ADP
fcis-28893	55	24	the	the	DET
fcis-28893	55	25	spatial	spatial	ADJ
fcis-28893	55	26	location	location	NOUN
fcis-28893	55	27	of	of	ADP
fcis-28893	55	28	key	key	ADJ
fcis-28893	55	29	pedestrian	pedestrian	NOUN
fcis-28893	55	30	attributes	attribute	NOUN
fcis-28893	55	31	.	.	PUNCT
fcis-28893	56	1	for	for	ADP
fcis-28893	56	2	example	example	NOUN
fcis-28893	56	3	,	,	PUNCT
fcis-28893	56	4	a	a	DET
fcis-28893	56	5	pedestrian	pedestrian	NOUN
fcis-28893	56	6	image	image	NOUN
fcis-28893	56	7	containing	contain	VERB
fcis-28893	56	8	the	the	DET
fcis-28893	56	9	phrase	phrase	NOUN
fcis-28893	56	10	"	"	PUNCT
fcis-28893	56	11	wearing	wear	VERB
fcis-28893	56	12	a	a	DET
fcis-28893	56	13	red	red	ADJ
fcis-28893	56	14	jacket	jacket	NOUN
fcis-28893	56	15	and	and	CCONJ
fcis-28893	56	16	white	white	ADJ
fcis-28893	56	17	short	short	ADJ
fcis-28893	56	18	sleeves	sleeve	NOUN
fcis-28893	56	19	"	"	PUNCT
fcis-28893	56	20	may	may	AUX
fcis-28893	56	21	have	have	VERB
fcis-28893	56	22	a	a	DET
fcis-28893	56	23	higher	high	ADJ
fcis-28893	56	24	similarity	similarity	NOUN
fcis-28893	56	25	to	to	ADP
fcis-28893	56	26	the	the	DET
fcis-28893	56	27	text	text	NOUN
fcis-28893	56	28	describing	describe	VERB
fcis-28893	56	29	"	"	PUNCT
fcis-28893	56	30	red	red	ADJ
fcis-28893	56	31	short	short	ADJ
fcis-28893	56	32	sleeves	sleeve	NOUN
fcis-28893	56	33	"	"	PUNCT
fcis-28893	56	34	.	.	PUNCT
fcis-28893	57	1	secondly	secondly	ADV
fcis-28893	57	2	,	,	PUNCT
fcis-28893	57	3	the	the	DET
fcis-28893	57	4	graphical	graphical	ADJ
fcis-28893	57	5	similarity	similarity	NOUN
fcis-28893	57	6	is	be	AUX
fcis-28893	57	7	obtained	obtain	VERB
fcis-28893	57	8	by	by	ADP
fcis-28893	57	9	simply	simply	ADV
fcis-28893	57	10	adding	add	VERB
fcis-28893	57	11	the	the	DET
fcis-28893	57	12	similarity	similarity	NOUN
fcis-28893	57	13	of	of	ADP
fcis-28893	57	14	each	each	DET
fcis-28893	57	15	word	word	NOUN
fcis-28893	57	16	and	and	CCONJ
fcis-28893	57	17	image	image	NOUN
fcis-28893	57	18	,	,	PUNCT
fcis-28893	57	19	and	and	CCONJ
fcis-28893	57	20	the	the	DET
fcis-28893	57	21	length	length	NOUN
fcis-28893	57	22	of	of	ADP
fcis-28893	57	23	the	the	DET
fcis-28893	57	24	sentence	sentence	NOUN
fcis-28893	57	25	will	will	AUX
fcis-28893	57	26	significantly	significantly	ADV
fcis-28893	57	27	affect	affect	VERB
fcis-28893	57	28	the	the	DET
fcis-28893	57	29	final	final	ADJ
fcis-28893	57	30	similarity	similarity	NOUN
fcis-28893	57	31	score	score	NOUN
fcis-28893	57	32	.	.	PUNCT
fcis-28893	58	1	in	in	ADP
fcis-28893	58	2	addition	addition	NOUN
fcis-28893	58	3	,	,	PUNCT
fcis-28893	58	4	the	the	DET
fcis-28893	58	5	top	top	ADJ
fcis-28893	58	6	1	1	NUM
fcis-28893	58	7	accuracy	accuracy	NOUN
fcis-28893	58	8	of	of	ADP
fcis-28893	58	9	gna	gna	PROPN
fcis-28893	58	10	-	-	PUNCT
fcis-28893	58	11	rnn	rnn	PROPN
fcis-28893	58	12	is	be	AUX
fcis-28893	58	13	only	only	ADV
fcis-28893	58	14	19.05	19.05	NUM
fcis-28893	58	15	,	,	PUNCT
fcis-28893	58	16	needs	need	VERB
fcis-28893	58	17	to	to	PART
fcis-28893	58	18	be	be	AUX
fcis-28893	58	19	improved	improve	VERB
fcis-28893	58	20	.	.	PUNCT
fcis-28893	59	1	fig	fig	NOUN
fcis-28893	59	2	2	2	NUM
fcis-28893	59	3	.	.	PUNCT
fcis-28893	59	4	framework	framework	NOUN
fcis-28893	59	5	of	of	ADP
fcis-28893	59	6	gna	gna	PROPN
fcis-28893	59	7	-	-	PUNCT
fcis-28893	59	8	rnn	rnn	PROPN
fcis-28893	59	9	after	after	ADP
fcis-28893	59	10	gna	gna	PROPN
fcis-28893	59	11	-	-	PUNCT
fcis-28893	59	12	rnn	rnn	PROPN
fcis-28893	59	13	,	,	PUNCT
fcis-28893	59	14	li	li	PROPN
fcis-28893	59	15	et	et	PROPN
fcis-28893	59	16	al.[5	al.[5	PROPN
fcis-28893	59	17	]	]	PUNCT
fcis-28893	59	18	noticed	notice	VERB
fcis-28893	59	19	that	that	SCONJ
fcis-28893	59	20	the	the	DET
fcis-28893	59	21	labeling	labeling	NOUN
fcis-28893	59	22	information	information	NOUN
fcis-28893	59	23	with	with	ADP
fcis-28893	59	24	pedestrian	pedestrian	NOUN
fcis-28893	59	25	identities	identity	NOUN
fcis-28893	59	26	was	be	AUX
fcis-28893	59	27	underutilized	underutilize	VERB
fcis-28893	59	28	and	and	CCONJ
fcis-28893	59	29	proposed	propose	VERB
fcis-28893	59	30	a	a	DET
fcis-28893	59	31	two	two	NUM
fcis-28893	59	32	-	-	PUNCT
fcis-28893	59	33	stage	stage	NOUN
fcis-28893	59	34	network	network	NOUN
fcis-28893	59	35	for	for	ADP
fcis-28893	59	36	iatv	iatv	PROPN
fcis-28893	59	37	with	with	ADP
fcis-28893	59	38	identityawareness	identityawareness	NOUN
fcis-28893	59	39	capability	capability	NOUN
fcis-28893	59	40	.	.	PUNCT
fcis-28893	60	1	the	the	DET
fcis-28893	60	2	first	first	ADJ
fcis-28893	60	3	stage	stage	NOUN
fcis-28893	60	4	network	network	NOUN
fcis-28893	60	5	of	of	ADP
fcis-28893	60	6	iatv	iatv	PROPN
fcis-28893	60	7	consists	consist	VERB
fcis-28893	60	8	of	of	ADP
fcis-28893	60	9	a	a	DET
fcis-28893	60	10	cnn	cnn	PROPN
fcis-28893	60	11	-	-	PUNCT
fcis-28893	60	12	lstm	lstm	ADJ
fcis-28893	60	13	network	network	NOUN
fcis-28893	60	14	,	,	PUNCT
fcis-28893	60	15	which	which	PRON
fcis-28893	60	16	is	be	AUX
fcis-28893	60	17	responsible	responsible	ADJ
fcis-28893	60	18	for	for	ADP
fcis-28893	60	19	the	the	DET
fcis-28893	60	20	extraction	extraction	NOUN
fcis-28893	60	21	of	of	ADP
fcis-28893	60	22	graphic	graphic	ADJ
fcis-28893	60	23	features	feature	NOUN
fcis-28893	60	24	and	and	CCONJ
fcis-28893	60	25	initial	initial	ADJ
fcis-28893	60	26	alignment	alignment	NOUN
fcis-28893	60	27	.	.	PUNCT
fcis-28893	61	1	in	in	ADP
fcis-28893	61	2	this	this	DET
fcis-28893	61	3	stage	stage	NOUN
fcis-28893	61	4	,	,	PUNCT
fcis-28893	61	5	li	li	PROPN
fcis-28893	61	6	et	et	PROPN
fcis-28893	61	7	al	al	PROPN
fcis-28893	61	8	.	.	PROPN
fcis-28893	61	9	designed	design	VERB
fcis-28893	61	10	a	a	DET
fcis-28893	61	11	new	new	ADJ
fcis-28893	61	12	cross	cross	ADJ
fcis-28893	61	13	-	-	ADJ
fcis-28893	61	14	modal	modal	ADJ
fcis-28893	61	15	cross	cross	ADJ
fcis-28893	61	16	-	-	ADJ
fcis-28893	61	17	entropy	entropy	ADJ
fcis-28893	61	18	loss	loss	NOUN
fcis-28893	61	19	function	function	NOUN
fcis-28893	61	20	cmce	cmce	NOUN
fcis-28893	61	21	,	,	PUNCT
fcis-28893	61	22	which	which	PRON
fcis-28893	61	23	introduces	introduce	VERB
fcis-28893	61	24	a	a	DET
fcis-28893	61	25	feature	feature	NOUN
fcis-28893	61	26	buffer	buffer	NOUN
fcis-28893	61	27	mechanism	mechanism	NOUN
fcis-28893	61	28	to	to	PART
fcis-28893	61	29	simplify	simplify	VERB
fcis-28893	61	30	the	the	DET
fcis-28893	61	31	screening	screening	NOUN
fcis-28893	61	32	process	process	NOUN
fcis-28893	61	33	of	of	ADP
fcis-28893	61	34	difficult	difficult	ADJ
fcis-28893	61	35	negative	negative	ADJ
fcis-28893	61	36	samples	sample	NOUN
fcis-28893	61	37	and	and	CCONJ
fcis-28893	61	38	enables	enable	VERB
fcis-28893	61	39	the	the	DET
fcis-28893	61	40	model	model	NOUN
fcis-28893	61	41	to	to	PART
fcis-28893	61	42	efficiently	efficiently	ADV
fcis-28893	61	43	calculate	calculate	VERB
fcis-28893	61	44	the	the	DET
fcis-28893	61	45	similarity	similarity	NOUN
fcis-28893	61	46	between	between	ADP
fcis-28893	61	47	samples	sample	NOUN
fcis-28893	61	48	and	and	CCONJ
fcis-28893	61	49	other	other	ADJ
fcis-28893	61	50	identities	identity	NOUN
fcis-28893	61	51	,	,	PUNCT
fcis-28893	61	52	providing	provide	VERB
fcis-28893	61	53	the	the	DET
fcis-28893	61	54	basis	basis	NOUN
fcis-28893	61	55	for	for	ADP
fcis-28893	61	56	the	the	DET
fcis-28893	61	57	second	second	ADJ
fcis-28893	61	58	stage	stage	NOUN
fcis-28893	61	59	network	network	NOUN
fcis-28893	61	60	.	.	PUNCT
fcis-28893	62	1	in	in	ADP
fcis-28893	62	2	the	the	DET
fcis-28893	62	3	second	second	ADJ
fcis-28893	62	4	stage	stage	NOUN
fcis-28893	62	5	,	,	PUNCT
fcis-28893	62	6	li	li	PROPN
fcis-28893	62	7	et	et	PROPN
fcis-28893	62	8	al	al	PROPN
fcis-28893	62	9	.	.	PROPN
fcis-28893	62	10	introduced	introduce	VERB
fcis-28893	62	11	a	a	DET
fcis-28893	62	12	potential	potential	ADJ
fcis-28893	62	13	collaborative	collaborative	ADJ
fcis-28893	62	14	attention	attention	NOUN
fcis-28893	62	15	mechanism	mechanism	NOUN
fcis-28893	62	16	and	and	CCONJ
fcis-28893	62	17	a	a	DET
fcis-28893	62	18	spatial	spatial	ADJ
fcis-28893	62	19	attention	attention	NOUN
fcis-28893	62	20	mechanism	mechanism	NOUN
fcis-28893	62	21	for	for	ADP
fcis-28893	62	22	aligning	align	VERB
fcis-28893	62	23	texts	text	NOUN
fcis-28893	62	24	with	with	ADP
fcis-28893	62	25	different	different	ADJ
fcis-28893	62	26	sentence	sentence	NOUN
fcis-28893	62	27	structures	structure	NOUN
fcis-28893	62	28	and	and	CCONJ
fcis-28893	62	29	reducing	reduce	VERB
fcis-28893	62	30	the	the	DET
fcis-28893	62	31	impact	impact	NOUN
fcis-28893	62	32	of	of	ADP
fcis-28893	62	33	sentence	sentence	NOUN
fcis-28893	62	34	differences	difference	NOUN
fcis-28893	62	35	on	on	ADP
fcis-28893	62	36	the	the	DET
fcis-28893	62	37	model	model	NOUN
fcis-28893	62	38	.	.	PUNCT
fcis-28893	63	1	the	the	DET
fcis-28893	63	2	introduction	introduction	NOUN
fcis-28893	63	3	of	of	ADP
fcis-28893	63	4	the	the	DET
fcis-28893	63	5	cmce	cmce	NOUN
fcis-28893	63	6	loss	loss	NOUN
fcis-28893	63	7	function	function	NOUN
fcis-28893	63	8	significantly	significantly	ADV
fcis-28893	63	9	advances	advance	VERB
fcis-28893	63	10	the	the	DET
fcis-28893	63	11	development	development	NOUN
fcis-28893	63	12	of	of	ADP
fcis-28893	63	13	text	text	NOUN
fcis-28893	63	14	-	-	PUNCT
fcis-28893	63	15	based	base	VERB
fcis-28893	63	16	pedestrian	pedestrian	NOUN
fcis-28893	63	17	retrieval	retrieval	NOUN
fcis-28893	63	18	tasks	task	NOUN
fcis-28893	63	19	.	.	PUNCT
fcis-28893	64	1	however	however	ADV
fcis-28893	64	2	,	,	PUNCT
fcis-28893	64	3	iatv	iatv	PROPN
fcis-28893	64	4	uses	use	VERB
fcis-28893	64	5	a	a	DET
fcis-28893	64	6	two	two	NUM
fcis-28893	64	7	-	-	PUNCT
fcis-28893	64	8	stage	stage	NOUN
fcis-28893	64	9	network	network	NOUN
fcis-28893	64	10	framework	framework	NOUN
fcis-28893	64	11	in	in	ADP
fcis-28893	64	12	pursuit	pursuit	NOUN
fcis-28893	64	13	of	of	ADP
fcis-28893	64	14	higher	high	ADJ
fcis-28893	64	15	accuracy	accuracy	NOUN
fcis-28893	64	16	,	,	PUNCT
fcis-28893	64	17	and	and	CCONJ
fcis-28893	64	18	the	the	DET
fcis-28893	64	19	training	training	NOUN
fcis-28893	64	20	process	process	NOUN
fcis-28893	64	21	needs	need	VERB
fcis-28893	64	22	to	to	PART
fcis-28893	64	23	be	be	AUX
fcis-28893	64	24	carried	carry	VERB
fcis-28893	64	25	out	out	ADP
fcis-28893	64	26	separately	separately	ADV
fcis-28893	64	27	,	,	PUNCT
fcis-28893	64	28	while	while	SCONJ
fcis-28893	64	29	the	the	DET
fcis-28893	64	30	addition	addition	NOUN
fcis-28893	64	31	of	of	ADP
fcis-28893	64	32	multiple	multiple	ADJ
fcis-28893	64	33	attention	attention	NOUN
fcis-28893	64	34	mechanisms	mechanism	NOUN
fcis-28893	64	35	leads	lead	VERB
fcis-28893	64	36	to	to	ADP
fcis-28893	64	37	a	a	DET
fcis-28893	64	38	significant	significant	ADJ
fcis-28893	64	39	increase	increase	NOUN
fcis-28893	64	40	in	in	ADP
fcis-28893	64	41	the	the	DET
fcis-28893	64	42	model	model	NOUN
fcis-28893	64	43	structure	structure	NOUN
fcis-28893	64	44	and	and	CCONJ
fcis-28893	64	45	computational	computational	ADJ
fcis-28893	64	46	complexity	complexity	NOUN
fcis-28893	64	47	,	,	PUNCT
fcis-28893	64	48	which	which	PRON
fcis-28893	64	49	prevents	prevent	VERB
fcis-28893	64	50	end	end	NOUN
fcis-28893	64	51	-	-	PUNCT
fcis-28893	64	52	to	to	ADP
fcis-28893	64	53	-	-	PUNCT
fcis-28893	64	54	end	end	NOUN
fcis-28893	64	55	training	training	NOUN
fcis-28893	64	56	.	.	PUNCT
fcis-28893	65	1	chen	chen	PROPN
fcis-28893	65	2	et	et	PROPN
fcis-28893	65	3	al.[6	al.[6	PROPN
fcis-28893	65	4	]	]	PUNCT
fcis-28893	65	5	proposed	propose	VERB
fcis-28893	65	6	an	an	DET
fcis-28893	65	7	improvement	improvement	NOUN
fcis-28893	65	8	idea	idea	NOUN
fcis-28893	65	9	to	to	PART
fcis-28893	65	10	address	address	VERB
fcis-28893	65	11	the	the	DET
fcis-28893	65	12	limitations	limitation	NOUN
fcis-28893	65	13	of	of	ADP
fcis-28893	65	14	other	other	ADJ
fcis-28893	65	15	methods	method	NOUN
fcis-28893	65	16	in	in	ADP
fcis-28893	65	17	the	the	DET
fcis-28893	65	18	field	field	NOUN
fcis-28893	65	19	,	,	PUNCT
fcis-28893	65	20	especially	especially	ADV
fcis-28893	65	21	gnarnn	gnarnn	NOUN
fcis-28893	65	22	.	.	PUNCT
fcis-28893	66	1	chen	chen	PROPN
fcis-28893	66	2	et	et	PROPN
fcis-28893	66	3	al	al	PROPN
fcis-28893	66	4	.	.	PROPN
fcis-28893	66	5	proposed	propose	VERB
fcis-28893	66	6	an	an	DET
fcis-28893	66	7	adaptive	adaptive	ADJ
fcis-28893	66	8	thresholding	thresholding	NOUN
fcis-28893	66	9	image	image	NOUN
fcis-28893	66	10	block	block	NOUN
fcis-28893	66	11	and	and	CCONJ
fcis-28893	66	12	word	word	NOUN
fcis-28893	66	13	matching	matching	NOUN
fcis-28893	66	14	network	network	NOUN
fcis-28893	66	15	architecture	architecture	NOUN
fcis-28893	66	16	,	,	PUNCT
fcis-28893	66	17	pwm	pwm	NOUN
fcis-28893	66	18	(	(	PUNCT
fcis-28893	66	19	patchword	patchword	NOUN
fcis-28893	66	20	matching	matching	NOUN
fcis-28893	66	21	)	)	PUNCT
fcis-28893	66	22	.	.	PUNCT
fcis-28893	67	1	when	when	SCONJ
fcis-28893	67	2	performing	perform	VERB
fcis-28893	67	3	image	image	NOUN
fcis-28893	67	4	block	block	NOUN
fcis-28893	67	5	and	and	CCONJ
fcis-28893	67	6	word	word	NOUN
fcis-28893	67	7	matching	matching	NOUN
fcis-28893	67	8	,	,	PUNCT
fcis-28893	67	9	pwm	pwm	NOUN
fcis-28893	67	10	assigns	assign	VERB
fcis-28893	67	11	the	the	DET
fcis-28893	67	12	highest	high	ADJ
fcis-28893	67	13	matching	matching	NOUN
fcis-28893	67	14	score	score	NOUN
fcis-28893	67	15	to	to	ADP
fcis-28893	67	16	an	an	DET
fcis-28893	67	17	image	image	NOUN
fcis-28893	67	18	block	block	NOUN
fcis-28893	67	19	only	only	ADV
fcis-28893	67	20	when	when	SCONJ
fcis-28893	67	21	two	two	NUM
fcis-28893	67	22	semantically	semantically	ADV
fcis-28893	67	23	consecutive	consecutive	ADJ
fcis-28893	67	24	key	key	ADJ
fcis-28893	67	25	words	word	NOUN
fcis-28893	67	26	correspond	correspond	VERB
fcis-28893	67	27	to	to	ADP
fcis-28893	67	28	that	that	DET
fcis-28893	67	29	image	image	NOUN
fcis-28893	67	30	block	block	NOUN
fcis-28893	67	31	at	at	ADP
fcis-28893	67	32	the	the	DET
fcis-28893	67	33	same	same	ADJ
fcis-28893	67	34	time	time	NOUN
fcis-28893	67	35	,	,	PUNCT
fcis-28893	67	36	thus	thus	ADV
fcis-28893	67	37	effectively	effectively	ADV
fcis-28893	67	38	solving	solve	VERB
fcis-28893	67	39	the	the	DET
fcis-28893	67	40	problem	problem	NOUN
fcis-28893	67	41	of	of	ADP
fcis-28893	67	42	spatial	spatial	ADJ
fcis-28893	67	43	location	location	NOUN
fcis-28893	67	44	insensitivity	insensitivity	NOUN
fcis-28893	67	45	of	of	ADP
fcis-28893	67	46	the	the	DET
fcis-28893	67	47	key	key	ADJ
fcis-28893	67	48	attributes	attribute	NOUN
fcis-28893	67	49	of	of	ADP
fcis-28893	67	50	the	the	DET
fcis-28893	67	51	characters	character	NOUN
fcis-28893	67	52	in	in	ADP
fcis-28893	67	53	the	the	DET
fcis-28893	67	54	text	text	NOUN
fcis-28893	67	55	processed	process	VERB
fcis-28893	67	56	by	by	ADP
fcis-28893	67	57	the	the	DET
fcis-28893	67	58	gna	gna	PROPN
fcis-28893	67	59	-	-	PUNCT
fcis-28893	67	60	rnn	rnn	PROPN
fcis-28893	67	61	model	model	NOUN
fcis-28893	67	62	.	.	PUNCT
fcis-28893	68	1	the	the	DET
fcis-28893	68	2	adaptive	adaptive	ADJ
fcis-28893	68	3	threshold	threshold	NOUN
fcis-28893	68	4	mechanism	mechanism	NOUN
fcis-28893	68	5	predicts	predict	VERB
fcis-28893	68	6	a	a	DET
fcis-28893	68	7	matching	matching	NOUN
fcis-28893	68	8	threshold	threshold	NOUN
fcis-28893	68	9	for	for	ADP
fcis-28893	68	10	each	each	DET
fcis-28893	68	11	word	word	NOUN
fcis-28893	68	12	in	in	ADP
fcis-28893	68	13	advance	advance	NOUN
fcis-28893	68	14	,	,	PUNCT
fcis-28893	68	15	and	and	CCONJ
fcis-28893	68	16	when	when	SCONJ
fcis-28893	68	17	the	the	DET
fcis-28893	68	18	matching	matching	NOUN
fcis-28893	68	19	score	score	NOUN
fcis-28893	68	20	is	be	AUX
fcis-28893	68	21	lower	low	ADJ
fcis-28893	68	22	than	than	ADP
fcis-28893	68	23	the	the	DET
fcis-28893	68	24	threshold	threshold	NOUN
fcis-28893	68	25	,	,	PUNCT
fcis-28893	68	26	it	it	PRON
fcis-28893	68	27	is	be	AUX
fcis-28893	68	28	determined	determine	VERB
fcis-28893	68	29	as	as	ADP
fcis-28893	68	30	a	a	DET
fcis-28893	68	31	negative	negative	ADJ
fcis-28893	68	32	121	121	NUM
fcis-28893	68	33	sample	sample	NOUN
fcis-28893	68	34	,	,	PUNCT
fcis-28893	68	35	and	and	CCONJ
fcis-28893	68	36	no	no	DET
fcis-28893	68	37	operation	operation	NOUN
fcis-28893	68	38	is	be	AUX
fcis-28893	68	39	performed	perform	VERB
fcis-28893	68	40	;	;	PUNCT
fcis-28893	68	41	when	when	SCONJ
fcis-28893	68	42	the	the	DET
fcis-28893	68	43	score	score	NOUN
fcis-28893	68	44	is	be	AUX
fcis-28893	68	45	higher	high	ADJ
fcis-28893	68	46	than	than	ADP
fcis-28893	68	47	the	the	DET
fcis-28893	68	48	threshold	threshold	NOUN
fcis-28893	68	49	,	,	PUNCT
fcis-28893	68	50	it	it	PRON
fcis-28893	68	51	determines	determine	VERB
fcis-28893	68	52	that	that	SCONJ
fcis-28893	68	53	the	the	DET
fcis-28893	68	54	image	image	NOUN
fcis-28893	68	55	block	block	NOUN
fcis-28893	68	56	matches	match	VERB
fcis-28893	68	57	the	the	DET
fcis-28893	68	58	word	word	NOUN
fcis-28893	68	59	,	,	PUNCT
fcis-28893	68	60	and	and	CCONJ
fcis-28893	68	61	automatically	automatically	ADV
fcis-28893	68	62	adjusts	adjust	VERB
fcis-28893	68	63	the	the	DET
fcis-28893	68	64	matching	match	VERB
fcis-28893	68	65	score	score	NOUN
fcis-28893	68	66	to	to	ADP
fcis-28893	68	67	the	the	DET
fcis-28893	68	68	vicinity	vicinity	NOUN
fcis-28893	68	69	of	of	ADP
fcis-28893	68	70	the	the	DET
fcis-28893	68	71	threshold	threshold	NOUN
fcis-28893	68	72	,	,	PUNCT
fcis-28893	68	73	to	to	PART
fcis-28893	68	74	avoid	avoid	VERB
fcis-28893	68	75	erroneous	erroneous	ADJ
fcis-28893	68	76	judgment	judgment	NOUN
fcis-28893	68	77	caused	cause	VERB
fcis-28893	68	78	by	by	ADP
fcis-28893	68	79	the	the	DET
fcis-28893	68	80	matching	matching	NOUN
fcis-28893	68	81	score	score	NOUN
fcis-28893	68	82	of	of	ADP
fcis-28893	68	83	the	the	DET
fcis-28893	68	84	image	image	NOUN
fcis-28893	68	85	block	block	NOUN
fcis-28893	68	86	and	and	CCONJ
fcis-28893	68	87	the	the	DET
fcis-28893	68	88	word	word	NOUN
fcis-28893	68	89	being	be	AUX
fcis-28893	68	90	too	too	ADV
fcis-28893	68	91	high	high	ADJ
fcis-28893	68	92	.	.	PUNCT
fcis-28893	69	1	pwm	pwm	NOUN
fcis-28893	69	2	overcomes	overcome	VERB
fcis-28893	69	3	the	the	DET
fcis-28893	69	4	problems	problem	NOUN
fcis-28893	69	5	in	in	ADP
fcis-28893	69	6	the	the	DET
fcis-28893	69	7	gna	gna	PROPN
fcis-28893	69	8	-	-	PUNCT
fcis-28893	69	9	rnn	rnn	PROPN
fcis-28893	69	10	method	method	NOUN
fcis-28893	69	11	by	by	ADP
fcis-28893	69	12	virtue	virtue	NOUN
fcis-28893	69	13	of	of	ADP
fcis-28893	69	14	the	the	DET
fcis-28893	69	15	adaptive	adaptive	ADJ
fcis-28893	69	16	thresholding	thresholding	NOUN
fcis-28893	69	17	mechanism	mechanism	NOUN
fcis-28893	69	18	of	of	ADP
fcis-28893	69	19	the	the	DET
fcis-28893	69	20	continuous	continuous	ADJ
fcis-28893	69	21	matching	matching	NOUN
fcis-28893	69	22	mechanism	mechanism	NOUN
fcis-28893	69	23	of	of	ADP
fcis-28893	69	24	the	the	DET
fcis-28893	69	25	image	image	NOUN
fcis-28893	69	26	and	and	CCONJ
fcis-28893	69	27	text	text	NOUN
fcis-28893	69	28	and	and	CCONJ
fcis-28893	69	29	adopts	adopt	VERB
fcis-28893	69	30	a	a	DET
fcis-28893	69	31	hierarchical	hierarchical	ADJ
fcis-28893	69	32	computational	computational	ADJ
fcis-28893	69	33	approach	approach	NOUN
fcis-28893	69	34	to	to	PART
fcis-28893	69	35	compute	compute	VERB
fcis-28893	69	36	the	the	DET
fcis-28893	69	37	overall	overall	ADJ
fcis-28893	69	38	matching	matching	NOUN
fcis-28893	69	39	scores	score	NOUN
fcis-28893	69	40	of	of	ADP
fcis-28893	69	41	the	the	DET
fcis-28893	69	42	image	image	NOUN
fcis-28893	69	43	and	and	CCONJ
fcis-28893	69	44	text	text	NOUN
fcis-28893	69	45	layer	layer	NOUN
fcis-28893	69	46	by	by	ADP
fcis-28893	69	47	layer	layer	NOUN
fcis-28893	69	48	,	,	PUNCT
fcis-28893	69	49	which	which	PRON
fcis-28893	69	50	significantly	significantly	ADV
fcis-28893	69	51	improves	improve	VERB
fcis-28893	69	52	the	the	DET
fcis-28893	69	53	sota	sota	NOUN
fcis-28893	69	54	(	(	PUNCT
fcis-28893	69	55	state	state	NOUN
fcis-28893	69	56	-	-	PUNCT
fcis-28893	69	57	of	of	ADP
fcis-28893	69	58	-	-	PUNCT
fcis-28893	69	59	the	the	DET
fcis-28893	69	60	-	-	PUNCT
fcis-28893	69	61	art	art	NOUN
fcis-28893	69	62	)	)	PUNCT
fcis-28893	69	63	level	level	NOUN
fcis-28893	69	64	of	of	ADP
fcis-28893	69	65	the	the	DET
fcis-28893	69	66	text	text	NOUN
fcis-28893	69	67	-	-	PUNCT
fcis-28893	69	68	based	base	VERB
fcis-28893	69	69	pedestrian	pedestrian	NOUN
fcis-28893	69	70	retrieval	retrieval	NOUN
fcis-28893	69	71	task	task	NOUN
fcis-28893	69	72	and	and	CCONJ
fcis-28893	69	73	pushes	push	VERB
fcis-28893	69	74	forward	forward	ADV
fcis-28893	69	75	the	the	DET
fcis-28893	69	76	further	further	ADJ
fcis-28893	69	77	development	development	NOUN
fcis-28893	69	78	of	of	ADP
fcis-28893	69	79	the	the	DET
fcis-28893	69	80	field	field	NOUN
fcis-28893	69	81	.	.	PUNCT
fcis-28893	70	1	zhang	zhang	PROPN
fcis-28893	70	2	et	et	PROPN
fcis-28893	70	3	al.[7	al.[7	PROPN
fcis-28893	70	4	]	]	PUNCT
fcis-28893	70	5	agree	agree	VERB
fcis-28893	70	6	with	with	ADP
fcis-28893	70	7	iatv	iatv	PROPN
fcis-28893	70	8	that	that	PRON
fcis-28893	70	9	individual	individual	ADJ
fcis-28893	70	10	identity	identity	NOUN
fcis-28893	70	11	information	information	NOUN
fcis-28893	70	12	must	must	AUX
fcis-28893	70	13	be	be	AUX
fcis-28893	70	14	fully	fully	ADV
fcis-28893	70	15	utilized	utilize	VERB
fcis-28893	70	16	,	,	PUNCT
fcis-28893	70	17	but	but	CCONJ
fcis-28893	70	18	the	the	DET
fcis-28893	70	19	cmce	cmce	NOUN
fcis-28893	70	20	loss	loss	NOUN
fcis-28893	70	21	function	function	NOUN
fcis-28893	70	22	proposed	propose	VERB
fcis-28893	70	23	by	by	ADP
fcis-28893	70	24	li	li	PROPN
fcis-28893	70	25	et	et	PROPN
fcis-28893	70	26	al	al	PROPN
fcis-28893	70	27	.	.	PROPN
fcis-28893	70	28	requires	require	VERB
fcis-28893	70	29	additional	additional	ADJ
fcis-28893	70	30	feature	feature	NOUN
fcis-28893	70	31	buffer	buffer	NOUN
fcis-28893	70	32	allocation	allocation	NOUN
fcis-28893	70	33	,	,	PUNCT
fcis-28893	70	34	which	which	PRON
fcis-28893	70	35	leads	lead	VERB
fcis-28893	70	36	to	to	ADP
fcis-28893	70	37	high	high	ADJ
fcis-28893	70	38	memory	memory	NOUN
fcis-28893	70	39	consumption	consumption	NOUN
fcis-28893	70	40	when	when	SCONJ
fcis-28893	70	41	processing	process	VERB
fcis-28893	70	42	a	a	DET
fcis-28893	70	43	large	large	ADJ
fcis-28893	70	44	number	number	NOUN
fcis-28893	70	45	of	of	ADP
fcis-28893	70	46	targets	target	NOUN
fcis-28893	70	47	.	.	PUNCT
fcis-28893	71	1	to	to	PART
fcis-28893	71	2	solve	solve	VERB
fcis-28893	71	3	this	this	DET
fcis-28893	71	4	problem	problem	NOUN
fcis-28893	71	5	,	,	PUNCT
fcis-28893	71	6	zhang	zhang	PROPN
fcis-28893	71	7	et	et	PROPN
fcis-28893	71	8	al	al	PROPN
fcis-28893	71	9	.	.	PROPN
fcis-28893	71	10	proposed	propose	VERB
fcis-28893	71	11	the	the	DET
fcis-28893	71	12	cross	cross	ADJ
fcis-28893	71	13	-	-	ADJ
fcis-28893	71	14	modal	modal	ADJ
fcis-28893	71	15	mapping	mapping	NOUN
fcis-28893	71	16	matching	matching	NOUN
fcis-28893	71	17	loss	loss	NOUN
fcis-28893	71	18	function	function	NOUN
fcis-28893	71	19	cmpm	cmpm	NOUN
fcis-28893	71	20	and	and	CCONJ
fcis-28893	71	21	the	the	DET
fcis-28893	71	22	cross	cross	ADJ
fcis-28893	71	23	-	-	ADJ
fcis-28893	71	24	modal	modal	ADJ
fcis-28893	71	25	mapping	mapping	NOUN
fcis-28893	71	26	classification	classification	NOUN
fcis-28893	71	27	loss	loss	NOUN
fcis-28893	71	28	function	function	NOUN
fcis-28893	71	29	cmpc	cmpc	PROPN
fcis-28893	71	30	.	.	PUNCT
fcis-28893	72	1	the	the	DET
fcis-28893	72	2	network	network	NOUN
fcis-28893	72	3	schematic	schematic	ADJ
fcis-28893	72	4	is	be	AUX
fcis-28893	72	5	shown	show	VERB
fcis-28893	72	6	in	in	ADP
fcis-28893	72	7	fig	fig	NOUN
fcis-28893	72	8	.	.	PUNCT
fcis-28893	73	1	3	3	X
fcis-28893	73	2	.	.	X
fcis-28893	73	3	the	the	DET
fcis-28893	73	4	visual	visual	ADJ
fcis-28893	73	5	part	part	NOUN
fcis-28893	73	6	is	be	AUX
fcis-28893	73	7	based	base	VERB
fcis-28893	73	8	on	on	ADP
fcis-28893	73	9	mobilenet	mobilenet	NOUN
fcis-28893	73	10	as	as	ADP
fcis-28893	73	11	the	the	DET
fcis-28893	73	12	backbone	backbone	NOUN
fcis-28893	73	13	network	network	NOUN
fcis-28893	73	14	and	and	CCONJ
fcis-28893	73	15	the	the	DET
fcis-28893	73	16	text	text	NOUN
fcis-28893	73	17	part	part	NOUN
fcis-28893	73	18	is	be	AUX
fcis-28893	73	19	a	a	DET
fcis-28893	73	20	bidirectional	bidirectional	ADJ
fcis-28893	73	21	lstm	lstm	NOUN
fcis-28893	73	22	network	network	NOUN
fcis-28893	73	23	.	.	PUNCT
fcis-28893	74	1	unlike	unlike	ADP
fcis-28893	74	2	previous	previous	ADJ
fcis-28893	74	3	work	work	NOUN
fcis-28893	74	4	after	after	ADP
fcis-28893	74	5	inputting	inputte	VERB
fcis-28893	74	6	the	the	DET
fcis-28893	74	7	images	image	NOUN
fcis-28893	74	8	,	,	PUNCT
fcis-28893	74	9	cmpm	cmpm	PROPN
fcis-28893	74	10	considers	consider	VERB
fcis-28893	74	11	all	all	DET
fcis-28893	74	12	the	the	DET
fcis-28893	74	13	candidate	candidate	NOUN
fcis-28893	74	14	texts	text	NOUN
fcis-28893	74	15	in	in	ADP
fcis-28893	74	16	a	a	DET
fcis-28893	74	17	small	small	ADJ
fcis-28893	74	18	batch	batch	NOUN
fcis-28893	74	19	(	(	PUNCT
fcis-28893	74	20	mini	mini	NOUN
fcis-28893	74	21	batch	batch	NOUN
fcis-28893	74	22	)	)	PUNCT
fcis-28893	74	23	and	and	CCONJ
fcis-28893	74	24	incorporates	incorporate	VERB
fcis-28893	74	25	them	they	PRON
fcis-28893	74	26	into	into	ADP
fcis-28893	74	27	the	the	DET
fcis-28893	74	28	computation	computation	NOUN
fcis-28893	74	29	of	of	ADP
fcis-28893	74	30	the	the	DET
fcis-28893	74	31	classification	classification	NOUN
fcis-28893	74	32	loss	loss	NOUN
fcis-28893	74	33	function	function	NOUN
fcis-28893	74	34	,	,	PUNCT
fcis-28893	74	35	avoiding	avoid	VERB
fcis-28893	74	36	the	the	DET
fcis-28893	74	37	cumbersome	cumbersome	ADJ
fcis-28893	74	38	sampling	sampling	NOUN
fcis-28893	74	39	process	process	NOUN
fcis-28893	74	40	in	in	ADP
fcis-28893	74	41	the	the	DET
fcis-28893	74	42	traditional	traditional	ADJ
fcis-28893	74	43	bidirectional	bidirectional	ADJ
fcis-28893	74	44	lstm	lstm	PROPN
fcis-28893	74	45	model	model	NOUN
fcis-28893	74	46	.	.	PUNCT
fcis-28893	75	1	for	for	ADP
fcis-28893	75	2	the	the	DET
fcis-28893	75	3	classification	classification	NOUN
fcis-28893	75	4	task	task	NOUN
fcis-28893	75	5	with	with	ADP
fcis-28893	75	6	identity	identity	NOUN
fcis-28893	75	7	annotation	annotation	NOUN
fcis-28893	75	8	information	information	NOUN
fcis-28893	75	9	,	,	PUNCT
fcis-28893	75	10	cmpc	cmpc	PROPN
fcis-28893	75	11	projects	project	VERB
fcis-28893	75	12	the	the	DET
fcis-28893	75	13	feature	feature	NOUN
fcis-28893	75	14	vectors	vector	NOUN
fcis-28893	75	15	onto	onto	ADP
fcis-28893	75	16	the	the	DET
fcis-28893	75	17	matching	matching	NOUN
fcis-28893	75	18	features	feature	NOUN
fcis-28893	75	19	in	in	ADP
fcis-28893	75	20	another	another	DET
fcis-28893	75	21	modality	modality	NOUN
fcis-28893	75	22	and	and	CCONJ
fcis-28893	75	23	then	then	ADV
fcis-28893	75	24	classifies	classify	VERB
fcis-28893	75	25	them	they	PRON
fcis-28893	75	26	,	,	PUNCT
fcis-28893	75	27	while	while	SCONJ
fcis-28893	75	28	integrating	integrate	VERB
fcis-28893	75	29	the	the	DET
fcis-28893	75	30	similarity	similarity	NOUN
fcis-28893	75	31	scores	score	NOUN
fcis-28893	75	32	of	of	ADP
fcis-28893	75	33	the	the	DET
fcis-28893	75	34	graphic	graphic	ADJ
fcis-28893	75	35	sample	sample	NOUN
fcis-28893	75	36	pairs	pair	NOUN
fcis-28893	75	37	into	into	ADP
fcis-28893	75	38	the	the	DET
fcis-28893	75	39	classification	classification	NOUN
fcis-28893	75	40	task	task	NOUN
fcis-28893	75	41	to	to	PART
fcis-28893	75	42	strengthen	strengthen	VERB
fcis-28893	75	43	the	the	DET
fcis-28893	75	44	connection	connection	NOUN
fcis-28893	75	45	between	between	ADP
fcis-28893	75	46	the	the	DET
fcis-28893	75	47	graphic	graphic	ADJ
fcis-28893	75	48	sample	sample	NOUN
fcis-28893	75	49	pairs	pair	NOUN
fcis-28893	75	50	.	.	PUNCT
fcis-28893	76	1	the	the	DET
fcis-28893	76	2	cmpm+cmpc	cmpm+cmpc	PROPN
fcis-28893	76	3	model	model	NOUN
fcis-28893	76	4	demonstrates	demonstrate	VERB
fcis-28893	76	5	great	great	ADJ
fcis-28893	76	6	stability	stability	NOUN
fcis-28893	76	7	and	and	CCONJ
fcis-28893	76	8	superiority	superiority	NOUN
fcis-28893	76	9	,	,	PUNCT
fcis-28893	76	10	which	which	PRON
fcis-28893	76	11	significantly	significantly	ADV
fcis-28893	76	12	strengthens	strengthen	VERB
fcis-28893	76	13	the	the	DET
fcis-28893	76	14	match	match	NOUN
fcis-28893	76	15	between	between	ADP
fcis-28893	76	16	the	the	DET
fcis-28893	76	17	features	feature	NOUN
fcis-28893	76	18	and	and	CCONJ
fcis-28893	76	19	the	the	DET
fcis-28893	76	20	individual	individual	ADJ
fcis-28893	76	21	categories	category	NOUN
fcis-28893	76	22	.	.	PUNCT
fcis-28893	77	1	further	far	ADV
fcis-28893	77	2	advancing	advance	VERB
fcis-28893	77	3	cross	cross	ADJ
fcis-28893	77	4	-	-	ADJ
fcis-28893	77	5	modal	modal	ADJ
fcis-28893	77	6	learning	learning	NOUN
fcis-28893	77	7	in	in	ADP
fcis-28893	77	8	text	text	NOUN
fcis-28893	77	9	-	-	PUNCT
fcis-28893	77	10	based	base	VERB
fcis-28893	77	11	pedestrian	pedestrian	NOUN
fcis-28893	77	12	retrieval	retrieval	NOUN
fcis-28893	77	13	tasks	task	NOUN
fcis-28893	77	14	.	.	PUNCT
fcis-28893	78	1	fig	fig	NOUN
fcis-28893	78	2	3	3	NUM
fcis-28893	78	3	.	.	PUNCT
fcis-28893	78	4	schematic	schematic	ADJ
fcis-28893	78	5	diagram	diagram	NOUN
fcis-28893	78	6	of	of	ADP
fcis-28893	78	7	the	the	DET
fcis-28893	78	8	cmpm+cmpc	cmpm+cmpc	PROPN
fcis-28893	78	9	network	network	NOUN
fcis-28893	78	10	framework	framework	NOUN
fcis-28893	78	11	4.2	4.2	NUM
fcis-28893	78	12	.	.	PUNCT
fcis-28893	79	1	method	method	NOUN
fcis-28893	79	2	based	base	VERB
fcis-28893	79	3	on	on	ADP
fcis-28893	79	4	multi	multi	ADJ
fcis-28893	79	5	-	-	NOUN
fcis-28893	79	6	granularity	granularity	NOUN
fcis-28893	79	7	jing	jing	NOUN
fcis-28893	79	8	et	et	PROPN
fcis-28893	79	9	al.[8	al.[8	PROPN
fcis-28893	79	10	]	]	PUNCT
fcis-28893	79	11	proposed	propose	VERB
fcis-28893	79	12	a	a	DET
fcis-28893	79	13	multi	multi	ADJ
fcis-28893	79	14	-	-	ADJ
fcis-28893	79	15	granularity	granularity	ADJ
fcis-28893	79	16	textual	textual	ADJ
fcis-28893	79	17	pedestrian	pedestrian	NOUN
fcis-28893	79	18	retrieval	retrieval	NOUN
fcis-28893	79	19	model	model	NOUN
fcis-28893	79	20	pma	pma	PROPN
fcis-28893	79	21	(	(	PUNCT
fcis-28893	79	22	pose	pose	NOUN
fcis-28893	79	23	-	-	PUNCT
fcis-28893	79	24	guided	guide	VERB
fcis-28893	79	25	multigranularity	multigranularity	NOUN
fcis-28893	79	26	attention	attention	NOUN
fcis-28893	79	27	network	network	NOUN
fcis-28893	79	28	)	)	PUNCT
fcis-28893	79	29	for	for	ADP
fcis-28893	79	30	fully	fully	ADV
fcis-28893	79	31	exploiting	exploit	VERB
fcis-28893	79	32	human	human	ADJ
fcis-28893	79	33	pose	pose	NOUN
fcis-28893	79	34	information	information	NOUN
fcis-28893	79	35	.	.	PUNCT
fcis-28893	80	1	the	the	DET
fcis-28893	80	2	architecture	architecture	NOUN
fcis-28893	80	3	of	of	ADP
fcis-28893	80	4	the	the	DET
fcis-28893	80	5	pma	pma	NOUN
fcis-28893	80	6	model	model	NOUN
fcis-28893	80	7	is	be	AUX
fcis-28893	80	8	shown	show	VERB
fcis-28893	80	9	in	in	ADP
fcis-28893	80	10	fig	fig	NOUN
fcis-28893	80	11	.	.	PUNCT
fcis-28893	81	1	4	4	X
fcis-28893	81	2	.	.	X
fcis-28893	81	3	the	the	DET
fcis-28893	81	4	model	model	NOUN
fcis-28893	81	5	adopts	adopt	VERB
fcis-28893	81	6	the	the	DET
fcis-28893	81	7	paf	paf	PROPN
fcis-28893	81	8	(	(	PUNCT
fcis-28893	81	9	part	part	NOUN
fcis-28893	81	10	affinity	affinity	NOUN
fcis-28893	81	11	fields	field	NOUN
fcis-28893	81	12	)	)	PUNCT
fcis-28893	81	13	method	method	NOUN
fcis-28893	81	14	for	for	ADP
fcis-28893	81	15	human	human	ADJ
fcis-28893	81	16	posture	posture	NOUN
fcis-28893	81	17	estimation	estimation	NOUN
fcis-28893	81	18	and	and	CCONJ
fcis-28893	81	19	combines	combine	VERB
fcis-28893	81	20	the	the	DET
fcis-28893	81	21	posture	posture	NOUN
fcis-28893	81	22	confidence	confidence	NOUN
fcis-28893	81	23	maps	map	NOUN
fcis-28893	81	24	with	with	ADP
fcis-28893	81	25	the	the	DET
fcis-28893	81	26	original	original	ADJ
fcis-28893	81	27	images	image	NOUN
fcis-28893	81	28	in	in	ADP
fcis-28893	81	29	the	the	DET
fcis-28893	81	30	evaluation	evaluation	NOUN
fcis-28893	81	31	stage	stage	NOUN
fcis-28893	81	32	,	,	PUNCT
fcis-28893	81	33	so	so	SCONJ
fcis-28893	81	34	that	that	SCONJ
fcis-28893	81	35	the	the	DET
fcis-28893	81	36	global	global	ADJ
fcis-28893	81	37	visual	visual	ADJ
fcis-28893	81	38	features	feature	NOUN
fcis-28893	81	39	are	be	AUX
fcis-28893	81	40	more	more	ADV
fcis-28893	81	41	focused	focused	ADJ
fcis-28893	81	42	on	on	ADP
fcis-28893	81	43	the	the	DET
fcis-28893	81	44	regions	region	NOUN
fcis-28893	81	45	related	relate	VERB
fcis-28893	81	46	to	to	ADP
fcis-28893	81	47	the	the	DET
fcis-28893	81	48	human	human	ADJ
fcis-28893	81	49	body	body	NOUN
fcis-28893	81	50	.	.	PUNCT
fcis-28893	82	1	the	the	DET
fcis-28893	82	2	role	role	NOUN
fcis-28893	82	3	of	of	ADP
fcis-28893	82	4	the	the	DET
fcis-28893	82	5	pose	pose	NOUN
fcis-28893	82	6	confidence	confidence	NOUN
fcis-28893	82	7	maps	map	NOUN
fcis-28893	82	8	is	be	AUX
fcis-28893	82	9	mainly	mainly	ADV
fcis-28893	82	10	reflected	reflect	VERB
fcis-28893	82	11	in	in	ADP
fcis-28893	82	12	two	two	NUM
fcis-28893	82	13	aspects	aspect	NOUN
fcis-28893	82	14	:	:	PUNCT
fcis-28893	82	15	on	on	ADP
fcis-28893	82	16	the	the	DET
fcis-28893	82	17	one	one	NUM
fcis-28893	82	18	hand	hand	NOUN
fcis-28893	82	19	,	,	PUNCT
fcis-28893	82	20	the	the	DET
fcis-28893	82	21	confidence	confidence	NOUN
fcis-28893	82	22	maps	map	NOUN
fcis-28893	82	23	work	work	VERB
fcis-28893	82	24	together	together	ADV
fcis-28893	82	25	with	with	ADP
fcis-28893	82	26	the	the	DET
fcis-28893	82	27	input	input	NOUN
fcis-28893	82	28	images	image	NOUN
fcis-28893	82	29	to	to	PART
fcis-28893	82	30	enhance	enhance	VERB
fcis-28893	82	31	the	the	DET
fcis-28893	82	32	representation	representation	NOUN
fcis-28893	82	33	of	of	ADP
fcis-28893	82	34	visual	visual	ADJ
fcis-28893	82	35	features	feature	NOUN
fcis-28893	82	36	;	;	PUNCT
fcis-28893	82	37	on	on	ADP
fcis-28893	82	38	the	the	DET
fcis-28893	82	39	other	other	ADJ
fcis-28893	82	40	hand	hand	NOUN
fcis-28893	82	41	,	,	PUNCT
fcis-28893	82	42	the	the	DET
fcis-28893	82	43	confidence	confidence	NOUN
fcis-28893	82	44	maps	map	NOUN
fcis-28893	82	45	are	be	AUX
fcis-28893	82	46	used	use	VERB
fcis-28893	82	47	to	to	PART
fcis-28893	82	48	learn	learn	VERB
fcis-28893	82	49	the	the	DET
fcis-28893	82	50	potential	potential	ADJ
fcis-28893	82	51	semantic	semantic	ADJ
fcis-28893	82	52	alignment	alignment	NOUN
fcis-28893	82	53	relationship	relationship	NOUN
fcis-28893	82	54	between	between	ADP
fcis-28893	82	55	noun	noun	NOUN
fcis-28893	82	56	phrases	phrase	NOUN
fcis-28893	82	57	and	and	CCONJ
fcis-28893	82	58	image	image	NOUN
fcis-28893	82	59	regions	region	NOUN
fcis-28893	82	60	.	.	PUNCT
fcis-28893	83	1	pma	pma	PROPN
fcis-28893	83	2	also	also	ADV
fcis-28893	83	3	contains	contain	VERB
fcis-28893	83	4	two	two	NUM
fcis-28893	83	5	parts	part	NOUN
fcis-28893	83	6	,	,	PUNCT
fcis-28893	83	7	the	the	DET
fcis-28893	83	8	coarse	coarse	ADV
fcis-28893	83	9	-	-	PUNCT
fcis-28893	83	10	grained	grain	VERB
fcis-28893	83	11	alignment	alignment	NOUN
fcis-28893	83	12	network	network	NOUN
fcis-28893	83	13	and	and	CCONJ
fcis-28893	83	14	the	the	DET
fcis-28893	83	15	fine	fine	ADV
fcis-28893	83	16	-	-	PUNCT
fcis-28893	83	17	grained	grain	VERB
fcis-28893	83	18	network	network	NOUN
fcis-28893	83	19	alignment	alignment	NOUN
fcis-28893	83	20	network	network	NOUN
fcis-28893	83	21	.	.	PUNCT
fcis-28893	84	1	the	the	DET
fcis-28893	84	2	coarse	coarse	ADV
fcis-28893	84	3	-	-	PUNCT
fcis-28893	84	4	grained	grain	VERB
fcis-28893	84	5	network	network	NOUN
fcis-28893	84	6	selects	select	VERB
fcis-28893	84	7	image	image	NOUN
fcis-28893	84	8	parts	part	NOUN
fcis-28893	84	9	that	that	PRON
fcis-28893	84	10	are	be	AUX
fcis-28893	84	11	motivated	motivate	VERB
fcis-28893	84	12	by	by	ADP
fcis-28893	84	13	the	the	DET
fcis-28893	84	14	highest	high	ADJ
fcis-28893	84	15	correlation	correlation	NOUN
fcis-28893	84	16	with	with	ADP
fcis-28893	84	17	the	the	DET
fcis-28893	84	18	text	text	NOUN
fcis-28893	84	19	parts	part	NOUN
fcis-28893	84	20	,	,	PUNCT
fcis-28893	84	21	and	and	CCONJ
fcis-28893	84	22	the	the	DET
fcis-28893	84	23	fine	fine	ADV
fcis-28893	84	24	-	-	PUNCT
fcis-28893	84	25	grained	grain	VERB
fcis-28893	84	26	network	network	NOUN
fcis-28893	84	27	further	far	ADV
fcis-28893	84	28	learns	learn	VERB
fcis-28893	84	29	the	the	DET
fcis-28893	84	30	potential	potential	ADJ
fcis-28893	84	31	semantic	semantic	ADJ
fcis-28893	84	32	links	link	NOUN
fcis-28893	84	33	between	between	ADP
fcis-28893	84	34	noun	noun	NOUN
fcis-28893	84	35	phrases	phrase	NOUN
fcis-28893	84	36	and	and	CCONJ
fcis-28893	84	37	image	image	NOUN
fcis-28893	84	38	regions	region	NOUN
fcis-28893	84	39	to	to	PART
fcis-28893	84	40	extract	extract	VERB
fcis-28893	84	41	the	the	DET
fcis-28893	84	42	relevant	relevant	ADJ
fcis-28893	84	43	image	image	NOUN
fcis-28893	84	44	content	content	NOUN
fcis-28893	84	45	in	in	ADP
fcis-28893	84	46	the	the	DET
fcis-28893	84	47	phrase	phrase	NOUN
fcis-28893	84	48	dimension	dimension	NOUN
fcis-28893	84	49	.	.	PUNCT
fcis-28893	85	1	the	the	DET
fcis-28893	85	2	pma	pma	PROPN
fcis-28893	85	3	uses	use	VERB
fcis-28893	85	4	triplet	triplet	NOUN
fcis-28893	85	5	ranking	rank	VERB
fcis-28893	85	6	loss	loss	NOUN
fcis-28893	85	7	and	and	CCONJ
fcis-28893	85	8	introduces	introduce	VERB
fcis-28893	85	9	the	the	DET
fcis-28893	85	10	identification	identification	NOUN
fcis-28893	85	11	loss	loss	NOUN
fcis-28893	85	12	,	,	PUNCT
fcis-28893	85	13	which	which	PRON
fcis-28893	85	14	ensures	ensure	VERB
fcis-28893	85	15	that	that	SCONJ
fcis-28893	85	16	the	the	DET
fcis-28893	85	17	pedestrian	pedestrian	NOUN
fcis-28893	85	18	features	feature	NOUN
fcis-28893	85	19	can	can	AUX
fcis-28893	85	20	be	be	AUX
fcis-28893	85	21	grouped	group	VERB
fcis-28893	85	22	together	together	ADV
fcis-28893	85	23	to	to	PART
fcis-28893	85	24	ensure	ensure	VERB
fcis-28893	85	25	the	the	DET
fcis-28893	85	26	accurate	accurate	ADJ
fcis-28893	85	27	matching	matching	NOUN
fcis-28893	85	28	at	at	ADP
fcis-28893	85	29	the	the	DET
fcis-28893	85	30	individual	individual	ADJ
fcis-28893	85	31	level	level	NOUN
fcis-28893	85	32	.	.	PUNCT
fcis-28893	86	1	pma	pma	NOUN
fcis-28893	86	2	enhances	enhance	VERB
fcis-28893	86	3	the	the	DET
fcis-28893	86	4	potential	potential	ADJ
fcis-28893	86	5	semantic	semantic	ADJ
fcis-28893	86	6	alignment	alignment	NOUN
fcis-28893	86	7	of	of	ADP
fcis-28893	86	8	images	image	NOUN
fcis-28893	86	9	and	and	CCONJ
fcis-28893	86	10	texts	text	NOUN
fcis-28893	86	11	by	by	ADP
fcis-28893	86	12	fully	fully	ADV
fcis-28893	86	13	utilizing	utilize	VERB
fcis-28893	86	14	the	the	DET
fcis-28893	86	15	human	human	ADJ
fcis-28893	86	16	posture	posture	NOUN
fcis-28893	86	17	information	information	NOUN
fcis-28893	86	18	through	through	ADP
fcis-28893	86	19	the	the	DET
fcis-28893	86	20	posture	posture	NOUN
fcis-28893	86	21	confidence	confidence	NOUN
fcis-28893	86	22	map	map	NOUN
fcis-28893	86	23	;	;	PUNCT
fcis-28893	86	24	learns	learn	VERB
fcis-28893	86	25	the	the	DET
fcis-28893	86	26	semantic	semantic	ADJ
fcis-28893	86	27	information	information	NOUN
fcis-28893	86	28	of	of	ADP
fcis-28893	86	29	different	different	ADJ
fcis-28893	86	30	granularity	granularity	NOUN
fcis-28893	86	31	in	in	ADP
fcis-28893	86	32	images	image	NOUN
fcis-28893	86	33	and	and	CCONJ
fcis-28893	86	34	texts	text	NOUN
fcis-28893	86	35	by	by	ADP
fcis-28893	86	36	designing	design	VERB
fcis-28893	86	37	coarse	coarse	NOUN
fcis-28893	86	38	-	-	PUNCT
fcis-28893	86	39	grained	grain	VERB
fcis-28893	86	40	and	and	CCONJ
fcis-28893	86	41	fine	fine	ADV
fcis-28893	86	42	-	-	PUNCT
fcis-28893	86	43	grained	grain	VERB
fcis-28893	86	44	alignment	alignment	NOUN
fcis-28893	86	45	networks	network	NOUN
fcis-28893	86	46	;	;	PUNCT
fcis-28893	86	47	and	and	CCONJ
fcis-28893	86	48	achieves	achieve	VERB
fcis-28893	86	49	high	high	ADJ
fcis-28893	86	50	-	-	PUNCT
fcis-28893	86	51	precision	precision	NOUN
fcis-28893	86	52	retrieval	retrieval	NOUN
fcis-28893	86	53	by	by	ADP
fcis-28893	86	54	using	use	VERB
fcis-28893	86	55	triplet	triplet	NOUN
fcis-28893	86	56	ranking	ranking	NOUN
fcis-28893	86	57	and	and	CCONJ
fcis-28893	86	58	identification	identification	NOUN
fcis-28893	86	59	loss	loss	NOUN
fcis-28893	86	60	.	.	PUNCT
fcis-28893	87	1	niu	niu	PROPN
fcis-28893	87	2	et	et	PROPN
fcis-28893	87	3	al.[9	al.[9	PROPN
fcis-28893	87	4	]	]	PUNCT
fcis-28893	87	5	argued	argue	VERB
fcis-28893	87	6	that	that	SCONJ
fcis-28893	87	7	the	the	DET
fcis-28893	87	8	key	key	NOUN
fcis-28893	87	9	to	to	ADP
fcis-28893	87	10	cross	cross	ADJ
fcis-28893	87	11	-	-	ADJ
fcis-28893	87	12	modal	modal	ADJ
fcis-28893	87	13	learning	learning	NOUN
fcis-28893	87	14	is	be	AUX
fcis-28893	87	15	adaptive	adaptive	ADJ
fcis-28893	87	16	local	local	ADJ
fcis-28893	87	17	alignment	alignment	NOUN
fcis-28893	87	18	and	and	CCONJ
fcis-28893	87	19	the	the	DET
fcis-28893	87	20	potential	potential	ADJ
fcis-28893	87	21	relationship	relationship	NOUN
fcis-28893	87	22	between	between	ADP
fcis-28893	87	23	local	local	ADJ
fcis-28893	87	24	and	and	CCONJ
fcis-28893	87	25	global	global	ADJ
fcis-28893	87	26	,	,	PUNCT
fcis-28893	87	27	and	and	CCONJ
fcis-28893	87	28	in	in	ADP
fcis-28893	87	29	order	order	NOUN
fcis-28893	87	30	to	to	PART
fcis-28893	87	31	address	address	VERB
fcis-28893	87	32	these	these	DET
fcis-28893	87	33	issues	issue	NOUN
fcis-28893	87	34	,	,	PUNCT
fcis-28893	87	35	niu	niu	PROPN
fcis-28893	87	36	et	et	PROPN
fcis-28893	87	37	al	al	PROPN
fcis-28893	87	38	.	.	PROPN
fcis-28893	87	39	proposed	propose	VERB
fcis-28893	87	40	a	a	DET
fcis-28893	87	41	mia	mia	PROPN
fcis-28893	87	42	model	model	NOUN
fcis-28893	87	43	consisting	consist	VERB
fcis-28893	87	44	of	of	ADP
fcis-28893	87	45	two	two	NUM
fcis-28893	87	46	parts	part	NOUN
fcis-28893	87	47	:	:	PUNCT
fcis-28893	87	48	global	global	ADJ
fcis-28893	87	49	and	and	CCONJ
fcis-28893	87	50	local	local	ADJ
fcis-28893	87	51	feature	feature	NOUN
fcis-28893	87	52	extraction	extraction	NOUN
fcis-28893	87	53	and	and	CCONJ
fcis-28893	87	54	graphic	graphic	ADJ
fcis-28893	87	55	and	and	CCONJ
fcis-28893	87	56	textual	textual	ADJ
fcis-28893	87	57	alignment	alignment	NOUN
fcis-28893	87	58	at	at	ADP
fcis-28893	87	59	multiple	multiple	ADJ
fcis-28893	87	60	granularities	granularity	NOUN
fcis-28893	87	61	.	.	PUNCT
fcis-28893	88	1	cnn	cnn	PROPN
fcis-28893	88	2	networks	network	NOUN
fcis-28893	88	3	and	and	CCONJ
fcis-28893	88	4	bidirectional	bidirectional	ADJ
fcis-28893	88	5	gated	gate	VERB
fcis-28893	88	6	recurrent	recurrent	ADJ
fcis-28893	88	7	units	unit	NOUN
fcis-28893	88	8	(	(	PUNCT
fcis-28893	88	9	bi	bi	NOUN
fcis-28893	88	10	-	-	NOUN
fcis-28893	88	11	grus	grus	NOUN
fcis-28893	88	12	)	)	PUNCT
fcis-28893	88	13	are	be	AUX
fcis-28893	88	14	utilized	utilize	VERB
fcis-28893	88	15	in	in	ADP
fcis-28893	88	16	global	global	ADJ
fcis-28893	88	17	and	and	CCONJ
fcis-28893	88	18	local	local	ADJ
fcis-28893	88	19	feature	feature	NOUN
fcis-28893	88	20	extraction	extraction	NOUN
fcis-28893	88	21	to	to	PART
fcis-28893	88	22	extract	extract	VERB
fcis-28893	88	23	visual	visual	ADJ
fcis-28893	88	24	feature	feature	NOUN
fcis-28893	88	25	maps	map	NOUN
fcis-28893	88	26	and	and	CCONJ
fcis-28893	88	27	encode	encode	ADJ
fcis-28893	88	28	sentences	sentence	NOUN
fcis-28893	88	29	and	and	CCONJ
fcis-28893	88	30	phrases	phrase	NOUN
fcis-28893	88	31	,	,	PUNCT
fcis-28893	88	32	respectively	respectively	ADV
fcis-28893	88	33	.	.	PUNCT
fcis-28893	89	1	the	the	DET
fcis-28893	89	2	multi	multi	ADJ
fcis-28893	89	3	-	-	ADJ
fcis-28893	89	4	granularity	granularity	NOUN
fcis-28893	89	5	graph	graph	NOUN
fcis-28893	89	6	alignment	alignment	NOUN
fcis-28893	89	7	part	part	NOUN
fcis-28893	89	8	contains	contain	VERB
fcis-28893	89	9	three	three	NUM
fcis-28893	89	10	granularity	granularity	NOUN
fcis-28893	89	11	alignment	alignment	NOUN
fcis-28893	89	12	modules	module	NOUN
fcis-28893	89	13	:	:	PUNCT
fcis-28893	89	14	the	the	DET
fcis-28893	89	15	global	global	ADJ
fcis-28893	89	16	comparison	comparison	NOUN
fcis-28893	89	17	module	module	NOUN
fcis-28893	89	18	(	(	PUNCT
fcis-28893	89	19	gc	gc	PROPN
fcis-28893	89	20	)	)	PUNCT
fcis-28893	89	21	,	,	PUNCT
fcis-28893	89	22	the	the	DET
fcis-28893	89	23	relation	relation	NOUN
fcis-28893	89	24	-	-	PUNCT
fcis-28893	89	25	guided	guide	VERB
fcis-28893	89	26	global	global	ADJ
fcis-28893	89	27	-	-	PUNCT
fcis-28893	89	28	local	local	ADJ
fcis-28893	89	29	alignment	alignment	NOUN
fcis-28893	89	30	module	module	NOUN
fcis-28893	89	31	(	(	PUNCT
fcis-28893	89	32	rga	rga	PROPN
fcis-28893	89	33	)	)	PUNCT
fcis-28893	89	34	,	,	PUNCT
fcis-28893	89	35	and	and	CCONJ
fcis-28893	89	36	the	the	DET
fcis-28893	89	37	bidirectional	bidirectional	ADJ
fcis-28893	89	38	fine	fine	ADV
fcis-28893	89	39	-	-	PUNCT
fcis-28893	89	40	grained	grain	VERB
fcis-28893	89	41	matching	matching	NOUN
fcis-28893	89	42	module	module	NOUN
fcis-28893	89	43	(	(	PUNCT
fcis-28893	89	44	bfm	bfm	PROPN
fcis-28893	89	45	)	)	PUNCT
fcis-28893	89	46	.	.	PUNCT
fcis-28893	90	1	in	in	ADP
fcis-28893	90	2	the	the	DET
fcis-28893	90	3	rga	rga	NOUN
fcis-28893	90	4	module	module	NOUN
fcis-28893	90	5	,	,	PUNCT
fcis-28893	90	6	the	the	DET
fcis-28893	90	7	feature	feature	NOUN
fcis-28893	90	8	map	map	NOUN
fcis-28893	90	9	is	be	AUX
fcis-28893	90	10	uniformly	uniformly	ADV
fcis-28893	90	11	segmented	segment	VERB
fcis-28893	90	12	into	into	ADP
fcis-28893	90	13	multiple	multiple	ADJ
fcis-28893	90	14	non	non	ADJ
fcis-28893	90	15	-	-	ADJ
fcis-28893	90	16	overlapping	overlapping	ADJ
fcis-28893	90	17	regions	region	NOUN
fcis-28893	90	18	along	along	ADP
fcis-28893	90	19	the	the	DET
fcis-28893	90	20	vertical	vertical	ADJ
fcis-28893	90	21	direction[10	direction[10	X
fcis-28893	90	22	]	]	PUNCT
fcis-28893	90	23	,	,	PUNCT
fcis-28893	90	24	and	and	CCONJ
fcis-28893	90	25	this	this	DET
fcis-28893	90	26	segmentation	segmentation	NOUN
fcis-28893	90	27	method	method	NOUN
fcis-28893	90	28	does	do	AUX
fcis-28893	90	29	not	not	PART
fcis-28893	90	30	incur	incur	VERB
fcis-28893	90	31	additional	additional	ADJ
fcis-28893	90	32	overhead	overhead	NOUN
fcis-28893	90	33	,	,	PUNCT
fcis-28893	90	34	and	and	CCONJ
fcis-28893	90	35	is	be	AUX
fcis-28893	90	36	performed	perform	VERB
fcis-28893	90	37	in	in	ADP
fcis-28893	90	38	parallel	parallel	NOUN
fcis-28893	90	39	with	with	ADP
fcis-28893	90	40	the	the	DET
fcis-28893	90	41	training	training	NOUN
fcis-28893	90	42	process	process	NOUN
fcis-28893	90	43	,	,	PUNCT
fcis-28893	90	44	which	which	PRON
fcis-28893	90	45	realizes	realize	VERB
fcis-28893	90	46	end	end	NOUN
fcis-28893	90	47	-	-	PUNCT
fcis-28893	90	48	toend	toend	NOUN
fcis-28893	90	49	training	training	NOUN
fcis-28893	90	50	.	.	PUNCT
fcis-28893	91	1	the	the	DET
fcis-28893	91	2	mia	mia	PROPN
fcis-28893	91	3	model	model	NOUN
fcis-28893	91	4	is	be	AUX
fcis-28893	91	5	the	the	DET
fcis-28893	91	6	first	first	ADJ
fcis-28893	91	7	of	of	ADP
fcis-28893	91	8	its	its	PRON
fcis-28893	91	9	kind	kind	NOUN
fcis-28893	91	10	that	that	SCONJ
fcis-28893	91	11	122	122	NUM
fcis-28893	91	12	combines	combine	VERB
fcis-28893	91	13	the	the	DET
fcis-28893	91	14	local	local	ADJ
fcis-28893	91	15	features	feature	NOUN
fcis-28893	91	16	of	of	ADP
fcis-28893	91	17	human	human	ADJ
fcis-28893	91	18	body	body	NOUN
fcis-28893	91	19	with	with	ADP
fcis-28893	91	20	noun	noun	NOUN
fcis-28893	91	21	phrases	phrase	NOUN
fcis-28893	91	22	for	for	ADP
fcis-28893	91	23	the	the	DET
fcis-28893	91	24	text	text	NOUN
fcis-28893	91	25	-	-	PUNCT
fcis-28893	91	26	based	base	VERB
fcis-28893	91	27	pedestrian	pedestrian	NOUN
fcis-28893	91	28	re	re	NOUN
fcis-28893	91	29	-	-	NOUN
fcis-28893	91	30	recognition	recognition	NOUN
fcis-28893	91	31	task	task	NOUN
fcis-28893	91	32	,	,	PUNCT
fcis-28893	91	33	which	which	PRON
fcis-28893	91	34	opens	open	VERB
fcis-28893	91	35	a	a	DET
fcis-28893	91	36	new	new	ADJ
fcis-28893	91	37	research	research	NOUN
fcis-28893	91	38	direction	direction	NOUN
fcis-28893	91	39	in	in	ADP
fcis-28893	91	40	this	this	DET
fcis-28893	91	41	field	field	NOUN
fcis-28893	91	42	of	of	ADP
fcis-28893	91	43	textual	textual	ADJ
fcis-28893	91	44	pedestrian	pedestrian	NOUN
fcis-28893	91	45	retrieval	retrieval	NOUN
fcis-28893	91	46	tasks	task	NOUN
fcis-28893	91	47	.	.	PUNCT
fcis-28893	92	1	the	the	DET
fcis-28893	92	2	mia	mia	PROPN
fcis-28893	92	3	model	model	NOUN
fcis-28893	92	4	is	be	AUX
fcis-28893	92	5	the	the	DET
fcis-28893	92	6	first	first	ADJ
fcis-28893	92	7	to	to	PART
fcis-28893	92	8	combine	combine	VERB
fcis-28893	92	9	human	human	ADJ
fcis-28893	92	10	local	local	ADJ
fcis-28893	92	11	features	feature	NOUN
fcis-28893	92	12	with	with	ADP
fcis-28893	92	13	noun	noun	NOUN
fcis-28893	92	14	phrases	phrase	NOUN
fcis-28893	92	15	for	for	ADP
fcis-28893	92	16	the	the	DET
fcis-28893	92	17	text	text	NOUN
fcis-28893	92	18	-	-	PUNCT
fcis-28893	92	19	based	base	VERB
fcis-28893	92	20	pedestrian	pedestrian	NOUN
fcis-28893	92	21	re	re	NOUN
fcis-28893	92	22	-	-	NOUN
fcis-28893	92	23	recognition	recognition	NOUN
fcis-28893	92	24	task	task	NOUN
fcis-28893	92	25	,	,	PUNCT
fcis-28893	92	26	opening	open	VERB
fcis-28893	92	27	a	a	DET
fcis-28893	92	28	new	new	ADJ
fcis-28893	92	29	research	research	NOUN
fcis-28893	92	30	direction	direction	NOUN
fcis-28893	92	31	for	for	ADP
fcis-28893	92	32	the	the	DET
fcis-28893	92	33	text	text	NOUN
fcis-28893	92	34	-	-	PUNCT
fcis-28893	92	35	based	base	VERB
fcis-28893	92	36	pedestrian	pedestrian	NOUN
fcis-28893	92	37	retrieval	retrieval	NOUN
fcis-28893	92	38	task	task	NOUN
fcis-28893	92	39	in	in	ADP
fcis-28893	92	40	this	this	DET
fcis-28893	92	41	area	area	NOUN
fcis-28893	92	42	.	.	PUNCT
fcis-28893	93	1	fig	fig	NOUN
fcis-28893	93	2	4	4	NUM
fcis-28893	93	3	.	.	PUNCT
fcis-28893	94	1	pma	pma	NOUN
fcis-28893	94	2	network	network	NOUN
fcis-28893	94	3	framework	framework	NOUN
fcis-28893	94	4	wang	wang	PROPN
fcis-28893	94	5	et	et	PROPN
fcis-28893	94	6	al.[11	al.[11	PROPN
fcis-28893	94	7	]	]	PUNCT
fcis-28893	94	8	pointed	point	VERB
fcis-28893	94	9	out	out	ADP
fcis-28893	94	10	that	that	SCONJ
fcis-28893	94	11	previous	previous	ADJ
fcis-28893	94	12	approaches	approach	NOUN
fcis-28893	94	13	viewed	view	VERB
fcis-28893	94	14	the	the	DET
fcis-28893	94	15	text	text	NOUN
fcis-28893	94	16	-	-	PUNCT
fcis-28893	94	17	based	base	VERB
fcis-28893	94	18	pedestrian	pedestrian	NOUN
fcis-28893	94	19	retrieval	retrieval	NOUN
fcis-28893	94	20	task	task	NOUN
fcis-28893	94	21	as	as	ADP
fcis-28893	94	22	a	a	DET
fcis-28893	94	23	holistic	holistic	ADJ
fcis-28893	94	24	graphical	graphical	ADJ
fcis-28893	94	25	feature	feature	NOUN
fcis-28893	94	26	matching	matching	NOUN
fcis-28893	94	27	.	.	PUNCT
fcis-28893	95	1	wang	wang	PROPN
fcis-28893	95	2	et	et	PROPN
fcis-28893	95	3	al	al	PROPN
fcis-28893	95	4	.	.	PROPN
fcis-28893	95	5	instead	instead	ADV
fcis-28893	95	6	adopted	adopt	VERB
fcis-28893	95	7	an	an	DET
fcis-28893	95	8	attribute	attribute	NOUN
fcis-28893	95	9	-	-	PUNCT
fcis-28893	95	10	alignment	alignment	NOUN
fcis-28893	95	11	-	-	PUNCT
fcis-28893	95	12	based	base	VERB
fcis-28893	95	13	perspective	perspective	NOUN
fcis-28893	95	14	on	on	ADP
fcis-28893	95	15	the	the	DET
fcis-28893	95	16	task	task	NOUN
fcis-28893	95	17	.	.	PUNCT
fcis-28893	96	1	based	base	VERB
fcis-28893	96	2	on	on	ADP
fcis-28893	96	3	this	this	DET
fcis-28893	96	4	motivation	motivation	NOUN
fcis-28893	96	5	,	,	PUNCT
fcis-28893	96	6	wang	wang	PROPN
fcis-28893	96	7	et	et	PROPN
fcis-28893	96	8	al	al	PROPN
fcis-28893	96	9	.	.	PROPN
fcis-28893	96	10	proposed	propose	VERB
fcis-28893	96	11	the	the	DET
fcis-28893	96	12	attribute	attribute	NOUN
fcis-28893	96	13	feature	feature	NOUN
fcis-28893	96	14	alignment	alignment	NOUN
fcis-28893	96	15	retrieval	retrieval	NOUN
fcis-28893	96	16	model	model	NOUN
fcis-28893	96	17	vitaa	vitaa	PROPN
fcis-28893	96	18	.	.	PUNCT
fcis-28893	97	1	vitaa	vitaa	PROPN
fcis-28893	97	2	contains	contain	VERB
fcis-28893	97	3	a	a	DET
fcis-28893	97	4	global	global	ADJ
fcis-28893	97	5	feature	feature	NOUN
fcis-28893	97	6	branch	branch	NOUN
fcis-28893	97	7	and	and	CCONJ
fcis-28893	97	8	several	several	ADJ
fcis-28893	97	9	local	local	ADJ
fcis-28893	97	10	branches	branch	NOUN
fcis-28893	97	11	for	for	ADP
fcis-28893	97	12	generating	generate	VERB
fcis-28893	97	13	global	global	ADJ
fcis-28893	97	14	visual	visual	ADJ
fcis-28893	97	15	features	feature	NOUN
fcis-28893	97	16	and	and	CCONJ
fcis-28893	97	17	attribute	attribute	VERB
fcis-28893	97	18	visual	visual	ADJ
fcis-28893	97	19	features	feature	NOUN
fcis-28893	97	20	.	.	PUNCT
fcis-28893	98	1	each	each	DET
fcis-28893	98	2	local	local	ADJ
fcis-28893	98	3	branch	branch	NOUN
fcis-28893	98	4	is	be	AUX
fcis-28893	98	5	also	also	ADV
fcis-28893	98	6	equipped	equip	VERB
fcis-28893	98	7	with	with	ADP
fcis-28893	98	8	a	a	DET
fcis-28893	98	9	lightweight	lightweight	ADJ
fcis-28893	98	10	maskhead	maskhead	ADJ
fcis-28893	98	11	architecture	architecture	NOUN
fcis-28893	98	12	to	to	PART
fcis-28893	98	13	assist	assist	VERB
fcis-28893	98	14	in	in	ADP
fcis-28893	98	15	generating	generate	VERB
fcis-28893	98	16	segmentation	segmentation	NOUN
fcis-28893	98	17	maps	map	NOUN
fcis-28893	98	18	for	for	ADP
fcis-28893	98	19	specific	specific	ADJ
fcis-28893	98	20	attribute	attribute	NOUN
fcis-28893	98	21	categories	category	NOUN
fcis-28893	98	22	.	.	PUNCT
fcis-28893	99	1	the	the	DET
fcis-28893	99	2	text	text	NOUN
fcis-28893	99	3	part	part	NOUN
fcis-28893	99	4	first	first	ADJ
fcis-28893	99	5	parses	parse	NOUN
fcis-28893	99	6	and	and	CCONJ
fcis-28893	99	7	extracts	extract	NOUN
fcis-28893	99	8	attribute	attribute	NOUN
fcis-28893	99	9	-	-	PUNCT
fcis-28893	99	10	related	relate	VERB
fcis-28893	99	11	noun	noun	NOUN
fcis-28893	99	12	phrases	phrase	NOUN
fcis-28893	99	13	,	,	PUNCT
fcis-28893	99	14	then	then	ADV
fcis-28893	99	15	generates	generate	VERB
fcis-28893	99	16	global	global	ADJ
fcis-28893	99	17	and	and	CCONJ
fcis-28893	99	18	local	local	ADJ
fcis-28893	99	19	features	feature	NOUN
fcis-28893	99	20	of	of	ADP
fcis-28893	99	21	the	the	DET
fcis-28893	99	22	text	text	NOUN
fcis-28893	99	23	through	through	ADP
fcis-28893	99	24	a	a	DET
fcis-28893	99	25	bidirectional	bidirectional	ADJ
fcis-28893	99	26	lstm	lstm	NOUN
fcis-28893	99	27	network	network	NOUN
fcis-28893	99	28	,	,	PUNCT
fcis-28893	99	29	and	and	CCONJ
fcis-28893	99	30	further	far	ADV
fcis-28893	99	31	employs	employ	VERB
fcis-28893	99	32	a	a	DET
fcis-28893	99	33	contrast	contrast	NOUN
fcis-28893	99	34	learning	learn	VERB
fcis-28893	99	35	approach	approach	NOUN
fcis-28893	99	36	to	to	PART
fcis-28893	99	37	achieve	achieve	VERB
fcis-28893	99	38	cross	cross	ADJ
fcis-28893	99	39	-	-	ADJ
fcis-28893	99	40	modal	modal	ADJ
fcis-28893	99	41	alignment	alignment	NOUN
fcis-28893	99	42	of	of	ADP
fcis-28893	99	43	the	the	DET
fcis-28893	99	44	graph	graph	NOUN
fcis-28893	99	45	and	and	CCONJ
fcis-28893	99	46	text	text	NOUN
fcis-28893	99	47	.	.	PUNCT
fcis-28893	100	1	since	since	SCONJ
fcis-28893	100	2	the	the	DET
fcis-28893	100	3	text	text	NOUN
fcis-28893	100	4	description	description	NOUN
fcis-28893	100	5	usually	usually	ADV
fcis-28893	100	6	targets	target	VERB
fcis-28893	100	7	the	the	DET
fcis-28893	100	8	appearance	appearance	NOUN
fcis-28893	100	9	and	and	CCONJ
fcis-28893	100	10	dress	dress	NOUN
fcis-28893	100	11	of	of	ADP
fcis-28893	100	12	pedestrians	pedestrian	NOUN
fcis-28893	100	13	,	,	PUNCT
fcis-28893	100	14	which	which	PRON
fcis-28893	100	15	are	be	AUX
fcis-28893	100	16	highly	highly	ADV
fcis-28893	100	17	consistent	consistent	ADJ
fcis-28893	100	18	with	with	ADP
fcis-28893	100	19	the	the	DET
fcis-28893	100	20	attribute	attribute	NOUN
fcis-28893	100	21	features	feature	VERB
fcis-28893	100	22	,	,	PUNCT
fcis-28893	100	23	the	the	DET
fcis-28893	100	24	mismatch	mismatch	NOUN
fcis-28893	100	25	due	due	ADP
fcis-28893	100	26	to	to	ADP
fcis-28893	100	27	the	the	DET
fcis-28893	100	28	similarity	similarity	NOUN
fcis-28893	100	29	of	of	ADP
fcis-28893	100	30	pedestrians	pedestrian	NOUN
fcis-28893	100	31	'	'	PART
fcis-28893	100	32	dress	dress	NOUN
fcis-28893	100	33	is	be	AUX
fcis-28893	100	34	effectively	effectively	ADV
fcis-28893	100	35	reduced	reduce	VERB
fcis-28893	100	36	.	.	PUNCT
fcis-28893	101	1	with	with	ADP
fcis-28893	101	2	these	these	DET
fcis-28893	101	3	advantages	advantage	NOUN
fcis-28893	101	4	,	,	PUNCT
fcis-28893	101	5	vitaa	vitaa	PRON
fcis-28893	101	6	significantly	significantly	ADV
fcis-28893	101	7	improves	improve	VERB
fcis-28893	101	8	the	the	DET
fcis-28893	101	9	sota	sota	NOUN
fcis-28893	101	10	scores	score	NOUN
fcis-28893	101	11	on	on	ADP
fcis-28893	101	12	the	the	DET
fcis-28893	101	13	cuhkpedes	cuhkpede	NOUN
fcis-28893	101	14	dataset	dataset	VERB
fcis-28893	101	15	.	.	PUNCT
fcis-28893	102	1	ding	de	VERB
fcis-28893	102	2	et	et	PROPN
fcis-28893	102	3	al.[2	al.[2	PROPN
fcis-28893	102	4	]	]	PUNCT
fcis-28893	102	5	argued	argue	VERB
fcis-28893	102	6	that	that	SCONJ
fcis-28893	102	7	semantic	semantic	ADJ
fcis-28893	102	8	information	information	NOUN
fcis-28893	102	9	alignment	alignment	NOUN
fcis-28893	102	10	is	be	AUX
fcis-28893	102	11	very	very	ADV
fcis-28893	102	12	effective	effective	ADJ
fcis-28893	102	13	in	in	ADP
fcis-28893	102	14	bridging	bridge	VERB
fcis-28893	102	15	the	the	DET
fcis-28893	102	16	gap	gap	NOUN
fcis-28893	102	17	between	between	ADP
fcis-28893	102	18	image	image	NOUN
fcis-28893	102	19	and	and	CCONJ
fcis-28893	102	20	text	text	NOUN
fcis-28893	102	21	modalities	modality	NOUN
fcis-28893	102	22	,	,	PUNCT
fcis-28893	102	23	and	and	CCONJ
fcis-28893	102	24	therefore	therefore	ADV
fcis-28893	102	25	proposed	propose	VERB
fcis-28893	102	26	the	the	DET
fcis-28893	102	27	semantic	semantic	ADJ
fcis-28893	102	28	information	information	NOUN
fcis-28893	102	29	self	self	NOUN
fcis-28893	102	30	-	-	PUNCT
fcis-28893	102	31	aligning	align	VERB
fcis-28893	102	32	network	network	NOUN
fcis-28893	102	33	ssan	ssan	NOUN
fcis-28893	102	34	.	.	PUNCT
fcis-28893	103	1	for	for	ADP
fcis-28893	103	2	feature	feature	NOUN
fcis-28893	103	3	extraction	extraction	NOUN
fcis-28893	103	4	ssan	ssan	NOUN
fcis-28893	103	5	uses	use	VERB
fcis-28893	103	6	resnet-50	resnet-50	PROPN
fcis-28893	103	7	and	and	CCONJ
fcis-28893	103	8	bi	bi	ADJ
fcis-28893	103	9	-	-	ADJ
fcis-28893	103	10	directional	directional	ADJ
fcis-28893	103	11	lstm	lstm	NOUN
fcis-28893	103	12	as	as	ADP
fcis-28893	103	13	a	a	DET
fcis-28893	103	14	backbone	backbone	NOUN
fcis-28893	103	15	network	network	NOUN
fcis-28893	103	16	for	for	ADP
fcis-28893	103	17	extracting	extract	VERB
fcis-28893	103	18	graphic	graphic	ADJ
fcis-28893	103	19	and	and	CCONJ
fcis-28893	103	20	textual	textual	ADJ
fcis-28893	103	21	features	feature	NOUN
fcis-28893	103	22	.	.	PUNCT
fcis-28893	104	1	ssan	ssan	PROPN
fcis-28893	104	2	utilizes	utilize	VERB
fcis-28893	104	3	easy	easy	ADJ
fcis-28893	104	4	-	-	PUNCT
fcis-28893	104	5	to	to	ADP
fcis-28893	104	6	-	-	PUNCT
fcis-28893	104	7	align	align	VERB
fcis-28893	104	8	human	human	ADJ
fcis-28893	104	9	body	body	NOUN
fcis-28893	104	10	parts	part	NOUN
fcis-28893	104	11	as	as	ADP
fcis-28893	104	12	a	a	DET
fcis-28893	104	13	supervised	supervised	ADJ
fcis-28893	104	14	acquisition	acquisition	NOUN
fcis-28893	104	15	of	of	ADP
fcis-28893	104	16	local	local	ADJ
fcis-28893	104	17	features	feature	NOUN
fcis-28893	104	18	of	of	ADP
fcis-28893	104	19	the	the	DET
fcis-28893	104	20	text	text	NOUN
fcis-28893	104	21	,	,	PUNCT
fcis-28893	104	22	avoiding	avoid	VERB
fcis-28893	104	23	the	the	DET
fcis-28893	104	24	introduction	introduction	NOUN
fcis-28893	104	25	of	of	ADP
fcis-28893	104	26	additional	additional	ADJ
fcis-28893	104	27	tools	tool	NOUN
fcis-28893	104	28	to	to	PART
fcis-28893	104	29	ensure	ensure	VERB
fcis-28893	104	30	the	the	DET
fcis-28893	104	31	integrity	integrity	NOUN
fcis-28893	104	32	of	of	ADP
fcis-28893	104	33	contextual	contextual	ADJ
fcis-28893	104	34	links	link	NOUN
fcis-28893	104	35	in	in	ADP
fcis-28893	104	36	the	the	DET
fcis-28893	104	37	text	text	NOUN
fcis-28893	104	38	part	part	NOUN
fcis-28893	104	39	.	.	PUNCT
fcis-28893	105	1	meanwhile	meanwhile	ADV
fcis-28893	105	2	,	,	PUNCT
fcis-28893	105	3	to	to	PART
fcis-28893	105	4	better	well	ADV
fcis-28893	105	5	mine	mine	VERB
fcis-28893	105	6	the	the	DET
fcis-28893	105	7	local	local	ADJ
fcis-28893	105	8	connections	connection	NOUN
fcis-28893	105	9	in	in	ADP
fcis-28893	105	10	text	text	NOUN
fcis-28893	105	11	descriptions	description	NOUN
fcis-28893	105	12	and	and	CCONJ
fcis-28893	105	13	help	help	VERB
fcis-28893	105	14	the	the	DET
fcis-28893	105	15	model	model	NOUN
fcis-28893	105	16	adapt	adapt	VERB
fcis-28893	105	17	to	to	ADP
fcis-28893	105	18	text	text	NOUN
fcis-28893	105	19	-	-	PUNCT
fcis-28893	105	20	to	to	ADP
fcis-28893	105	21	-	-	PUNCT
fcis-28893	105	22	figure	figure	NOUN
fcis-28893	105	23	pedestrian	pedestrian	NOUN
fcis-28893	105	24	retrieval	retrieval	NOUN
fcis-28893	105	25	,	,	PUNCT
fcis-28893	105	26	ssan	ssan	NOUN
fcis-28893	105	27	introduces	introduce	VERB
fcis-28893	105	28	a	a	DET
fcis-28893	105	29	multi	multi	ADJ
fcis-28893	105	30	-	-	ADJ
fcis-28893	105	31	view	view	ADJ
fcis-28893	105	32	non	non	ADJ
fcis-28893	105	33	-	-	ADJ
fcis-28893	105	34	local	local	ADJ
fcis-28893	105	35	network	network	NOUN
fcis-28893	105	36	mv	mv	PROPN
fcis-28893	105	37	-	-	PROPN
fcis-28893	105	38	nln	nln	NOUN
fcis-28893	105	39	.	.	PUNCT
fcis-28893	106	1	the	the	DET
fcis-28893	106	2	loss	loss	NOUN
fcis-28893	106	3	function	function	NOUN
fcis-28893	106	4	aspect	aspect	NOUN
fcis-28893	106	5	uses	use	VERB
fcis-28893	106	6	ding	de	VERB
fcis-28893	106	7	et	et	PROPN
fcis-28893	106	8	al	al	PROPN
fcis-28893	106	9	.	.	PROPN
fcis-28893	106	10	to	to	PART
fcis-28893	106	11	design	design	VERB
fcis-28893	106	12	compound	compound	NOUN
fcis-28893	106	13	ranking	ranking	NOUN
fcis-28893	106	14	(	(	PUNCT
fcis-28893	106	15	cr	cr	NOUN
fcis-28893	106	16	)	)	PUNCT
fcis-28893	106	17	loss	loss	NOUN
fcis-28893	106	18	as	as	ADV
fcis-28893	106	19	well	well	ADV
fcis-28893	106	20	as	as	ADP
fcis-28893	106	21	the	the	DET
fcis-28893	106	22	commonly	commonly	ADV
fcis-28893	106	23	used	use	VERB
fcis-28893	106	24	i	i	PROPN
fcis-28893	106	25	d	d	PROPN
fcis-28893	106	26	loss	loss	NOUN
fcis-28893	106	27	for	for	ADP
fcis-28893	106	28	local	local	ADJ
fcis-28893	106	29	and	and	CCONJ
fcis-28893	106	30	global	global	ADJ
fcis-28893	106	31	feature	feature	NOUN
fcis-28893	106	32	optimization	optimization	NOUN
fcis-28893	106	33	.	.	PUNCT
fcis-28893	107	1	notably	notably	ADV
fcis-28893	107	2	,	,	PUNCT
fcis-28893	107	3	ding	ding	NOUN
fcis-28893	107	4	et	et	PROPN
fcis-28893	107	5	al	al	PROPN
fcis-28893	107	6	.	.	PROPN
fcis-28893	107	7	concluded	conclude	VERB
fcis-28893	107	8	that	that	SCONJ
fcis-28893	107	9	the	the	DET
fcis-28893	107	10	text	text	NOUN
fcis-28893	107	11	annotations	annotation	NOUN
fcis-28893	107	12	in	in	ADP
fcis-28893	107	13	the	the	DET
fcis-28893	107	14	cuhk	cuhk	ADJ
fcis-28893	107	15	-	-	PUNCT
fcis-28893	107	16	pedes	pes	NOUN
fcis-28893	107	17	dataset	dataset	NOUN
fcis-28893	107	18	contain	contain	VERB
fcis-28893	107	19	too	too	ADV
fcis-28893	107	20	much	much	ADJ
fcis-28893	107	21	descriptive	descriptive	ADJ
fcis-28893	107	22	information	information	NOUN
fcis-28893	107	23	that	that	PRON
fcis-28893	107	24	is	be	AUX
fcis-28893	107	25	not	not	PART
fcis-28893	107	26	relevant	relevant	ADJ
fcis-28893	107	27	to	to	ADP
fcis-28893	107	28	individuals	individual	NOUN
fcis-28893	107	29	,	,	PUNCT
fcis-28893	107	30	so	so	ADV
fcis-28893	107	31	ding	ding	NOUN
fcis-28893	107	32	et	et	PROPN
fcis-28893	107	33	al	al	PROPN
fcis-28893	107	34	.	.	PROPN
fcis-28893	107	35	created	create	VERB
fcis-28893	107	36	a	a	DET
fcis-28893	107	37	new	new	ADJ
fcis-28893	107	38	dataset	dataset	NOUN
fcis-28893	107	39	,	,	PUNCT
fcis-28893	107	40	icfg	icfg	NOUN
fcis-28893	107	41	-	-	NOUN
fcis-28893	107	42	pedes	pes	NOUN
fcis-28893	107	43	.	.	PUNCT
fcis-28893	108	1	ding	ding	NOUN
fcis-28893	108	2	et	et	PROPN
fcis-28893	108	3	al	al	PROPN
fcis-28893	108	4	.	.	PROPN
fcis-28893	108	5	designed	design	VERB
fcis-28893	108	6	a	a	DET
fcis-28893	108	7	new	new	ADJ
fcis-28893	108	8	loss	loss	NOUN
fcis-28893	108	9	function	function	NOUN
fcis-28893	108	10	while	while	SCONJ
fcis-28893	108	11	making	make	VERB
fcis-28893	108	12	full	full	ADJ
fcis-28893	108	13	use	use	NOUN
fcis-28893	108	14	of	of	ADP
fcis-28893	108	15	the	the	DET
fcis-28893	108	16	semantic	semantic	ADJ
fcis-28893	108	17	information	information	NOUN
fcis-28893	108	18	and	and	CCONJ
fcis-28893	108	19	constructed	construct	VERB
fcis-28893	108	20	a	a	DET
fcis-28893	108	21	completely	completely	ADV
fcis-28893	108	22	new	new	ADJ
fcis-28893	108	23	dataset	dataset	NOUN
fcis-28893	108	24	,	,	PUNCT
fcis-28893	108	25	which	which	PRON
fcis-28893	108	26	makes	make	VERB
fcis-28893	108	27	an	an	DET
fcis-28893	108	28	excellent	excellent	ADJ
fcis-28893	108	29	contribution	contribution	NOUN
fcis-28893	108	30	to	to	ADP
fcis-28893	108	31	the	the	DET
fcis-28893	108	32	text	text	NOUN
fcis-28893	108	33	-	-	PUNCT
fcis-28893	108	34	based	base	VERB
fcis-28893	108	35	pedestrian	pedestrian	NOUN
fcis-28893	108	36	retrieval	retrieval	NOUN
fcis-28893	108	37	field	field	NOUN
fcis-28893	108	38	development	development	NOUN
fcis-28893	108	39	with	with	ADP
fcis-28893	108	40	excellent	excellent	ADJ
fcis-28893	108	41	contributions	contribution	NOUN
fcis-28893	108	42	.	.	PUNCT
fcis-28893	109	1	4.3	4.3	NUM
fcis-28893	109	2	.	.	PUNCT
fcis-28893	109	3	methods	method	NOUN
fcis-28893	109	4	based	base	VERB
fcis-28893	109	5	on	on	ADP
fcis-28893	109	6	adversarial	adversarial	ADJ
fcis-28893	109	7	learning	learn	VERB
fcis-28893	109	8	liu	liu	PROPN
fcis-28893	109	9	et	et	PROPN
fcis-28893	109	10	al.[12	al.[12	PROPN
fcis-28893	109	11	]	]	PUNCT
fcis-28893	109	12	first	first	ADV
fcis-28893	109	13	proposed	propose	VERB
fcis-28893	109	14	a	a	DET
fcis-28893	109	15	deep	deep	ADJ
fcis-28893	109	16	adversarial	adversarial	ADJ
fcis-28893	109	17	graph	graph	NOUN
fcis-28893	109	18	attention	attention	NOUN
fcis-28893	109	19	convolutional	convolutional	ADJ
fcis-28893	109	20	network	network	NOUN
fcis-28893	109	21	framework	framework	NOUN
fcis-28893	109	22	,	,	PUNCT
fcis-28893	109	23	a	a	DET
fcis-28893	109	24	-	-	PUNCT
fcis-28893	109	25	ganet	ganet	NOUN
fcis-28893	109	26	,	,	PUNCT
fcis-28893	109	27	which	which	PRON
fcis-28893	109	28	consists	consist	VERB
fcis-28893	109	29	of	of	ADP
fcis-28893	109	30	an	an	DET
fcis-28893	109	31	image	image	NOUN
fcis-28893	109	32	graph	graph	NOUN
fcis-28893	109	33	attention	attention	NOUN
fcis-28893	109	34	network	network	NOUN
fcis-28893	109	35	,	,	PUNCT
fcis-28893	109	36	a	a	DET
fcis-28893	109	37	text	text	NOUN
fcis-28893	109	38	graph	graph	NOUN
fcis-28893	109	39	attention	attention	NOUN
fcis-28893	109	40	network	network	NOUN
fcis-28893	109	41	,	,	PUNCT
fcis-28893	109	42	and	and	CCONJ
fcis-28893	109	43	an	an	DET
fcis-28893	109	44	adversarial	adversarial	ADJ
fcis-28893	109	45	learning	learning	NOUN
fcis-28893	109	46	module	module	NOUN
fcis-28893	109	47	.	.	PUNCT
fcis-28893	110	1	the	the	DET
fcis-28893	110	2	image	image	NOUN
fcis-28893	110	3	graph	graph	NOUN
fcis-28893	110	4	attention	attention	NOUN
fcis-28893	110	5	network	network	NOUN
fcis-28893	110	6	uses	use	VERB
fcis-28893	110	7	resnet-50	resnet-50	PROPN
fcis-28893	110	8	as	as	ADP
fcis-28893	110	9	the	the	DET
fcis-28893	110	10	backbone	backbone	NOUN
fcis-28893	110	11	network	network	NOUN
fcis-28893	110	12	,	,	PUNCT
fcis-28893	110	13	which	which	PRON
fcis-28893	110	14	contains	contain	VERB
fcis-28893	110	15	five	five	NUM
fcis-28893	110	16	residual	residual	ADJ
fcis-28893	110	17	blocks	block	NOUN
fcis-28893	110	18	,	,	PUNCT
fcis-28893	110	19	a	a	DET
fcis-28893	110	20	visual	visual	ADJ
fcis-28893	110	21	scene	scene	NOUN
fcis-28893	110	22	graph	graph	NOUN
fcis-28893	110	23	module	module	NOUN
fcis-28893	110	24	,	,	PUNCT
fcis-28893	110	25	and	and	CCONJ
fcis-28893	110	26	a	a	DET
fcis-28893	110	27	joint	joint	ADJ
fcis-28893	110	28	embedding	embed	VERB
fcis-28893	110	29	layer	layer	NOUN
fcis-28893	110	30	.	.	PUNCT
fcis-28893	111	1	liu	liu	PROPN
fcis-28893	111	2	et	et	PROPN
fcis-28893	111	3	al	al	PROPN
fcis-28893	111	4	.	.	PROPN
fcis-28893	112	1	designed	design	VERB
fcis-28893	112	2	graph	graph	NOUN
fcis-28893	112	3	attention	attention	NOUN
fcis-28893	112	4	convolutional	convolutional	ADJ
fcis-28893	112	5	layer	layer	NOUN
fcis-28893	112	6	(	(	PUNCT
fcis-28893	112	7	gaconv	gaconv	NOUN
fcis-28893	112	8	)	)	PUNCT
fcis-28893	112	9	for	for	ADP
fcis-28893	112	10	extracting	extract	VERB
fcis-28893	112	11	structured	structured	ADJ
fcis-28893	112	12	semantic	semantic	ADJ
fcis-28893	112	13	visual	visual	ADJ
fcis-28893	112	14	features	feature	NOUN
fcis-28893	112	15	.	.	PUNCT
fcis-28893	113	1	finally	finally	ADV
fcis-28893	113	2	,	,	PUNCT
fcis-28893	113	3	the	the	DET
fcis-28893	113	4	joint	joint	NOUN
fcis-28893	113	5	embedding	embed	VERB
fcis-28893	113	6	layer	layer	NOUN
fcis-28893	113	7	maps	map	VERB
fcis-28893	113	8	the	the	DET
fcis-28893	113	9	virtual	virtual	ADJ
fcis-28893	113	10	node	node	ADJ
fcis-28893	113	11	feature	feature	NOUN
fcis-28893	113	12	values	value	NOUN
fcis-28893	113	13	output	output	VERB
fcis-28893	113	14	from	from	ADP
fcis-28893	113	15	the	the	DET
fcis-28893	113	16	visual	visual	ADJ
fcis-28893	113	17	scene	scene	NOUN
fcis-28893	113	18	graph	graph	NOUN
fcis-28893	113	19	module	module	NOUN
fcis-28893	113	20	to	to	ADP
fcis-28893	113	21	the	the	DET
fcis-28893	113	22	shared	share	VERB
fcis-28893	113	23	feature	feature	NOUN
fcis-28893	113	24	space	space	NOUN
fcis-28893	113	25	by	by	ADP
fcis-28893	113	26	joining	join	VERB
fcis-28893	113	27	them	they	PRON
fcis-28893	113	28	with	with	ADP
fcis-28893	113	29	the	the	DET
fcis-28893	113	30	global	global	ADJ
fcis-28893	113	31	visual	visual	ADJ
fcis-28893	113	32	features	feature	NOUN
fcis-28893	113	33	.	.	PUNCT
fcis-28893	114	1	the	the	DET
fcis-28893	114	2	textual	textual	ADJ
fcis-28893	114	3	graph	graph	NOUN
fcis-28893	114	4	attention	attention	NOUN
fcis-28893	114	5	network	network	NOUN
fcis-28893	114	6	consists	consist	VERB
fcis-28893	114	7	of	of	ADP
fcis-28893	114	8	a	a	DET
fcis-28893	114	9	bidirectional	bidirectional	ADJ
fcis-28893	114	10	lstm	lstm	NOUN
fcis-28893	114	11	network	network	NOUN
fcis-28893	114	12	,	,	PUNCT
fcis-28893	114	13	a	a	DET
fcis-28893	114	14	textual	textual	ADJ
fcis-28893	114	15	scene	scene	NOUN
fcis-28893	114	16	graph	graph	NOUN
fcis-28893	114	17	module	module	NOUN
fcis-28893	114	18	with	with	ADP
fcis-28893	114	19	a	a	DET
fcis-28893	114	20	joint	joint	ADJ
fcis-28893	114	21	embedding	embed	VERB
fcis-28893	114	22	layer	layer	NOUN
fcis-28893	114	23	.	.	PUNCT
fcis-28893	115	1	first	first	ADJ
fcis-28893	115	2	bi	bi	ADJ
fcis-28893	115	3	-	-	ADJ
fcis-28893	115	4	directional	directional	ADJ
fcis-28893	115	5	lstm	lstm	ADJ
fcis-28893	115	6	extracts	extract	NOUN
fcis-28893	115	7	text	text	NOUN
fcis-28893	115	8	features	feature	NOUN
fcis-28893	115	9	and	and	CCONJ
fcis-28893	115	10	understands	understand	VERB
fcis-28893	115	11	potential	potential	ADJ
fcis-28893	115	12	semantic	semantic	ADJ
fcis-28893	115	13	dependencies	dependency	NOUN
fcis-28893	115	14	between	between	ADP
fcis-28893	115	15	words	word	NOUN
fcis-28893	115	16	,	,	PUNCT
fcis-28893	115	17	similar	similar	ADJ
fcis-28893	115	18	to	to	ADP
fcis-28893	115	19	the	the	DET
fcis-28893	115	20	visual	visual	ADJ
fcis-28893	115	21	scene	scene	NOUN
fcis-28893	115	22	graph	graph	NOUN
fcis-28893	115	23	module	module	NOUN
fcis-28893	115	24	,	,	PUNCT
fcis-28893	115	25	the	the	DET
fcis-28893	115	26	text	text	NOUN
fcis-28893	115	27	scene	scene	NOUN
fcis-28893	115	28	graph	graph	NOUN
fcis-28893	115	29	module	module	NOUN
fcis-28893	115	30	detects	detect	NOUN
fcis-28893	115	31	objects	object	VERB
fcis-28893	115	32	present	present	ADJ
fcis-28893	115	33	in	in	ADP
fcis-28893	115	34	the	the	DET
fcis-28893	115	35	text	text	NOUN
fcis-28893	115	36	and	and	CCONJ
fcis-28893	115	37	predicts	predict	VERB
fcis-28893	115	38	the	the	DET
fcis-28893	115	39	relationships	relationship	NOUN
fcis-28893	115	40	between	between	ADP
fcis-28893	115	41	them	they	PRON
fcis-28893	115	42	,	,	PUNCT
fcis-28893	115	43	and	and	CCONJ
fcis-28893	115	44	at	at	ADP
fcis-28893	115	45	the	the	DET
fcis-28893	115	46	same	same	ADJ
fcis-28893	115	47	time	time	NOUN
fcis-28893	115	48	,	,	PUNCT
fcis-28893	115	49	a	a	DET
fcis-28893	115	50	graph	graph	NOUN
fcis-28893	115	51	attention	attention	NOUN
fcis-28893	115	52	convolutional	convolutional	ADJ
fcis-28893	115	53	layer	layer	NOUN
fcis-28893	115	54	is	be	AUX
fcis-28893	115	55	used	use	VERB
fcis-28893	115	56	to	to	PART
fcis-28893	115	57	obtain	obtain	VERB
fcis-28893	115	58	structured	structured	ADJ
fcis-28893	115	59	text	text	NOUN
fcis-28893	115	60	semantic	semantic	ADJ
fcis-28893	115	61	features	feature	NOUN
fcis-28893	115	62	,	,	PUNCT
fcis-28893	115	63	and	and	CCONJ
fcis-28893	115	64	finally	finally	ADV
fcis-28893	115	65	,	,	PUNCT
fcis-28893	115	66	the	the	DET
fcis-28893	115	67	structured	structured	ADJ
fcis-28893	115	68	semantic	semantic	ADJ
fcis-28893	115	69	text	text	NOUN
fcis-28893	115	70	features	feature	NOUN
fcis-28893	115	71	and	and	CCONJ
fcis-28893	115	72	the	the	DET
fcis-28893	115	73	initial	initial	ADJ
fcis-28893	115	74	text	text	NOUN
fcis-28893	115	75	features	feature	NOUN
fcis-28893	115	76	are	be	AUX
fcis-28893	115	77	fused	fuse	VERB
fcis-28893	115	78	,	,	PUNCT
fcis-28893	115	79	which	which	PRON
fcis-28893	115	80	are	be	AUX
fcis-28893	115	81	mapped	map	VERB
fcis-28893	115	82	by	by	ADP
fcis-28893	115	83	the	the	DET
fcis-28893	115	84	joint	joint	ADJ
fcis-28893	115	85	embedding	embed	VERB
fcis-28893	115	86	layer	layer	NOUN
fcis-28893	115	87	into	into	ADP
fcis-28893	115	88	the	the	DET
fcis-28893	115	89	shared	share	VERB
fcis-28893	115	90	feature	feature	NOUN
fcis-28893	115	91	space	space	NOUN
fcis-28893	115	92	.	.	PUNCT
fcis-28893	116	1	the	the	DET
fcis-28893	116	2	adversarial	adversarial	ADJ
fcis-28893	116	3	learning	learning	NOUN
fcis-28893	116	4	module	module	NOUN
fcis-28893	116	5	consists	consist	VERB
fcis-28893	116	6	of	of	ADP
fcis-28893	116	7	a	a	DET
fcis-28893	116	8	modal	modal	ADJ
fcis-28893	116	9	discriminator	discriminator	NOUN
fcis-28893	116	10	and	and	CCONJ
fcis-28893	116	11	a	a	DET
fcis-28893	116	12	feature	feature	NOUN
fcis-28893	116	13	converter	converter	NOUN
fcis-28893	116	14	,	,	PUNCT
fcis-28893	116	15	which	which	PRON
fcis-28893	116	16	are	be	AUX
fcis-28893	116	17	used	use	VERB
fcis-28893	116	18	to	to	PART
fcis-28893	116	19	optimize	optimize	VERB
fcis-28893	116	20	the	the	DET
fcis-28893	116	21	features	feature	NOUN
fcis-28893	116	22	extracted	extract	VERB
fcis-28893	116	23	by	by	ADP
fcis-28893	116	24	the	the	DET
fcis-28893	116	25	image	image	NOUN
fcis-28893	116	26	graph	graph	NOUN
fcis-28893	116	27	attention	attention	NOUN
fcis-28893	116	28	network	network	NOUN
fcis-28893	116	29	and	and	CCONJ
fcis-28893	116	30	the	the	DET
fcis-28893	116	31	text	text	NOUN
fcis-28893	116	32	graph	graph	NOUN
fcis-28893	116	33	attention	attention	NOUN
fcis-28893	116	34	network	network	NOUN
fcis-28893	116	35	,	,	PUNCT
fcis-28893	116	36	and	and	CCONJ
fcis-28893	116	37	the	the	DET
fcis-28893	116	38	process	process	NOUN
fcis-28893	116	39	is	be	AUX
fcis-28893	116	40	like	like	ADP
fcis-28893	116	41	that	that	PRON
fcis-28893	116	42	of	of	ADP
fcis-28893	116	43	gan	gan	PROPN
fcis-28893	116	44	.	.	PROPN
fcis-28893	117	1	based	base	VERB
fcis-28893	117	2	on	on	ADP
fcis-28893	117	3	the	the	DET
fcis-28893	117	4	training	training	NOUN
fcis-28893	117	5	of	of	ADP
fcis-28893	117	6	the	the	DET
fcis-28893	117	7	feature	feature	NOUN
fcis-28893	117	8	converter	converter	NOUN
fcis-28893	117	9	,	,	PUNCT
fcis-28893	117	10	the	the	DET
fcis-28893	117	11	goal	goal	NOUN
fcis-28893	117	12	of	of	ADP
fcis-28893	117	13	the	the	DET
fcis-28893	117	14	modal	modal	ADJ
fcis-28893	117	15	discriminator	discriminator	NOUN
fcis-28893	117	16	is	be	AUX
fcis-28893	117	17	to	to	PART
fcis-28893	117	18	utilize	utilize	VERB
fcis-28893	117	19	the	the	DET
fcis-28893	117	20	adversarial	adversarial	ADJ
fcis-28893	117	21	loss	loss	NOUN
fcis-28893	117	22	function	function	NOUN
fcis-28893	117	23	to	to	PART
fcis-28893	117	24	distinguish	distinguish	VERB
fcis-28893	117	25	the	the	DET
fcis-28893	117	26	sample	sample	NOUN
fcis-28893	117	27	modalities	modality	NOUN
fcis-28893	117	28	as	as	ADV
fcis-28893	117	29	much	much	ADV
fcis-28893	117	30	as	as	ADP
fcis-28893	117	31	possible	possible	ADJ
fcis-28893	117	32	,	,	PUNCT
fcis-28893	117	33	while	while	SCONJ
fcis-28893	117	34	the	the	DET
fcis-28893	117	35	feature	feature	NOUN
fcis-28893	117	36	converter	converter	NOUN
fcis-28893	117	37	maps	map	NOUN
fcis-28893	117	38	the	the	DET
fcis-28893	117	39	graphic	graphic	ADJ
fcis-28893	117	40	and	and	CCONJ
fcis-28893	117	41	textual	textual	ADJ
fcis-28893	117	42	modal	modal	NOUN
fcis-28893	117	43	features	feature	NOUN
fcis-28893	117	44	into	into	ADP
fcis-28893	117	45	a	a	DET
fcis-28893	117	46	common	common	ADJ
fcis-28893	117	47	feature	feature	NOUN
fcis-28893	117	48	space	space	NOUN
fcis-28893	117	49	,	,	PUNCT
fcis-28893	117	50	where	where	SCONJ
fcis-28893	117	51	the	the	DET
fcis-28893	117	52	recognition	recognition	NOUN
fcis-28893	117	53	loss	loss	NOUN
fcis-28893	117	54	function	function	NOUN
fcis-28893	117	55	is	be	AUX
fcis-28893	117	56	used	use	VERB
fcis-28893	117	57	to	to	PART
fcis-28893	117	58	make	make	VERB
fcis-28893	117	59	the	the	DET
fcis-28893	117	60	features	feature	NOUN
fcis-28893	117	61	discriminative	discriminative	NOUN
fcis-28893	117	62	,	,	PUNCT
fcis-28893	117	63	and	and	CCONJ
fcis-28893	117	64	the	the	DET
fcis-28893	117	65	pairwise	pairwise	NOUN
fcis-28893	117	66	loss	loss	NOUN
fcis-28893	117	67	function	function	NOUN
fcis-28893	117	68	is	be	AUX
fcis-28893	117	69	used	use	VERB
fcis-28893	117	70	to	to	PART
fcis-28893	117	71	ensure	ensure	VERB
fcis-28893	117	72	that	that	SCONJ
fcis-28893	117	73	the	the	DET
fcis-28893	117	74	graphic	graphic	ADJ
fcis-28893	117	75	and	and	CCONJ
fcis-28893	117	76	textual	textual	ADJ
fcis-28893	117	77	features	feature	NOUN
fcis-28893	117	78	belonging	belong	VERB
fcis-28893	117	79	to	to	ADP
fcis-28893	117	80	the	the	DET
fcis-28893	117	81	same	same	ADJ
fcis-28893	117	82	identity	identity	NOUN
fcis-28893	117	83	have	have	VERB
fcis-28893	117	84	a	a	DET
fcis-28893	117	85	higher	high	ADJ
fcis-28893	117	86	level	level	NOUN
fcis-28893	117	87	of	of	ADP
fcis-28893	117	88	semantic	semantic	ADJ
fcis-28893	117	89	similarity	similarity	NOUN
fcis-28893	117	90	and	and	CCONJ
fcis-28893	117	91	123	123	NUM
fcis-28893	117	92	cross	cross	ADJ
fcis-28893	117	93	-	-	ADJ
fcis-28893	117	94	modal	modal	ADJ
fcis-28893	117	95	invariance	invariance	NOUN
fcis-28893	117	96	.	.	PUNCT
fcis-28893	118	1	a	a	DET
fcis-28893	118	2	-	-	PUNCT
fcis-28893	118	3	ganet	ganet	NOUN
fcis-28893	118	4	fully	fully	ADV
fcis-28893	118	5	mines	mine	NOUN
fcis-28893	118	6	the	the	DET
fcis-28893	118	7	directed	direct	VERB
fcis-28893	118	8	semantic	semantic	ADJ
fcis-28893	118	9	scene	scene	NOUN
fcis-28893	118	10	graph	graph	NOUN
fcis-28893	118	11	to	to	PART
fcis-28893	118	12	extract	extract	VERB
fcis-28893	118	13	more	more	ADJ
fcis-28893	118	14	discriminative	discriminative	NOUN
fcis-28893	118	15	graphical	graphical	ADJ
fcis-28893	118	16	and	and	CCONJ
fcis-28893	118	17	textual	textual	ADJ
fcis-28893	118	18	features	feature	NOUN
fcis-28893	118	19	,	,	PUNCT
fcis-28893	118	20	which	which	PRON
fcis-28893	118	21	improves	improve	VERB
fcis-28893	118	22	the	the	DET
fcis-28893	118	23	retrieval	retrieval	NOUN
fcis-28893	118	24	accuracy	accuracy	NOUN
fcis-28893	118	25	without	without	ADP
fcis-28893	118	26	adding	add	VERB
fcis-28893	118	27	extra	extra	ADJ
fcis-28893	118	28	modules	module	NOUN
fcis-28893	118	29	.	.	PUNCT
fcis-28893	119	1	the	the	DET
fcis-28893	119	2	adversarial	adversarial	ADJ
fcis-28893	119	3	idea	idea	NOUN
fcis-28893	119	4	is	be	AUX
fcis-28893	119	5	firstly	firstly	ADV
fcis-28893	119	6	introduced	introduce	VERB
fcis-28893	119	7	into	into	ADP
fcis-28893	119	8	the	the	DET
fcis-28893	119	9	field	field	NOUN
fcis-28893	119	10	of	of	ADP
fcis-28893	119	11	text	text	NOUN
fcis-28893	119	12	-	-	PUNCT
fcis-28893	119	13	based	base	VERB
fcis-28893	119	14	pedestrian	pedestrian	NOUN
fcis-28893	119	15	retrieval	retrieval	NOUN
fcis-28893	119	16	,	,	PUNCT
fcis-28893	119	17	which	which	PRON
fcis-28893	119	18	provides	provide	VERB
fcis-28893	119	19	new	new	ADJ
fcis-28893	119	20	ideas	idea	NOUN
fcis-28893	119	21	for	for	ADP
fcis-28893	119	22	subsequent	subsequent	ADJ
fcis-28893	119	23	research	research	NOUN
fcis-28893	119	24	.	.	PUNCT
fcis-28893	120	1	however	however	ADV
fcis-28893	120	2	,	,	PUNCT
fcis-28893	120	3	the	the	DET
fcis-28893	120	4	introduction	introduction	NOUN
fcis-28893	120	5	of	of	ADP
fcis-28893	120	6	graph	graph	NOUN
fcis-28893	120	7	attention	attention	NOUN
fcis-28893	120	8	mechanism	mechanism	NOUN
fcis-28893	120	9	increases	increase	VERB
fcis-28893	120	10	the	the	DET
fcis-28893	120	11	complexity	complexity	NOUN
fcis-28893	120	12	of	of	ADP
fcis-28893	120	13	the	the	DET
fcis-28893	120	14	training	training	NOUN
fcis-28893	120	15	process	process	NOUN
fcis-28893	120	16	,	,	PUNCT
fcis-28893	120	17	while	while	SCONJ
fcis-28893	120	18	the	the	DET
fcis-28893	120	19	balance	balance	NOUN
fcis-28893	120	20	between	between	ADP
fcis-28893	120	21	modal	modal	ADJ
fcis-28893	120	22	discriminator	discriminator	NOUN
fcis-28893	120	23	and	and	CCONJ
fcis-28893	120	24	feature	feature	NOUN
fcis-28893	120	25	converter	converter	NOUN
fcis-28893	120	26	is	be	AUX
fcis-28893	120	27	harder	hard	ADJ
fcis-28893	120	28	to	to	PART
fcis-28893	120	29	define	define	VERB
fcis-28893	120	30	.	.	PUNCT
fcis-28893	121	1	in	in	ADP
fcis-28893	121	2	the	the	DET
fcis-28893	121	3	same	same	ADJ
fcis-28893	121	4	year	year	NOUN
fcis-28893	121	5	,	,	PUNCT
fcis-28893	121	6	sarafianos	sarafiano	VERB
fcis-28893	121	7	et	et	NOUN
fcis-28893	121	8	al.[13	al.[13	NOUN
fcis-28893	121	9	]	]	PUNCT
fcis-28893	121	10	also	also	ADV
fcis-28893	121	11	proposed	propose	VERB
fcis-28893	121	12	timam	timam	NUM
fcis-28893	121	13	,	,	PUNCT
fcis-28893	121	14	a	a	DET
fcis-28893	121	15	network	network	NOUN
fcis-28893	121	16	framework	framework	NOUN
fcis-28893	121	17	based	base	VERB
fcis-28893	121	18	on	on	ADP
fcis-28893	121	19	adversarial	adversarial	ADJ
fcis-28893	121	20	ideas	idea	NOUN
fcis-28893	121	21	.	.	PUNCT
fcis-28893	122	1	sarafianos	sarafianos	VERB
fcis-28893	122	2	et	et	PROPN
fcis-28893	122	3	al	al	PROPN
fcis-28893	122	4	.	.	PROPN
fcis-28893	122	5	argued	argue	VERB
fcis-28893	122	6	that	that	SCONJ
fcis-28893	122	7	the	the	DET
fcis-28893	122	8	lack	lack	NOUN
fcis-28893	122	9	of	of	ADP
fcis-28893	122	10	textual	textual	ADJ
fcis-28893	122	11	feature	feature	NOUN
fcis-28893	122	12	recognition	recognition	NOUN
fcis-28893	122	13	makes	make	VERB
fcis-28893	122	14	the	the	DET
fcis-28893	122	15	main	main	ADJ
fcis-28893	122	16	reason	reason	NOUN
fcis-28893	122	17	for	for	ADP
fcis-28893	122	18	the	the	DET
fcis-28893	122	19	poor	poor	ADJ
fcis-28893	122	20	performance	performance	NOUN
fcis-28893	122	21	of	of	ADP
fcis-28893	122	22	current	current	ADJ
fcis-28893	122	23	computer	computer	NOUN
fcis-28893	122	24	vision	vision	NOUN
fcis-28893	122	25	methods	method	NOUN
fcis-28893	122	26	in	in	ADP
fcis-28893	122	27	graphic	graphic	ADJ
fcis-28893	122	28	matching	matching	NOUN
fcis-28893	122	29	tasks	task	NOUN
fcis-28893	122	30	.	.	PUNCT
fcis-28893	123	1	for	for	ADP
fcis-28893	123	2	this	this	DET
fcis-28893	123	3	reason	reason	NOUN
fcis-28893	123	4	,	,	PUNCT
fcis-28893	123	5	sarafianos	sarafiano	VERB
fcis-28893	123	6	et	et	PROPN
fcis-28893	123	7	al	al	PROPN
fcis-28893	123	8	.	.	PROPN
fcis-28893	124	1	introduced	introduce	VERB
fcis-28893	124	2	bert[14	bert[14	PROPN
fcis-28893	124	3	]	]	PUNCT
fcis-28893	124	4	for	for	ADP
fcis-28893	124	5	the	the	DET
fcis-28893	124	6	first	first	ADJ
fcis-28893	124	7	time	time	NOUN
fcis-28893	124	8	to	to	ADP
fcis-28893	124	9	the	the	DET
fcis-28893	124	10	study	study	NOUN
fcis-28893	124	11	of	of	ADP
fcis-28893	124	12	text	text	NOUN
fcis-28893	124	13	-	-	PUNCT
fcis-28893	124	14	based	base	VERB
fcis-28893	124	15	pedestrian	pedestrian	NOUN
fcis-28893	124	16	retrieval	retrieval	NOUN
fcis-28893	124	17	.	.	PUNCT
fcis-28893	125	1	compared	compare	VERB
fcis-28893	125	2	with	with	ADP
fcis-28893	125	3	lstm	lstm	PROPN
fcis-28893	125	4	,	,	PUNCT
fcis-28893	125	5	a	a	DET
fcis-28893	125	6	text	text	NOUN
fcis-28893	125	7	feature	feature	NOUN
fcis-28893	125	8	extraction	extraction	NOUN
fcis-28893	125	9	network	network	NOUN
fcis-28893	125	10	commonly	commonly	ADV
fcis-28893	125	11	used	use	VERB
fcis-28893	125	12	in	in	ADP
fcis-28893	125	13	the	the	DET
fcis-28893	125	14	field	field	NOUN
fcis-28893	125	15	,	,	PUNCT
fcis-28893	125	16	bert	bert	PROPN
fcis-28893	125	17	uses	use	VERB
fcis-28893	125	18	a	a	DET
fcis-28893	125	19	large	large	ADJ
fcis-28893	125	20	amount	amount	NOUN
fcis-28893	125	21	of	of	ADP
fcis-28893	125	22	data	datum	NOUN
fcis-28893	125	23	for	for	ADP
fcis-28893	125	24	unsupervised	unsupervised	ADJ
fcis-28893	125	25	pre	pre	NOUN
fcis-28893	125	26	-	-	NOUN
fcis-28893	125	27	training	training	NOUN
fcis-28893	125	28	and	and	CCONJ
fcis-28893	125	29	can	can	AUX
fcis-28893	125	30	be	be	AUX
fcis-28893	125	31	easily	easily	ADV
fcis-28893	125	32	migrated	migrate	VERB
fcis-28893	125	33	to	to	ADP
fcis-28893	125	34	various	various	ADJ
fcis-28893	125	35	downstream	downstream	ADJ
fcis-28893	125	36	tasks	task	NOUN
fcis-28893	125	37	while	while	SCONJ
fcis-28893	125	38	possessing	possess	VERB
fcis-28893	125	39	strong	strong	ADJ
fcis-28893	125	40	semantic	semantic	ADJ
fcis-28893	125	41	understanding	understanding	NOUN
fcis-28893	125	42	capabilities	capability	NOUN
fcis-28893	125	43	.	.	PUNCT
fcis-28893	126	1	secondly	secondly	ADV
fcis-28893	126	2	,	,	PUNCT
fcis-28893	126	3	bert	bert	PROPN
fcis-28893	126	4	adopts	adopt	VERB
fcis-28893	126	5	the	the	DET
fcis-28893	126	6	attention	attention	NOUN
fcis-28893	126	7	mechanism	mechanism	NOUN
fcis-28893	126	8	,	,	PUNCT
fcis-28893	126	9	which	which	PRON
fcis-28893	126	10	can	can	AUX
fcis-28893	126	11	compute	compute	VERB
fcis-28893	126	12	the	the	DET
fcis-28893	126	13	relative	relative	ADJ
fcis-28893	126	14	weights	weight	NOUN
fcis-28893	126	15	of	of	ADP
fcis-28893	126	16	different	different	ADJ
fcis-28893	126	17	positions	position	NOUN
fcis-28893	126	18	in	in	ADP
fcis-28893	126	19	parallel	parallel	NOUN
fcis-28893	126	20	,	,	PUNCT
fcis-28893	126	21	making	make	VERB
fcis-28893	126	22	it	it	PRON
fcis-28893	126	23	more	more	ADV
fcis-28893	126	24	efficient	efficient	ADJ
fcis-28893	126	25	in	in	ADP
fcis-28893	126	26	processing	process	VERB
fcis-28893	126	27	long	long	ADJ
fcis-28893	126	28	text	text	NOUN
fcis-28893	126	29	.	.	PUNCT
fcis-28893	127	1	in	in	ADP
fcis-28893	127	2	addition	addition	NOUN
fcis-28893	127	3	,	,	PUNCT
fcis-28893	127	4	bert	bert	PROPN
fcis-28893	127	5	adopts	adopt	VERB
fcis-28893	127	6	bidirectional	bidirectional	ADJ
fcis-28893	127	7	encoding	encoding	NOUN
fcis-28893	127	8	and	and	CCONJ
fcis-28893	127	9	a	a	DET
fcis-28893	127	10	masking	masking	NOUN
fcis-28893	127	11	strategy	strategy	NOUN
fcis-28893	127	12	during	during	ADP
fcis-28893	127	13	training	training	NOUN
fcis-28893	127	14	,	,	PUNCT
fcis-28893	127	15	which	which	PRON
fcis-28893	127	16	allows	allow	VERB
fcis-28893	127	17	it	it	PRON
fcis-28893	127	18	to	to	PART
fcis-28893	127	19	understand	understand	VERB
fcis-28893	127	20	contextual	contextual	ADJ
fcis-28893	127	21	information	information	NOUN
fcis-28893	127	22	in	in	ADP
fcis-28893	127	23	greater	great	ADJ
fcis-28893	127	24	depth	depth	NOUN
fcis-28893	127	25	.	.	PUNCT
fcis-28893	128	1	in	in	ADP
fcis-28893	128	2	addition	addition	NOUN
fcis-28893	128	3	,	,	PUNCT
fcis-28893	128	4	timam	timam	PRON
fcis-28893	128	5	introduces	introduce	NOUN
fcis-28893	128	6	two	two	NUM
fcis-28893	128	7	loss	loss	NOUN
fcis-28893	128	8	functions	function	NOUN
fcis-28893	128	9	for	for	ADP
fcis-28893	128	10	identification	identification	NOUN
fcis-28893	128	11	and	and	CCONJ
fcis-28893	128	12	cross	cross	ADJ
fcis-28893	128	13	-	-	ADJ
fcis-28893	128	14	modal	modal	ADJ
fcis-28893	128	15	matching	matching	NOUN
fcis-28893	128	16	,	,	PUNCT
fcis-28893	128	17	i.e.	i.e.	X
fcis-28893	128	18	,	,	PUNCT
fcis-28893	128	19	norm	norm	NOUN
fcis-28893	128	20	-	-	PUNCT
fcis-28893	128	21	softmax	softmax	NOUN
fcis-28893	128	22	cross	cross	PROPN
fcis-28893	128	23	entropy	entropy	PROPN
fcis-28893	128	24	loss	loss	PROPN
fcis-28893	128	25	and	and	CCONJ
fcis-28893	128	26	cmpm	cmpm	PROPN
fcis-28893	128	27	loss	loss	NOUN
fcis-28893	128	28	function	function	NOUN
fcis-28893	128	29	proposed	propose	VERB
fcis-28893	128	30	by	by	ADP
fcis-28893	128	31	zhang	zhang	PROPN
fcis-28893	128	32	et	et	PROPN
fcis-28893	128	33	al	al	PROPN
fcis-28893	128	34	.	.	PROPN
fcis-28893	128	35	and	and	CCONJ
fcis-28893	128	36	the	the	DET
fcis-28893	128	37	generalized	generalized	ADJ
fcis-28893	128	38	features	feature	NOUN
fcis-28893	128	39	are	be	AUX
fcis-28893	128	40	further	far	ADV
fcis-28893	128	41	learned	learn	VERB
fcis-28893	128	42	by	by	ADP
fcis-28893	128	43	gan	gan	ADJ
fcis-28893	128	44	loss	loss	NOUN
fcis-28893	128	45	to	to	PART
fcis-28893	128	46	maximize	maximize	VERB
fcis-28893	128	47	the	the	DET
fcis-28893	128	48	matching	matching	NOUN
fcis-28893	128	49	accuracy	accuracy	NOUN
fcis-28893	128	50	.	.	PUNCT
fcis-28893	129	1	the	the	DET
fcis-28893	129	2	accuracy	accuracy	NOUN
fcis-28893	129	3	improvement	improvement	NOUN
fcis-28893	129	4	brought	bring	VERB
fcis-28893	129	5	by	by	ADP
fcis-28893	129	6	bert	bert	PROPN
fcis-28893	129	7	is	be	AUX
fcis-28893	129	8	verified	verify	VERB
fcis-28893	129	9	by	by	ADP
fcis-28893	129	10	ablation	ablation	NOUN
fcis-28893	129	11	experiments	experiment	NOUN
fcis-28893	129	12	.	.	PUNCT
fcis-28893	130	1	the	the	DET
fcis-28893	130	2	rank-1	rank-1	NUM
fcis-28893	130	3	accuracy	accuracy	NOUN
fcis-28893	130	4	comes	come	VERB
fcis-28893	130	5	from	from	ADP
fcis-28893	130	6	49.85	49.85	NUM
fcis-28893	130	7	%	%	NOUN
fcis-28893	130	8	to	to	ADP
fcis-28893	130	9	52.97	52.97	NUM
fcis-28893	130	10	%	%	NOUN
fcis-28893	130	11	.	.	PUNCT
fcis-28893	131	1	with	with	ADP
fcis-28893	131	2	the	the	DET
fcis-28893	131	3	help	help	NOUN
fcis-28893	131	4	of	of	ADP
fcis-28893	131	5	bert	bert	PROPN
fcis-28893	131	6	and	and	CCONJ
fcis-28893	131	7	the	the	DET
fcis-28893	131	8	adversarial	adversarial	ADJ
fcis-28893	131	9	representation	representation	NOUN
fcis-28893	131	10	learning	learning	NOUN
fcis-28893	131	11	paradigm	paradigm	NOUN
fcis-28893	131	12	,	,	PUNCT
fcis-28893	131	13	timam	timam	PROPN
fcis-28893	131	14	refreshed	refresh	VERB
fcis-28893	131	15	the	the	DET
fcis-28893	131	16	sota	sota	NOUN
fcis-28893	131	17	score	score	NOUN
fcis-28893	131	18	,	,	PUNCT
fcis-28893	131	19	verifying	verify	VERB
fcis-28893	131	20	the	the	DET
fcis-28893	131	21	advantages	advantage	NOUN
fcis-28893	131	22	of	of	ADP
fcis-28893	131	23	bert	bert	PROPN
fcis-28893	131	24	over	over	ADP
fcis-28893	131	25	traditional	traditional	ADJ
fcis-28893	131	26	lstm	lstm	NOUN
fcis-28893	131	27	networks	network	NOUN
fcis-28893	131	28	.	.	PUNCT
fcis-28893	132	1	since	since	SCONJ
fcis-28893	132	2	then	then	ADV
fcis-28893	132	3	,	,	PUNCT
fcis-28893	132	4	more	more	ADJ
fcis-28893	132	5	and	and	CCONJ
fcis-28893	132	6	more	more	ADJ
fcis-28893	132	7	researchers	researcher	NOUN
fcis-28893	132	8	have	have	AUX
fcis-28893	132	9	begun	begin	VERB
fcis-28893	132	10	to	to	PART
fcis-28893	132	11	use	use	VERB
fcis-28893	132	12	bert	bert	NOUN
fcis-28893	132	13	as	as	ADP
fcis-28893	132	14	a	a	DET
fcis-28893	132	15	text	text	NOUN
fcis-28893	132	16	feature	feature	NOUN
fcis-28893	132	17	encoder	encoder	NOUN
fcis-28893	132	18	,	,	PUNCT
fcis-28893	132	19	which	which	PRON
fcis-28893	132	20	promotes	promote	VERB
fcis-28893	132	21	the	the	DET
fcis-28893	132	22	further	further	ADJ
fcis-28893	132	23	development	development	NOUN
fcis-28893	132	24	of	of	ADP
fcis-28893	132	25	the	the	DET
fcis-28893	132	26	field	field	NOUN
fcis-28893	132	27	.	.	PUNCT
fcis-28893	133	1	4.4	4.4	NUM
fcis-28893	133	2	.	.	PUNCT
fcis-28893	133	3	method	method	PROPN
fcis-28893	133	4	based	base	VERB
fcis-28893	133	5	on	on	ADP
fcis-28893	133	6	cross	cross	ADJ
fcis-28893	133	7	-	-	ADJ
fcis-28893	133	8	modal	modal	ADJ
fcis-28893	133	9	attention	attention	NOUN
fcis-28893	133	10	mechanisms	mechanism	NOUN
fcis-28893	133	11	zhu	zhu	PROPN
fcis-28893	133	12	et	et	PROPN
fcis-28893	133	13	al.[3	al.[3	PROPN
fcis-28893	133	14	]	]	PUNCT
fcis-28893	133	15	argued	argue	VERB
fcis-28893	133	16	that	that	SCONJ
fcis-28893	133	17	image	image	NOUN
fcis-28893	133	18	information	information	NOUN
fcis-28893	133	19	can	can	AUX
fcis-28893	133	20	be	be	AUX
fcis-28893	133	21	divided	divide	VERB
fcis-28893	133	22	into	into	ADP
fcis-28893	133	23	character	character	NOUN
fcis-28893	133	24	information	information	NOUN
fcis-28893	133	25	and	and	CCONJ
fcis-28893	133	26	background	background	NOUN
fcis-28893	133	27	information	information	NOUN
fcis-28893	133	28	,	,	PUNCT
fcis-28893	133	29	which	which	PRON
fcis-28893	133	30	are	be	AUX
fcis-28893	133	31	mutually	mutually	ADV
fcis-28893	133	32	exclusive	exclusive	ADJ
fcis-28893	133	33	,	,	PUNCT
fcis-28893	133	34	while	while	SCONJ
fcis-28893	133	35	text	text	NOUN
fcis-28893	133	36	descriptions	description	NOUN
fcis-28893	133	37	are	be	AUX
fcis-28893	133	38	usually	usually	ADV
fcis-28893	133	39	mainly	mainly	ADV
fcis-28893	133	40	related	relate	VERB
fcis-28893	133	41	to	to	ADP
fcis-28893	133	42	character	character	NOUN
fcis-28893	133	43	information	information	NOUN
fcis-28893	133	44	.	.	PUNCT
fcis-28893	134	1	therefore	therefore	ADV
fcis-28893	134	2	,	,	PUNCT
fcis-28893	134	3	algorithms	algorithm	NOUN
fcis-28893	134	4	that	that	PRON
fcis-28893	134	5	accurately	accurately	ADV
fcis-28893	134	6	distinguish	distinguish	VERB
fcis-28893	134	7	character	character	NOUN
fcis-28893	134	8	and	and	CCONJ
fcis-28893	134	9	background	background	NOUN
fcis-28893	134	10	information	information	NOUN
fcis-28893	134	11	and	and	CCONJ
fcis-28893	134	12	filter	filter	VERB
fcis-28893	134	13	irrelevant	irrelevant	ADJ
fcis-28893	134	14	words	word	NOUN
fcis-28893	134	15	and	and	CCONJ
fcis-28893	134	16	syntactic	syntactic	ADJ
fcis-28893	134	17	noise	noise	NOUN
fcis-28893	134	18	in	in	ADP
fcis-28893	134	19	text	text	NOUN
fcis-28893	134	20	are	be	AUX
fcis-28893	134	21	particularly	particularly	ADV
fcis-28893	134	22	important	important	ADJ
fcis-28893	134	23	.	.	PUNCT
fcis-28893	135	1	to	to	ADP
fcis-28893	135	2	this	this	DET
fcis-28893	135	3	end	end	NOUN
fcis-28893	135	4	,	,	PUNCT
fcis-28893	135	5	the	the	DET
fcis-28893	135	6	authors	author	NOUN
fcis-28893	135	7	propose	propose	VERB
fcis-28893	135	8	the	the	DET
fcis-28893	135	9	deep	deep	ADJ
fcis-28893	135	10	background	background	NOUN
fcis-28893	135	11	-	-	PUNCT
fcis-28893	135	12	character	character	NOUN
fcis-28893	135	13	separation	separation	NOUN
fcis-28893	135	14	model	model	NOUN
fcis-28893	135	15	dssl	dssl	PROPN
fcis-28893	135	16	,	,	PUNCT
fcis-28893	135	17	which	which	PRON
fcis-28893	135	18	consists	consist	VERB
fcis-28893	135	19	of	of	ADP
fcis-28893	135	20	the	the	DET
fcis-28893	135	21	sdm	sdm	NOUN
fcis-28893	135	22	module	module	NOUN
fcis-28893	135	23	,	,	PUNCT
fcis-28893	135	24	the	the	DET
fcis-28893	135	25	spsm	spsm	NOUN
fcis-28893	135	26	module	module	NOUN
fcis-28893	135	27	,	,	PUNCT
fcis-28893	135	28	the	the	DET
fcis-28893	135	29	spfm	spfm	NOUN
fcis-28893	135	30	module	module	NOUN
fcis-28893	135	31	,	,	PUNCT
fcis-28893	135	32	the	the	DET
fcis-28893	135	33	pdm	pdm	NOUN
fcis-28893	135	34	module	module	NOUN
fcis-28893	135	35	,	,	PUNCT
fcis-28893	135	36	and	and	CCONJ
fcis-28893	135	37	the	the	DET
fcis-28893	135	38	sam	sam	PROPN
fcis-28893	135	39	module	module	NOUN
fcis-28893	135	40	.	.	PUNCT
fcis-28893	136	1	dssl	dssl	VERB
fcis-28893	136	2	first	first	ADJ
fcis-28893	136	3	extracts	extract	NOUN
fcis-28893	136	4	global	global	ADJ
fcis-28893	136	5	visual	visual	ADJ
fcis-28893	136	6	features	feature	VERB
fcis-28893	136	7	vg	vg	ADP
fcis-28893	136	8	,	,	PUNCT
fcis-28893	136	9	and	and	CCONJ
fcis-28893	136	10	local	local	ADJ
fcis-28893	136	11	visual	visual	ADJ
fcis-28893	136	12	feature	feature	NOUN
fcis-28893	136	13	matrix	matrix	NOUN
fcis-28893	136	14	mv	mv	X
fcis-28893	136	15	by	by	ADP
fcis-28893	136	16	resnet-50	resnet-50	PROPN
fcis-28893	136	17	,	,	PUNCT
fcis-28893	136	18	and	and	CCONJ
fcis-28893	136	19	the	the	DET
fcis-28893	136	20	text	text	NOUN
fcis-28893	136	21	part	part	NOUN
fcis-28893	136	22	uses	use	VERB
fcis-28893	136	23	bi	bi	ADJ
fcis-28893	136	24	-	-	ADJ
fcis-28893	136	25	directional	directional	ADJ
fcis-28893	136	26	gru	gru	NOUN
fcis-28893	136	27	to	to	PART
fcis-28893	136	28	obtain	obtain	VERB
fcis-28893	136	29	global	global	ADJ
fcis-28893	136	30	text	text	NOUN
fcis-28893	136	31	features	feature	VERB
fcis-28893	136	32	tg	tg	PROPN
fcis-28893	136	33	and	and	CCONJ
fcis-28893	136	34	local	local	ADJ
fcis-28893	136	35	text	text	NOUN
fcis-28893	136	36	feature	feature	NOUN
fcis-28893	136	37	matrix	matrix	NOUN
fcis-28893	136	38	mt	mt	PROPN
fcis-28893	136	39	.	.	PUNCT
fcis-28893	137	1	the	the	DET
fcis-28893	137	2	sdm	sdm	PROPN
fcis-28893	137	3	module	module	NOUN
fcis-28893	137	4	processes	process	VERB
fcis-28893	137	5	the	the	DET
fcis-28893	137	6	global	global	ADJ
fcis-28893	137	7	text	text	NOUN
fcis-28893	137	8	features	feature	NOUN
fcis-28893	137	9	and	and	CCONJ
fcis-28893	137	10	the	the	DET
fcis-28893	137	11	local	local	ADJ
fcis-28893	137	12	feature	feature	NOUN
fcis-28893	137	13	matrix	matrix	NOUN
fcis-28893	137	14	by	by	ADP
fcis-28893	137	15	setting	set	VERB
fcis-28893	137	16	a	a	DET
fcis-28893	137	17	fixed	fix	VERB
fcis-28893	137	18	number	number	NOUN
fcis-28893	137	19	of	of	ADP
fcis-28893	137	20	elements	element	NOUN
fcis-28893	137	21	in	in	ADP
fcis-28893	137	22	the	the	DET
fcis-28893	137	23	input	input	NOUN
fcis-28893	137	24	vectors	vector	NOUN
fcis-28893	137	25	to	to	ADP
fcis-28893	137	26	zero	zero	NUM
fcis-28893	137	27	with	with	ADP
fcis-28893	137	28	a	a	DET
fcis-28893	137	29	fixed	fix	VERB
fcis-28893	137	30	nulling	nulling	NOUN
fcis-28893	137	31	ratio	ratio	NOUN
fcis-28893	137	32	,	,	PUNCT
fcis-28893	137	33	and	and	CCONJ
fcis-28893	137	34	then	then	ADV
fcis-28893	137	35	reconstructs	reconstruct	VERB
fcis-28893	137	36	the	the	DET
fcis-28893	137	37	vectors	vector	NOUN
fcis-28893	137	38	in	in	ADP
fcis-28893	137	39	an	an	DET
fcis-28893	137	40	autoencoder	autoencoder	NOUN
fcis-28893	137	41	manner	manner	NOUN
fcis-28893	137	42	through	through	ADP
fcis-28893	137	43	ternary	ternary	ADJ
fcis-28893	137	44	permutation	permutation	NOUN
fcis-28893	137	45	loss	loss	NOUN
fcis-28893	137	46	function	function	NOUN
fcis-28893	137	47	optimization	optimization	NOUN
fcis-28893	137	48	to	to	PART
fcis-28893	137	49	obtain	obtain	VERB
fcis-28893	137	50	the	the	DET
fcis-28893	137	51	text	text	NOUN
fcis-28893	137	52	character	character	NOUN
fcis-28893	137	53	feature	feature	NOUN
fcis-28893	137	54	vector	vector	NOUN
fcis-28893	137	55	tp	tp	NOUN
fcis-28893	137	56	and	and	CCONJ
fcis-28893	137	57	the	the	DET
fcis-28893	137	58	local	local	ADJ
fcis-28893	137	59	text	text	NOUN
fcis-28893	137	60	character	character	NOUN
fcis-28893	137	61	feature	feature	NOUN
fcis-28893	137	62	matrix	matrix	NOUN
fcis-28893	137	63	tl	tl	PROPN
fcis-28893	137	64	,	,	PUNCT
fcis-28893	137	65	which	which	PRON
fcis-28893	137	66	fully	fully	ADV
fcis-28893	137	67	retains	retain	VERB
fcis-28893	137	68	the	the	DET
fcis-28893	137	69	valid	valid	ADJ
fcis-28893	137	70	information	information	NOUN
fcis-28893	137	71	while	while	SCONJ
fcis-28893	137	72	ignoring	ignore	VERB
fcis-28893	137	73	the	the	DET
fcis-28893	137	74	redundant	redundant	ADJ
fcis-28893	137	75	noise	noise	NOUN
fcis-28893	137	76	signals	signal	NOUN
fcis-28893	137	77	.	.	PUNCT
fcis-28893	138	1	the	the	DET
fcis-28893	138	2	spsm	spsm	NOUN
fcis-28893	138	3	module	module	NOUN
fcis-28893	138	4	separates	separate	VERB
fcis-28893	138	5	the	the	DET
fcis-28893	138	6	character	character	NOUN
fcis-28893	138	7	features	feature	NOUN
fcis-28893	138	8	from	from	ADP
fcis-28893	138	9	the	the	DET
fcis-28893	138	10	background	background	NOUN
fcis-28893	138	11	features	feature	VERB
fcis-28893	138	12	in	in	ADP
fcis-28893	138	13	the	the	DET
fcis-28893	138	14	image	image	NOUN
fcis-28893	138	15	to	to	PART
fcis-28893	138	16	obtain	obtain	VERB
fcis-28893	138	17	the	the	DET
fcis-28893	138	18	character	character	NOUN
fcis-28893	138	19	visual	visual	ADJ
fcis-28893	138	20	feature	feature	NOUN
fcis-28893	138	21	vp	vp	PROPN
fcis-28893	138	22	and	and	CCONJ
fcis-28893	138	23	the	the	DET
fcis-28893	138	24	environment	environment	NOUN
fcis-28893	138	25	visual	visual	ADJ
fcis-28893	138	26	feature	feature	NOUN
fcis-28893	138	27	vs	vs	ADP
fcis-28893	138	28	,	,	PUNCT
fcis-28893	138	29	while	while	SCONJ
fcis-28893	138	30	zhu	zhu	PROPN
fcis-28893	138	31	et	et	PROPN
fcis-28893	138	32	al	al	PROPN
fcis-28893	138	33	.	.	PROPN
fcis-28893	139	1	design	design	VERB
fcis-28893	139	2	the	the	DET
fcis-28893	139	3	mutual	mutual	ADJ
fcis-28893	139	4	exclusion	exclusion	NOUN
fcis-28893	139	5	loss	loss	NOUN
fcis-28893	139	6	function	function	NOUN
fcis-28893	139	7	according	accord	VERB
fcis-28893	139	8	to	to	ADP
fcis-28893	139	9	the	the	DET
fcis-28893	139	10	mutually	mutually	ADV
fcis-28893	139	11	exclusion	exclusion	NOUN
fcis-28893	139	12	constraint	constraint	NOUN
fcis-28893	139	13	(	(	PUNCT
fcis-28893	139	14	mec	mec	NOUN
fcis-28893	139	15	)	)	PUNCT
fcis-28893	139	16	strategy	strategy	NOUN
fcis-28893	139	17	to	to	PART
fcis-28893	139	18	ensure	ensure	VERB
fcis-28893	139	19	that	that	SCONJ
fcis-28893	139	20	the	the	DET
fcis-28893	139	21	character	character	NOUN
fcis-28893	139	22	and	and	CCONJ
fcis-28893	139	23	the	the	DET
fcis-28893	139	24	background	background	NOUN
fcis-28893	139	25	do	do	AUX
fcis-28893	139	26	not	not	PART
fcis-28893	139	27	overlap	overlap	VERB
fcis-28893	139	28	.	.	PUNCT
fcis-28893	140	1	the	the	DET
fcis-28893	140	2	spfm	spfm	NOUN
fcis-28893	140	3	module	module	NOUN
fcis-28893	140	4	,	,	PUNCT
fcis-28893	140	5	in	in	ADP
fcis-28893	140	6	the	the	DET
fcis-28893	140	7	form	form	NOUN
fcis-28893	140	8	of	of	ADP
fcis-28893	140	9	summation	summation	NOUN
fcis-28893	140	10	or	or	CCONJ
fcis-28893	140	11	concatenation	concatenation	NOUN
fcis-28893	140	12	,	,	PUNCT
fcis-28893	140	13	integrates	integrate	VERB
fcis-28893	140	14	the	the	DET
fcis-28893	140	15	text	text	NOUN
fcis-28893	140	16	feature	feature	NOUN
fcis-28893	140	17	tp	tp	NOUN
fcis-28893	140	18	and	and	CCONJ
fcis-28893	140	19	background	background	NOUN
fcis-28893	140	20	features	feature	VERB
fcis-28893	140	21	vs	vs	ADP
fcis-28893	140	22	fusion	fusion	NOUN
fcis-28893	140	23	,	,	PUNCT
fcis-28893	140	24	reconstructs	reconstruct	VERB
fcis-28893	140	25	the	the	DET
fcis-28893	140	26	visual	visual	ADJ
fcis-28893	140	27	modal	modal	NOUN
fcis-28893	140	28	features	feature	VERB
fcis-28893	140	29	vr	vr	NOUN
fcis-28893	140	30	containing	contain	VERB
fcis-28893	140	31	character	character	NOUN
fcis-28893	140	32	and	and	CCONJ
fcis-28893	140	33	background	background	NOUN
fcis-28893	140	34	information	information	NOUN
fcis-28893	140	35	through	through	ADP
fcis-28893	140	36	the	the	DET
fcis-28893	140	37	full	full	ADJ
fcis-28893	140	38	connectivity	connectivity	NOUN
fcis-28893	140	39	layer	layer	NOUN
fcis-28893	140	40	mlp	mlp	NOUN
fcis-28893	140	41	,	,	PUNCT
fcis-28893	140	42	and	and	CCONJ
fcis-28893	140	43	then	then	ADV
fcis-28893	140	44	aligns	align	VERB
fcis-28893	140	45	the	the	DET
fcis-28893	140	46	rejoined	rejoin	VERB
fcis-28893	140	47	visual	visual	ADJ
fcis-28893	140	48	modal	modal	NOUN
fcis-28893	140	49	features	feature	VERB
fcis-28893	140	50	vr	vr	PROPN
fcis-28893	140	51	with	with	ADP
fcis-28893	140	52	the	the	DET
fcis-28893	140	53	visual	visual	ADJ
fcis-28893	140	54	global	global	ADJ
fcis-28893	140	55	features	feature	NOUN
fcis-28893	140	56	vg	vg	PROPN
fcis-28893	140	57	.	.	PUNCT
fcis-28893	141	1	similarly	similarly	ADV
fcis-28893	141	2	,	,	PUNCT
fcis-28893	141	3	vp	vp	PROPN
fcis-28893	141	4	is	be	AUX
fcis-28893	141	5	rewired	rewire	VERB
fcis-28893	141	6	as	as	ADP
fcis-28893	141	7	a	a	DET
fcis-28893	141	8	textual	textual	ADJ
fcis-28893	141	9	modal	modal	NOUN
fcis-28893	141	10	feature	feature	NOUN
fcis-28893	141	11	tr	tr	VERB
fcis-28893	141	12	in	in	ADP
fcis-28893	141	13	the	the	DET
fcis-28893	141	14	pdm	pdm	NOUN
fcis-28893	141	15	module	module	NOUN
fcis-28893	141	16	by	by	ADP
fcis-28893	141	17	means	mean	NOUN
fcis-28893	141	18	of	of	ADP
fcis-28893	141	19	a	a	DET
fcis-28893	141	20	fully	fully	ADV
fcis-28893	141	21	connected	connect	VERB
fcis-28893	141	22	layer	layer	NOUN
fcis-28893	141	23	with	with	ADP
fcis-28893	141	24	a	a	DET
fcis-28893	141	25	tanh	tanh	NOUN
fcis-28893	141	26	activation	activation	NOUN
fcis-28893	141	27	function	function	NOUN
fcis-28893	141	28	,	,	PUNCT
fcis-28893	141	29	which	which	PRON
fcis-28893	141	30	is	be	AUX
fcis-28893	141	31	aligned	align	VERB
fcis-28893	141	32	with	with	ADP
fcis-28893	141	33	the	the	DET
fcis-28893	141	34	textual	textual	ADJ
fcis-28893	141	35	global	global	ADJ
fcis-28893	141	36	feature	feature	NOUN
fcis-28893	141	37	.	.	PUNCT
fcis-28893	142	1	sam	sam	PROPN
fcis-28893	142	2	first	first	ADV
fcis-28893	142	3	strengthens	strengthen	VERB
fcis-28893	142	4	the	the	DET
fcis-28893	142	5	character	character	NOUN
fcis-28893	142	6	information	information	NOUN
fcis-28893	142	7	in	in	ADP
fcis-28893	142	8	the	the	DET
fcis-28893	142	9	local	local	ADJ
fcis-28893	142	10	visual	visual	ADJ
fcis-28893	142	11	feature	feature	NOUN
fcis-28893	142	12	matrix	matrix	NOUN
fcis-28893	142	13	mv	mv	VERB
fcis-28893	142	14	to	to	PART
fcis-28893	142	15	obtain	obtain	VERB
fcis-28893	142	16	vl	vl	NOUN
fcis-28893	142	17	,	,	PUNCT
fcis-28893	142	18	and	and	CCONJ
fcis-28893	142	19	then	then	ADV
fcis-28893	142	20	aligns	align	VERB
fcis-28893	142	21	vl	vl	PROPN
fcis-28893	142	22	with	with	SCONJ
fcis-28893	142	23	the	the	DET
fcis-28893	142	24	textual	textual	ADJ
fcis-28893	142	25	character	character	NOUN
fcis-28893	142	26	features	feature	VERB
fcis-28893	142	27	tg	tg	PROPN
fcis-28893	142	28	through	through	ADP
fcis-28893	142	29	the	the	DET
fcis-28893	142	30	cross	cross	ADJ
fcis-28893	142	31	-	-	ADJ
fcis-28893	142	32	modal	modal	ADJ
fcis-28893	142	33	attention	attention	NOUN
fcis-28893	142	34	mechanism	mechanism	NOUN
fcis-28893	142	35	to	to	PART
fcis-28893	142	36	further	far	ADV
fcis-28893	142	37	enhance	enhance	VERB
fcis-28893	142	38	the	the	DET
fcis-28893	142	39	recognition	recognition	NOUN
fcis-28893	142	40	ability	ability	NOUN
fcis-28893	142	41	of	of	ADP
fcis-28893	142	42	the	the	DET
fcis-28893	142	43	model	model	NOUN
fcis-28893	142	44	.	.	PUNCT
fcis-28893	143	1	the	the	DET
fcis-28893	143	2	fifth	fifth	ADJ
fcis-28893	143	3	module	module	NOUN
fcis-28893	143	4	,	,	PUNCT
fcis-28893	143	5	on	on	ADP
fcis-28893	143	6	the	the	DET
fcis-28893	143	7	other	other	ADJ
fcis-28893	143	8	hand	hand	NOUN
fcis-28893	143	9	,	,	PUNCT
fcis-28893	143	10	is	be	AUX
fcis-28893	143	11	like	like	ADP
fcis-28893	143	12	the	the	DET
fcis-28893	143	13	fourth	fourth	ADJ
fcis-28893	143	14	module	module	NOUN
fcis-28893	143	15	,	,	PUNCT
fcis-28893	143	16	which	which	PRON
fcis-28893	143	17	mainly	mainly	ADV
fcis-28893	143	18	strengthens	strengthen	VERB
fcis-28893	143	19	the	the	DET
fcis-28893	143	20	character	character	NOUN
fcis-28893	143	21	information	information	NOUN
fcis-28893	143	22	in	in	ADP
fcis-28893	143	23	the	the	DET
fcis-28893	143	24	local	local	ADJ
fcis-28893	143	25	text	text	NOUN
fcis-28893	143	26	feature	feature	NOUN
fcis-28893	143	27	matrix	matrix	NOUN
fcis-28893	143	28	.	.	PUNCT
fcis-28893	144	1	dssl	dssl	PROPN
fcis-28893	144	2	divides	divide	VERB
fcis-28893	144	3	the	the	DET
fcis-28893	144	4	text	text	NOUN
fcis-28893	144	5	image	image	NOUN
fcis-28893	144	6	into	into	ADP
fcis-28893	144	7	character	character	NOUN
fcis-28893	144	8	part	part	NOUN
fcis-28893	144	9	and	and	CCONJ
fcis-28893	144	10	background	background	NOUN
fcis-28893	144	11	part	part	NOUN
fcis-28893	144	12	,	,	PUNCT
fcis-28893	144	13	and	and	CCONJ
fcis-28893	144	14	through	through	ADP
fcis-28893	144	15	segmentation	segmentation	NOUN
fcis-28893	144	16	,	,	PUNCT
fcis-28893	144	17	the	the	DET
fcis-28893	144	18	model	model	NOUN
fcis-28893	144	19	focuses	focus	VERB
fcis-28893	144	20	on	on	ADP
fcis-28893	144	21	the	the	DET
fcis-28893	144	22	character	character	NOUN
fcis-28893	144	23	information	information	NOUN
fcis-28893	144	24	in	in	ADP
fcis-28893	144	25	the	the	DET
fcis-28893	144	26	image	image	NOUN
fcis-28893	144	27	and	and	CCONJ
fcis-28893	144	28	text	text	NOUN
fcis-28893	144	29	,	,	PUNCT
fcis-28893	144	30	and	and	CCONJ
fcis-28893	144	31	filters	filter	VERB
fcis-28893	144	32	irrelevant	irrelevant	ADJ
fcis-28893	144	33	information	information	NOUN
fcis-28893	144	34	and	and	CCONJ
fcis-28893	144	35	noise	noise	NOUN
fcis-28893	144	36	in	in	ADP
fcis-28893	144	37	the	the	DET
fcis-28893	144	38	graphic	graphic	ADJ
fcis-28893	144	39	text	text	NOUN
fcis-28893	144	40	.	.	PUNCT
fcis-28893	145	1	however	however	ADV
fcis-28893	145	2	,	,	PUNCT
fcis-28893	145	3	dssl	dssl	PROPN
fcis-28893	145	4	requires	require	VERB
fcis-28893	145	5	multiple	multiple	ADJ
fcis-28893	145	6	modules	module	NOUN
fcis-28893	145	7	to	to	PART
fcis-28893	145	8	differentiate	differentiate	VERB
fcis-28893	145	9	characters	character	NOUN
fcis-28893	145	10	from	from	ADP
fcis-28893	145	11	the	the	DET
fcis-28893	145	12	environment	environment	NOUN
fcis-28893	145	13	and	and	CCONJ
fcis-28893	145	14	reconstruct	reconstruct	VERB
fcis-28893	145	15	graphic	graphic	ADJ
fcis-28893	145	16	features	feature	NOUN
fcis-28893	145	17	,	,	PUNCT
fcis-28893	145	18	which	which	PRON
fcis-28893	145	19	is	be	AUX
fcis-28893	145	20	a	a	DET
fcis-28893	145	21	cumbersome	cumbersome	ADJ
fcis-28893	145	22	and	and	CCONJ
fcis-28893	145	23	complex	complex	ADJ
fcis-28893	145	24	process	process	NOUN
fcis-28893	145	25	with	with	ADP
fcis-28893	145	26	high	high	ADJ
fcis-28893	145	27	training	training	NOUN
fcis-28893	145	28	overhead	overhead	NOUN
fcis-28893	145	29	.	.	PUNCT
fcis-28893	146	1	acsa	acsa	PROPN
fcis-28893	146	2	asymmetric	asymmetric	ADJ
fcis-28893	146	3	cross	cross	ADJ
fcis-28893	146	4	-	-	ADJ
fcis-28893	146	5	scale	scale	ADJ
fcis-28893	146	6	alignment	alignment	NOUN
fcis-28893	146	7	includes	include	VERB
fcis-28893	146	8	a	a	DET
fcis-28893	146	9	global	global	ADJ
fcis-28893	146	10	alignment	alignment	NOUN
fcis-28893	146	11	module	module	NOUN
fcis-28893	146	12	and	and	CCONJ
fcis-28893	146	13	an	an	DET
fcis-28893	146	14	asymmetric	asymmetric	ADJ
fcis-28893	146	15	cross	cross	ADJ
fcis-28893	146	16	-	-	ADJ
fcis-28893	146	17	attention	attention	ADJ
fcis-28893	146	18	module	module	NOUN
fcis-28893	146	19	.	.	PUNCT
fcis-28893	147	1	the	the	DET
fcis-28893	147	2	global	global	ADJ
fcis-28893	147	3	alignment	alignment	NOUN
fcis-28893	147	4	module	module	NOUN
fcis-28893	147	5	uses	use	VERB
fcis-28893	147	6	the	the	DET
fcis-28893	147	7	cmpm	cmpm	PROPN
fcis-28893	147	8	loss	loss	NOUN
fcis-28893	147	9	function	function	NOUN
fcis-28893	147	10	proposed	propose	VERB
fcis-28893	147	11	by	by	ADP
fcis-28893	147	12	zhang	zhang	PROPN
fcis-28893	147	13	et	et	PROPN
fcis-28893	147	14	al.[7	al.[7	PROPN
fcis-28893	147	15	]	]	PUNCT
fcis-28893	147	16	as	as	ADV
fcis-28893	147	17	well	well	ADV
fcis-28893	147	18	as	as	ADP
fcis-28893	147	19	the	the	DET
fcis-28893	147	20	cmpc	cmpc	PROPN
fcis-28893	147	21	loss	loss	NOUN
fcis-28893	147	22	function	function	NOUN
fcis-28893	147	23	to	to	PART
fcis-28893	147	24	perform	perform	VERB
fcis-28893	147	25	global	global	ADJ
fcis-28893	147	26	alignment	alignment	NOUN
fcis-28893	147	27	of	of	ADP
fcis-28893	147	28	graphic	graphic	ADJ
fcis-28893	147	29	and	and	CCONJ
fcis-28893	147	30	textual	textual	ADJ
fcis-28893	147	31	features	feature	NOUN
fcis-28893	147	32	and	and	CCONJ
fcis-28893	147	33	the	the	DET
fcis-28893	147	34	alignment	alignment	NOUN
fcis-28893	147	35	of	of	ADP
fcis-28893	147	36	images	image	NOUN
fcis-28893	147	37	or	or	CCONJ
fcis-28893	147	38	local	local	ADJ
fcis-28893	147	39	regions	region	NOUN
fcis-28893	147	40	with	with	ADP
fcis-28893	147	41	text	text	NOUN
fcis-28893	147	42	phrases	phrase	NOUN
fcis-28893	147	43	to	to	PART
fcis-28893	147	44	accomplish	accomplish	VERB
fcis-28893	147	45	the	the	DET
fcis-28893	147	46	first	first	ADJ
fcis-28893	147	47	and	and	CCONJ
fcis-28893	147	48	second	second	ADJ
fcis-28893	147	49	alignment	alignment	NOUN
fcis-28893	147	50	,	,	PUNCT
fcis-28893	147	51	and	and	CCONJ
fcis-28893	147	52	the	the	DET
fcis-28893	147	53	specific	specific	ADJ
fcis-28893	147	54	alignment	alignment	NOUN
fcis-28893	147	55	process	process	NOUN
fcis-28893	147	56	is	be	AUX
fcis-28893	147	57	similar	similar	ADJ
fcis-28893	147	58	to	to	ADP
fcis-28893	147	59	the	the	DET
fcis-28893	147	60	method	method	NOUN
fcis-28893	147	61	proposed	propose	VERB
fcis-28893	147	62	by	by	ADP
fcis-28893	147	63	zhang	zhang	PROPN
fcis-28893	147	64	et	et	PROPN
fcis-28893	147	65	al.[7	al.[7	PROPN
fcis-28893	147	66	]	]	PUNCT
fcis-28893	147	67	.	.	PUNCT
fcis-28893	148	1	the	the	DET
fcis-28893	148	2	asymmetric	asymmetric	ADJ
fcis-28893	148	3	cross	cross	ADJ
fcis-28893	148	4	-	-	ADJ
fcis-28893	148	5	modal	modal	ADJ
fcis-28893	148	6	attention	attention	NOUN
fcis-28893	148	7	module	module	NOUN
fcis-28893	148	8	is	be	AUX
fcis-28893	148	9	optimized	optimize	VERB
fcis-28893	148	10	by	by	ADP
fcis-28893	148	11	the	the	DET
fcis-28893	148	12	asymmetric	asymmetric	ADJ
fcis-28893	148	13	cross	cross	ADJ
fcis-28893	148	14	-	-	ADJ
fcis-28893	148	15	scale	scale	ADJ
fcis-28893	148	16	alignment	alignment	NOUN
fcis-28893	148	17	loss	loss	NOUN
fcis-28893	148	18	function	function	NOUN
fcis-28893	148	19	,	,	PUNCT
fcis-28893	148	20	which	which	PRON
fcis-28893	148	21	is	be	AUX
fcis-28893	148	22	responsible	responsible	ADJ
fcis-28893	148	23	for	for	ADP
fcis-28893	148	24	realizing	realize	VERB
fcis-28893	148	25	the	the	DET
fcis-28893	148	26	alignment	alignment	NOUN
fcis-28893	148	27	between	between	ADP
fcis-28893	148	28	local	local	ADJ
fcis-28893	148	29	phrases	phrase	NOUN
fcis-28893	148	30	and	and	CCONJ
fcis-28893	148	31	local	local	ADJ
fcis-28893	148	32	images	image	NOUN
fcis-28893	148	33	,	,	PUNCT
fcis-28893	148	34	and	and	CCONJ
fcis-28893	148	35	contains	contain	VERB
fcis-28893	148	36	two	two	NUM
fcis-28893	148	37	sets	set	NOUN
fcis-28893	148	38	of	of	ADP
fcis-28893	148	39	inputs	input	NOUN
fcis-28893	148	40	:	:	PUNCT
fcis-28893	148	41	the	the	DET
fcis-28893	148	42	visual	visual	ADJ
fcis-28893	148	43	part	part	NOUN
fcis-28893	148	44	connects	connect	VERB
fcis-28893	148	45	the	the	DET
fcis-28893	148	46	global	global	ADJ
fcis-28893	148	47	image	image	NOUN
fcis-28893	148	48	features	feature	VERB
fcis-28893	148	49	with	with	ADP
fcis-28893	148	50	the	the	DET
fcis-28893	148	51	regional	regional	ADJ
fcis-28893	148	52	features	feature	NOUN
fcis-28893	148	53	to	to	PART
fcis-28893	148	54	obtain	obtain	VERB
fcis-28893	148	55	a	a	DET
fcis-28893	148	56	set	set	NOUN
fcis-28893	148	57	of	of	ADP
fcis-28893	148	58	multi	multi	ADJ
fcis-28893	148	59	-	-	ADJ
fcis-28893	148	60	scale	scale	ADJ
fcis-28893	148	61	visual	visual	ADJ
fcis-28893	148	62	feature	feature	NOUN
fcis-28893	148	63	vectors	vector	NOUN
fcis-28893	148	64	as	as	ADP
fcis-28893	148	65	the	the	DET
fcis-28893	148	66	input	input	NOUN
fcis-28893	148	67	of	of	ADP
fcis-28893	148	68	the	the	DET
fcis-28893	148	69	visual	visual	ADJ
fcis-28893	148	70	part	part	NOUN
fcis-28893	148	71	;	;	PUNCT
fcis-28893	148	72	the	the	DET
fcis-28893	148	73	text	text	NOUN
fcis-28893	148	74	part	part	NOUN
fcis-28893	148	75	takes	take	VERB
fcis-28893	148	76	a	a	DET
fcis-28893	148	77	set	set	NOUN
fcis-28893	148	78	of	of	ADP
fcis-28893	148	79	noun	noun	NOUN
fcis-28893	148	80	phrase	phrase	NOUN
fcis-28893	148	81	feature	feature	NOUN
fcis-28893	148	82	vectors	vector	NOUN
fcis-28893	148	83	as	as	ADP
fcis-28893	148	84	the	the	DET
fcis-28893	148	85	text	text	NOUN
fcis-28893	148	86	input	input	NOUN
fcis-28893	148	87	.	.	PUNCT
fcis-28893	149	1	the	the	DET
fcis-28893	149	2	cross	cross	ADJ
fcis-28893	149	3	-	-	ADJ
fcis-28893	149	4	modal	modal	ADJ
fcis-28893	149	5	attention	attention	NOUN
fcis-28893	149	6	mechanism	mechanism	NOUN
fcis-28893	149	7	is	be	AUX
fcis-28893	149	8	utilized	utilize	VERB
fcis-28893	149	9	to	to	PART
fcis-28893	149	10	calculate	calculate	VERB
fcis-28893	149	11	and	and	CCONJ
fcis-28893	149	12	output	output	VERB
fcis-28893	149	13	the	the	DET
fcis-28893	149	14	similarity	similarity	NOUN
fcis-28893	149	15	scores	score	NOUN
fcis-28893	149	16	between	between	ADP
fcis-28893	149	17	pairs	pair	NOUN
fcis-28893	149	18	of	of	ADP
fcis-28893	149	19	graphical	graphical	ADJ
fcis-28893	149	20	and	and	CCONJ
fcis-28893	149	21	textual	textual	ADJ
fcis-28893	149	22	local	local	ADJ
fcis-28893	149	23	features	feature	NOUN
fcis-28893	149	24	.	.	PUNCT
fcis-28893	150	1	acsa	acsa	VERB
fcis-28893	150	2	automatically	automatically	ADV
fcis-28893	150	3	learns	learn	VERB
fcis-28893	150	4	multi	multi	ADJ
fcis-28893	150	5	-	-	ADJ
fcis-28893	150	6	scale	scale	ADJ
fcis-28893	150	7	alignment	alignment	NOUN
fcis-28893	150	8	through	through	ADP
fcis-28893	150	9	the	the	DET
fcis-28893	150	10	cross	cross	ADJ
fcis-28893	150	11	-	-	ADJ
fcis-28893	150	12	modal	modal	ADJ
fcis-28893	150	13	attention	attention	NOUN
fcis-28893	150	14	mechanism	mechanism	NOUN
fcis-28893	150	15	and	and	CCONJ
fcis-28893	150	16	is	be	AUX
fcis-28893	150	17	no	no	ADV
fcis-28893	150	18	longer	long	ADV
fcis-28893	150	19	limited	limit	VERB
fcis-28893	150	20	to	to	ADP
fcis-28893	150	21	a	a	DET
fcis-28893	150	22	specific	specific	ADJ
fcis-28893	150	23	scale	scale	NOUN
fcis-28893	150	24	range	range	NOUN
fcis-28893	150	25	.	.	PUNCT
fcis-28893	151	1	acsa	acsa	PROPN
fcis-28893	151	2	further	far	ADV
fcis-28893	151	3	improves	improve	VERB
fcis-28893	151	4	the	the	DET
fcis-28893	151	5	multiscale	multiscale	ADJ
fcis-28893	151	6	alignment	alignment	NOUN
fcis-28893	151	7	mechanism	mechanism	NOUN
fcis-28893	151	8	and	and	CCONJ
fcis-28893	151	9	significantly	significantly	ADV
fcis-28893	151	10	improves	improve	VERB
fcis-28893	151	11	the	the	DET
fcis-28893	151	12	retrieval	retrieval	ADJ
fcis-28893	151	13	accuracy	accuracy	NOUN
fcis-28893	151	14	.	.	PUNCT
fcis-28893	152	1	4.5	4.5	NUM
fcis-28893	152	2	.	.	PUNCT
fcis-28893	152	3	methods	method	NOUN
fcis-28893	152	4	based	base	VERB
fcis-28893	152	5	on	on	ADP
fcis-28893	152	6	vlp	vlp	PROPN
fcis-28893	152	7	yan	yan	PROPN
fcis-28893	152	8	et	et	PROPN
fcis-28893	152	9	al.[15	al.[15	PROPN
fcis-28893	152	10	]	]	PUNCT
fcis-28893	152	11	recognized	recognize	VERB
fcis-28893	152	12	the	the	DET
fcis-28893	152	13	power	power	NOUN
fcis-28893	152	14	of	of	ADP
fcis-28893	152	15	the	the	DET
fcis-28893	152	16	clip	clip	NOUN
fcis-28893	152	17	model[16	model[16	PROPN
fcis-28893	152	18	]	]	PUNCT
fcis-28893	152	19	in	in	ADP
fcis-28893	152	20	semantic	semantic	ADJ
fcis-28893	152	21	learning	learning	NOUN
fcis-28893	152	22	and	and	CCONJ
fcis-28893	152	23	cross	cross	ADJ
fcis-28893	152	24	-	-	ADJ
fcis-28893	152	25	modal	modal	ADJ
fcis-28893	152	26	knowledge	knowledge	NOUN
fcis-28893	152	27	extraction	extraction	NOUN
fcis-28893	152	28	,	,	PUNCT
fcis-28893	152	29	and	and	CCONJ
fcis-28893	152	30	proposed	propose	VERB
fcis-28893	152	31	cfine	cfine	NOUN
fcis-28893	152	32	,	,	PUNCT
fcis-28893	152	33	a	a	DET
fcis-28893	152	34	clip	clip	NOUN
fcis-28893	152	35	-	-	PUNCT
fcis-28893	152	36	based	base	VERB
fcis-28893	152	37	framework	framework	NOUN
fcis-28893	152	38	for	for	ADP
fcis-28893	152	39	finegrained	finegraine	VERB
fcis-28893	152	40	information	information	NOUN
fcis-28893	152	41	extraction	extraction	NOUN
fcis-28893	152	42	.	.	PUNCT
fcis-28893	153	1	the	the	DET
fcis-28893	153	2	clip	clip	NOUN
fcis-28893	153	3	model	model	NOUN
fcis-28893	153	4	is	be	AUX
fcis-28893	153	5	mainly	mainly	ADV
fcis-28893	153	6	124	124	NUM
fcis-28893	153	7	focused	focus	VERB
fcis-28893	153	8	on	on	ADP
fcis-28893	153	9	individual	individual	ADJ
fcis-28893	153	10	-	-	PUNCT
fcis-28893	153	11	level	level	NOUN
fcis-28893	153	12	features	feature	NOUN
fcis-28893	153	13	,	,	PUNCT
fcis-28893	153	14	which	which	PRON
fcis-28893	153	15	may	may	AUX
fcis-28893	153	16	lead	lead	VERB
fcis-28893	153	17	to	to	ADP
fcis-28893	153	18	poor	poor	ADJ
fcis-28893	153	19	results	result	NOUN
fcis-28893	153	20	when	when	SCONJ
fcis-28893	153	21	directly	directly	ADV
fcis-28893	153	22	applied	apply	VERB
fcis-28893	153	23	to	to	ADP
fcis-28893	153	24	new	new	ADJ
fcis-28893	153	25	tasks	task	NOUN
fcis-28893	153	26	.	.	PUNCT
fcis-28893	154	1	the	the	DET
fcis-28893	154	2	cfine	cfine	NOUN
fcis-28893	154	3	framework	framework	NOUN
fcis-28893	154	4	consists	consist	VERB
fcis-28893	154	5	of	of	ADP
fcis-28893	154	6	two	two	NUM
fcis-28893	154	7	main	main	ADJ
fcis-28893	154	8	components	component	NOUN
fcis-28893	154	9	:	:	PUNCT
fcis-28893	154	10	a	a	DET
fcis-28893	154	11	modalityspecific	modalityspecific	ADJ
fcis-28893	154	12	feature	feature	NOUN
fcis-28893	154	13	extraction	extraction	NOUN
fcis-28893	154	14	module	module	NOUN
fcis-28893	154	15	and	and	CCONJ
fcis-28893	154	16	fine	fine	ADV
fcis-28893	154	17	-	-	PUNCT
fcis-28893	154	18	grained	grain	VERB
fcis-28893	154	19	information	information	NOUN
fcis-28893	154	20	mining	mining	NOUN
fcis-28893	154	21	module	module	NOUN
fcis-28893	154	22	.	.	PUNCT
fcis-28893	155	1	in	in	ADP
fcis-28893	155	2	order	order	NOUN
fcis-28893	155	3	to	to	PART
fcis-28893	155	4	avoid	avoid	VERB
fcis-28893	155	5	intra	intra	ADJ
fcis-28893	155	6	-	-	ADJ
fcis-28893	155	7	modal	modal	ADJ
fcis-28893	155	8	information	information	NOUN
fcis-28893	155	9	distortion	distortion	NOUN
fcis-28893	155	10	,	,	PUNCT
fcis-28893	155	11	the	the	DET
fcis-28893	155	12	cfine	cfine	NOUN
fcis-28893	155	13	model	model	NOUN
fcis-28893	155	14	only	only	ADV
fcis-28893	155	15	uses	use	VERB
fcis-28893	155	16	the	the	DET
fcis-28893	155	17	encoder	encoder	NOUN
fcis-28893	155	18	of	of	ADP
fcis-28893	155	19	the	the	DET
fcis-28893	155	20	clip	clip	NOUN
fcis-28893	155	21	model	model	NOUN
fcis-28893	155	22	for	for	ADP
fcis-28893	155	23	the	the	DET
fcis-28893	155	24	image	image	NOUN
fcis-28893	155	25	part	part	NOUN
fcis-28893	155	26	,	,	PUNCT
fcis-28893	155	27	while	while	SCONJ
fcis-28893	155	28	the	the	DET
fcis-28893	155	29	text	text	NOUN
fcis-28893	155	30	part	part	NOUN
fcis-28893	155	31	directly	directly	ADV
fcis-28893	155	32	uses	use	VERB
fcis-28893	155	33	the	the	DET
fcis-28893	155	34	pre	pre	ADJ
fcis-28893	155	35	-	-	ADJ
fcis-28893	155	36	trained	train	VERB
fcis-28893	155	37	bert	bert	PROPN
fcis-28893	155	38	model	model	NOUN
fcis-28893	155	39	.	.	PUNCT
fcis-28893	156	1	however	however	ADV
fcis-28893	156	2	,	,	PUNCT
fcis-28893	156	3	the	the	DET
fcis-28893	156	4	graphical	graphical	ADJ
fcis-28893	156	5	and	and	CCONJ
fcis-28893	156	6	textual	textual	ADJ
fcis-28893	156	7	features	feature	NOUN
fcis-28893	156	8	extracted	extract	VERB
fcis-28893	156	9	by	by	ADP
fcis-28893	156	10	clip	clip	NOUN
fcis-28893	156	11	image	image	NOUN
fcis-28893	156	12	encoder	encoder	NOUN
fcis-28893	156	13	and	and	CCONJ
fcis-28893	156	14	bert	bert	PROPN
fcis-28893	156	15	only	only	ADV
fcis-28893	156	16	contain	contain	VERB
fcis-28893	156	17	individual	individual	ADJ
fcis-28893	156	18	dimensional	dimensional	ADJ
fcis-28893	156	19	modal	modal	ADJ
fcis-28893	156	20	information	information	NOUN
fcis-28893	156	21	and	and	CCONJ
fcis-28893	156	22	cross	cross	ADJ
fcis-28893	156	23	-	-	ADJ
fcis-28893	156	24	modal	modal	ADJ
fcis-28893	156	25	correspondences	correspondence	NOUN
fcis-28893	156	26	,	,	PUNCT
fcis-28893	156	27	but	but	CCONJ
fcis-28893	156	28	lack	lack	VERB
fcis-28893	156	29	the	the	DET
fcis-28893	156	30	crucial	crucial	ADJ
fcis-28893	156	31	fine	fine	ADV
fcis-28893	156	32	-	-	PUNCT
fcis-28893	156	33	grained	grain	VERB
fcis-28893	156	34	information	information	NOUN
fcis-28893	156	35	,	,	PUNCT
fcis-28893	156	36	so	so	ADV
fcis-28893	156	37	yan	yan	PROPN
fcis-28893	156	38	et	et	PROPN
fcis-28893	156	39	al	al	PROPN
fcis-28893	156	40	.	.	PROPN
fcis-28893	156	41	further	further	PROPN
fcis-28893	156	42	designed	design	VERB
fcis-28893	156	43	the	the	DET
fcis-28893	156	44	fine	fine	ADV
fcis-28893	156	45	-	-	PUNCT
fcis-28893	156	46	grained	grain	VERB
fcis-28893	156	47	information	information	NOUN
fcis-28893	156	48	mining	mining	NOUN
fcis-28893	156	49	module	module	NOUN
fcis-28893	156	50	to	to	PART
fcis-28893	156	51	fully	fully	ADV
fcis-28893	156	52	explore	explore	VERB
fcis-28893	156	53	the	the	DET
fcis-28893	156	54	fine	fine	ADV
fcis-28893	156	55	-	-	PUNCT
fcis-28893	156	56	grained	grain	VERB
fcis-28893	156	57	correspondences	correspondence	NOUN
fcis-28893	156	58	.	.	PUNCT
fcis-28893	157	1	the	the	DET
fcis-28893	157	2	fine	fine	ADV
fcis-28893	157	3	-	-	PUNCT
fcis-28893	157	4	grained	grain	VERB
fcis-28893	157	5	information	information	NOUN
fcis-28893	157	6	mining	mining	NOUN
fcis-28893	157	7	module	module	NOUN
fcis-28893	157	8	is	be	AUX
fcis-28893	157	9	divided	divide	VERB
fcis-28893	157	10	into	into	ADP
fcis-28893	157	11	three	three	NUM
fcis-28893	157	12	sub	sub	NOUN
fcis-28893	157	13	-	-	NOUN
fcis-28893	157	14	modules	module	NOUN
fcis-28893	157	15	:	:	PUNCT
fcis-28893	157	16	first	first	ADV
fcis-28893	157	17	,	,	PUNCT
fcis-28893	157	18	the	the	DET
fcis-28893	157	19	multi	multi	ADJ
fcis-28893	157	20	-	-	ADJ
fcis-28893	157	21	granular	granular	ADJ
fcis-28893	157	22	global	global	ADJ
fcis-28893	157	23	feature	feature	NOUN
fcis-28893	157	24	learning	learning	NOUN
fcis-28893	157	25	(	(	PUNCT
fcis-28893	157	26	mgf	mgf	PROPN
fcis-28893	157	27	)	)	PUNCT
fcis-28893	157	28	module	module	NOUN
fcis-28893	157	29	is	be	AUX
fcis-28893	157	30	responsible	responsible	ADJ
fcis-28893	157	31	for	for	ADP
fcis-28893	157	32	mining	mining	NOUN
fcis-28893	157	33	potential	potential	ADJ
fcis-28893	157	34	clues	clue	NOUN
fcis-28893	157	35	related	relate	VERB
fcis-28893	157	36	to	to	ADP
fcis-28893	157	37	individual	individual	ADJ
fcis-28893	157	38	identity	identity	NOUN
fcis-28893	157	39	,	,	PUNCT
fcis-28893	157	40	executing	execute	VERB
fcis-28893	157	41	the	the	DET
fcis-28893	157	42	token	token	ADJ
fcis-28893	157	43	selection	selection	NOUN
fcis-28893	157	44	process	process	NOUN
fcis-28893	157	45	according	accord	VERB
fcis-28893	157	46	to	to	ADP
fcis-28893	157	47	the	the	DET
fcis-28893	157	48	transformer	transformer	NOUN
fcis-28893	157	49	's	's	PART
fcis-28893	157	50	strategy	strategy	NOUN
fcis-28893	157	51	for	for	ADP
fcis-28893	157	52	text	text	NOUN
fcis-28893	157	53	and	and	CCONJ
fcis-28893	157	54	images	image	NOUN
fcis-28893	157	55	,	,	PUNCT
fcis-28893	157	56	respectively	respectively	ADV
fcis-28893	157	57	,	,	PUNCT
fcis-28893	157	58	and	and	CCONJ
fcis-28893	157	59	then	then	ADV
fcis-28893	157	60	inputting	inputte	VERB
fcis-28893	157	61	the	the	DET
fcis-28893	157	62	token	token	NOUN
fcis-28893	157	63	into	into	ADP
fcis-28893	157	64	the	the	DET
fcis-28893	157	65	global	global	ADJ
fcis-28893	157	66	-	-	PUNCT
fcis-28893	157	67	local	local	ADJ
fcis-28893	157	68	decoder	decoder	NOUN
fcis-28893	157	69	,	,	PUNCT
fcis-28893	157	70	which	which	PRON
fcis-28893	157	71	utilizes	utilize	VERB
fcis-28893	157	72	the	the	DET
fcis-28893	157	73	multi	multi	ADJ
fcis-28893	157	74	-	-	ADJ
fcis-28893	157	75	head	head	ADJ
fcis-28893	157	76	self	self	NOUN
fcis-28893	157	77	-	-	PUNCT
fcis-28893	157	78	attention	attention	NOUN
fcis-28893	157	79	layer	layer	NOUN
fcis-28893	157	80	to	to	PART
fcis-28893	157	81	make	make	VERB
fcis-28893	157	82	the	the	DET
fcis-28893	157	83	token	token	ADJ
fcis-28893	157	84	information	information	NOUN
fcis-28893	157	85	pass	pass	VERB
fcis-28893	157	86	into	into	ADP
fcis-28893	157	87	the	the	DET
fcis-28893	157	88	classification	classification	NOUN
fcis-28893	157	89	token	token	VERB
fcis-28893	157	90	,	,	PUNCT
fcis-28893	157	91	after	after	ADP
fcis-28893	157	92	which	which	PRON
fcis-28893	157	93	the	the	DET
fcis-28893	157	94	multi	multi	ADJ
fcis-28893	157	95	-	-	ADJ
fcis-28893	157	96	head	head	ADJ
fcis-28893	157	97	cross	cross	ADJ
fcis-28893	157	98	-	-	ADJ
fcis-28893	157	99	attention	attention	ADJ
fcis-28893	157	100	mechanism	mechanism	NOUN
fcis-28893	157	101	intervenes	intervene	VERB
fcis-28893	157	102	to	to	PART
fcis-28893	157	103	highlight	highlight	VERB
fcis-28893	157	104	the	the	DET
fcis-28893	157	105	token	token	ADJ
fcis-28893	157	106	information	information	NOUN
fcis-28893	157	107	and	and	CCONJ
fcis-28893	157	108	contextual	contextual	ADJ
fcis-28893	157	109	information	information	NOUN
fcis-28893	157	110	.	.	PUNCT
fcis-28893	158	1	cross	cross	ADJ
fcis-28893	158	2	-	-	ADJ
fcis-28893	158	3	granularity	granularity	NOUN
fcis-28893	158	4	feature	feature	NOUN
fcis-28893	158	5	optimization	optimization	NOUN
fcis-28893	158	6	(	(	PUNCT
fcis-28893	158	7	cfr	cfr	NOUN
fcis-28893	158	8	)	)	PUNCT
fcis-28893	158	9	module	module	NOUN
fcis-28893	158	10	removes	remove	VERB
fcis-28893	158	11	the	the	DET
fcis-28893	158	12	non	non	ADJ
fcis-28893	158	13	-	-	ADJ
fcis-28893	158	14	modal	modal	ADJ
fcis-28893	158	15	sharing	sharing	NOUN
fcis-28893	158	16	information	information	NOUN
fcis-28893	158	17	and	and	CCONJ
fcis-28893	158	18	initially	initially	ADV
fcis-28893	158	19	establishes	establish	VERB
fcis-28893	158	20	the	the	DET
fcis-28893	158	21	cross	cross	ADJ
fcis-28893	158	22	-	-	ADJ
fcis-28893	158	23	modal	modal	ADJ
fcis-28893	158	24	linkage	linkage	NOUN
fcis-28893	158	25	,	,	PUNCT
fcis-28893	158	26	firstly	firstly	ADV
fcis-28893	158	27	,	,	PUNCT
fcis-28893	158	28	the	the	DET
fcis-28893	158	29	similarity	similarity	NOUN
fcis-28893	158	30	between	between	ADP
fcis-28893	158	31	image	image	NOUN
fcis-28893	158	32	-	-	PUNCT
fcis-28893	158	33	word	word	NOUN
fcis-28893	158	34	and	and	CCONJ
fcis-28893	158	35	sentence	sentence	NOUN
fcis-28893	158	36	-	-	PUNCT
fcis-28893	158	37	image	image	NOUN
fcis-28893	158	38	blocks	block	NOUN
fcis-28893	158	39	is	be	AUX
fcis-28893	158	40	evaluated	evaluate	VERB
fcis-28893	158	41	by	by	ADP
fcis-28893	158	42	inner	inner	ADJ
fcis-28893	158	43	product	product	NOUN
fcis-28893	158	44	,	,	PUNCT
fcis-28893	158	45	and	and	CCONJ
fcis-28893	158	46	the	the	DET
fcis-28893	158	47	similarity	similarity	NOUN
fcis-28893	158	48	of	of	ADP
fcis-28893	158	49	individual	individual	ADJ
fcis-28893	158	50	dimensions	dimension	NOUN
fcis-28893	158	51	is	be	AUX
fcis-28893	158	52	obtained	obtain	VERB
fcis-28893	158	53	by	by	ADP
fcis-28893	158	54	fusion	fusion	NOUN
fcis-28893	158	55	,	,	PUNCT
fcis-28893	158	56	and	and	CCONJ
fcis-28893	158	57	the	the	DET
fcis-28893	158	58	weights	weight	NOUN
fcis-28893	158	59	are	be	AUX
fcis-28893	158	60	adaptively	adaptively	ADV
fcis-28893	158	61	assigned	assign	VERB
fcis-28893	158	62	by	by	ADP
fcis-28893	158	63	softmax	softmax	NOUN
fcis-28893	158	64	function	function	NOUN
fcis-28893	158	65	during	during	ADP
fcis-28893	158	66	the	the	DET
fcis-28893	158	67	process	process	NOUN
fcis-28893	158	68	,	,	PUNCT
fcis-28893	158	69	and	and	CCONJ
fcis-28893	158	70	the	the	DET
fcis-28893	158	71	cross	cross	ADJ
fcis-28893	158	72	-	-	ADJ
fcis-28893	158	73	modal	modal	ADJ
fcis-28893	158	74	correspondences	correspondence	NOUN
fcis-28893	158	75	between	between	ADP
fcis-28893	158	76	the	the	DET
fcis-28893	158	77	image	image	NOUN
fcis-28893	158	78	and	and	CCONJ
fcis-28893	158	79	the	the	DET
fcis-28893	158	80	text	text	NOUN
fcis-28893	158	81	have	have	AUX
fcis-28893	158	82	been	be	AUX
fcis-28893	158	83	roughly	roughly	ADV
fcis-28893	158	84	established	establish	VERB
fcis-28893	158	85	after	after	ADP
fcis-28893	158	86	passing	pass	VERB
fcis-28893	158	87	through	through	ADP
fcis-28893	158	88	the	the	DET
fcis-28893	158	89	cfr	cfr	NOUN
fcis-28893	158	90	module	module	NOUN
fcis-28893	158	91	.	.	PUNCT
fcis-28893	159	1	the	the	DET
fcis-28893	159	2	fine	fine	ADV
fcis-28893	159	3	-	-	PUNCT
fcis-28893	159	4	grained	grain	VERB
fcis-28893	159	5	connection	connection	NOUN
fcis-28893	159	6	discovery	discovery	NOUN
fcis-28893	159	7	(	(	PUNCT
fcis-28893	159	8	fcd	fcd	PROPN
fcis-28893	159	9	)	)	PUNCT
fcis-28893	159	10	module	module	NOUN
fcis-28893	159	11	is	be	AUX
fcis-28893	159	12	used	use	VERB
fcis-28893	159	13	to	to	PART
fcis-28893	159	14	establish	establish	VERB
fcis-28893	159	15	fine	fine	ADV
fcis-28893	159	16	-	-	PUNCT
fcis-28893	159	17	grained	grain	VERB
fcis-28893	159	18	connections	connection	NOUN
fcis-28893	159	19	between	between	ADP
fcis-28893	159	20	modalities	modality	NOUN
fcis-28893	159	21	,	,	PUNCT
fcis-28893	159	22	after	after	ADP
fcis-28893	159	23	going	go	VERB
fcis-28893	159	24	through	through	ADP
fcis-28893	159	25	the	the	DET
fcis-28893	159	26	mgf	mgf	PROPN
fcis-28893	159	27	and	and	CCONJ
fcis-28893	159	28	cfr	cfr	PROPN
fcis-28893	159	29	modules	module	NOUN
fcis-28893	159	30	,	,	PUNCT
fcis-28893	159	31	the	the	DET
fcis-28893	159	32	fcd	fcd	NOUN
fcis-28893	159	33	module	module	NOUN
fcis-28893	159	34	has	have	AUX
fcis-28893	159	35	made	make	VERB
fcis-28893	159	36	explicit	explicit	ADJ
fcis-28893	159	37	the	the	DET
fcis-28893	159	38	image	image	NOUN
fcis-28893	159	39	blocks	block	NOUN
fcis-28893	159	40	and	and	CCONJ
fcis-28893	159	41	text	text	NOUN
fcis-28893	159	42	phrases	phrase	NOUN
fcis-28893	159	43	that	that	PRON
fcis-28893	159	44	contain	contain	VERB
fcis-28893	159	45	the	the	DET
fcis-28893	159	46	most	most	ADJ
fcis-28893	159	47	information	information	NOUN
fcis-28893	159	48	,	,	PUNCT
fcis-28893	159	49	and	and	CCONJ
fcis-28893	159	50	the	the	DET
fcis-28893	159	51	fcd	fcd	NOUN
fcis-28893	159	52	calculates	calculate	VERB
fcis-28893	159	53	the	the	DET
fcis-28893	159	54	pre	pre	NOUN
fcis-28893	159	55	-	-	NOUN
fcis-28893	159	56	similarity	similarity	NOUN
fcis-28893	159	57	between	between	ADP
fcis-28893	159	58	the	the	DET
fcis-28893	159	59	two	two	NUM
fcis-28893	159	60	,	,	PUNCT
fcis-28893	159	61	and	and	CCONJ
fcis-28893	159	62	then	then	ADV
fcis-28893	159	63	the	the	DET
fcis-28893	159	64	most	most	ADV
fcis-28893	159	65	relevant	relevant	ADJ
fcis-28893	159	66	image	image	NOUN
fcis-28893	159	67	or	or	CCONJ
fcis-28893	159	68	text	text	NOUN
fcis-28893	159	69	features	feature	NOUN
fcis-28893	159	70	are	be	AUX
fcis-28893	159	71	selected	select	VERB
fcis-28893	159	72	using	use	VERB
fcis-28893	159	73	the	the	DET
fcis-28893	159	74	top	top	ADJ
fcis-28893	159	75	-	-	PUNCT
fcis-28893	159	76	k	k	NOUN
fcis-28893	159	77	operation	operation	NOUN
fcis-28893	159	78	,	,	PUNCT
fcis-28893	159	79	and	and	CCONJ
fcis-28893	159	80	the	the	DET
fcis-28893	159	81	average	average	ADJ
fcis-28893	159	82	pooling	pooling	NOUN
fcis-28893	159	83	results	result	NOUN
fcis-28893	159	84	in	in	ADP
fcis-28893	159	85	matching	match	VERB
fcis-28893	159	86	graphic	graphic	ADJ
fcis-28893	159	87	and	and	CCONJ
fcis-28893	159	88	text	text	NOUN
fcis-28893	159	89	feature	feature	NOUN
fcis-28893	159	90	pairs	pair	NOUN
fcis-28893	159	91	.	.	PUNCT
fcis-28893	160	1	to	to	PART
fcis-28893	160	2	bring	bring	VERB
fcis-28893	160	3	these	these	DET
fcis-28893	160	4	graphic	graphic	ADJ
fcis-28893	160	5	and	and	CCONJ
fcis-28893	160	6	text	text	NOUN
fcis-28893	160	7	feature	feature	NOUN
fcis-28893	160	8	pairs	pair	NOUN
fcis-28893	160	9	closer	close	ADV
fcis-28893	160	10	together	together	ADV
fcis-28893	160	11	in	in	ADP
fcis-28893	160	12	the	the	DET
fcis-28893	160	13	feature	feature	NOUN
fcis-28893	160	14	space	space	NOUN
fcis-28893	160	15	,	,	PUNCT
fcis-28893	160	16	fcd	fcd	PROPN
fcis-28893	160	17	again	again	ADV
fcis-28893	160	18	calculates	calculate	VERB
fcis-28893	160	19	the	the	DET
fcis-28893	160	20	individual	individual	ADJ
fcis-28893	160	21	dimensional	dimensional	ADJ
fcis-28893	160	22	similarity	similarity	NOUN
fcis-28893	160	23	by	by	ADP
fcis-28893	160	24	cosine	cosine	NOUN
fcis-28893	160	25	similarity	similarity	NOUN
fcis-28893	160	26	.	.	PUNCT
fcis-28893	161	1	cfine	cfine	PROPN
fcis-28893	161	2	avoids	avoid	VERB
fcis-28893	161	3	the	the	DET
fcis-28893	161	4	problem	problem	NOUN
fcis-28893	161	5	of	of	ADP
fcis-28893	161	6	intra	intra	ADJ
fcis-28893	161	7	-	-	ADJ
fcis-28893	161	8	modal	modal	ADJ
fcis-28893	161	9	information	information	NOUN
fcis-28893	161	10	distortion	distortion	NOUN
fcis-28893	161	11	that	that	PRON
fcis-28893	161	12	may	may	AUX
fcis-28893	161	13	arise	arise	VERB
fcis-28893	161	14	from	from	ADP
fcis-28893	161	15	the	the	DET
fcis-28893	161	16	direct	direct	ADJ
fcis-28893	161	17	use	use	NOUN
fcis-28893	161	18	of	of	ADP
fcis-28893	161	19	the	the	DET
fcis-28893	161	20	clip	clip	NOUN
fcis-28893	161	21	model	model	NOUN
fcis-28893	161	22	,	,	PUNCT
fcis-28893	161	23	and	and	CCONJ
fcis-28893	161	24	does	do	AUX
fcis-28893	161	25	not	not	PART
fcis-28893	161	26	introduce	introduce	VERB
fcis-28893	161	27	additional	additional	ADJ
fcis-28893	161	28	embedding	embed	VERB
fcis-28893	161	29	layers	layer	NOUN
fcis-28893	161	30	,	,	PUNCT
fcis-28893	161	31	supports	support	VERB
fcis-28893	161	32	end	end	NOUN
fcis-28893	161	33	-	-	PUNCT
fcis-28893	161	34	to	to	ADP
fcis-28893	161	35	-	-	PUNCT
fcis-28893	161	36	end	end	NOUN
fcis-28893	161	37	optimization	optimization	NOUN
fcis-28893	161	38	,	,	PUNCT
fcis-28893	161	39	and	and	CCONJ
fcis-28893	161	40	achieves	achieve	VERB
fcis-28893	161	41	sota	sota	NOUN
fcis-28893	161	42	scores	score	NOUN
fcis-28893	161	43	on	on	ADP
fcis-28893	161	44	all	all	DET
fcis-28893	161	45	three	three	NUM
fcis-28893	161	46	datasets	dataset	NOUN
fcis-28893	161	47	,	,	PUNCT
fcis-28893	161	48	which	which	PRON
fcis-28893	161	49	verifies	verify	VERB
fcis-28893	161	50	the	the	DET
fcis-28893	161	51	advantages	advantage	NOUN
fcis-28893	161	52	of	of	ADP
fcis-28893	161	53	the	the	DET
fcis-28893	161	54	vlp	vlp	PROPN
fcis-28893	161	55	model	model	NOUN
fcis-28893	161	56	,	,	PUNCT
fcis-28893	161	57	and	and	CCONJ
fcis-28893	161	58	strengthens	strengthen	VERB
fcis-28893	161	59	the	the	DET
fcis-28893	161	60	foundation	foundation	NOUN
fcis-28893	161	61	for	for	ADP
fcis-28893	161	62	the	the	DET
fcis-28893	161	63	continued	continue	VERB
fcis-28893	161	64	use	use	NOUN
fcis-28893	161	65	of	of	ADP
fcis-28893	161	66	the	the	DET
fcis-28893	161	67	vlp	vlp	PROPN
fcis-28893	161	68	model	model	NOUN
fcis-28893	161	69	in	in	ADP
fcis-28893	161	70	the	the	DET
fcis-28893	161	71	domain	domain	NOUN
fcis-28893	161	72	.	.	PUNCT
fcis-28893	162	1	however	however	ADV
fcis-28893	162	2	,	,	PUNCT
fcis-28893	162	3	cfine	cfine	VERB
fcis-28893	162	4	only	only	ADV
fcis-28893	162	5	uses	use	VERB
fcis-28893	162	6	the	the	DET
fcis-28893	162	7	image	image	NOUN
fcis-28893	162	8	encoder	encoder	NOUN
fcis-28893	162	9	of	of	ADP
fcis-28893	162	10	the	the	DET
fcis-28893	162	11	clip	clip	NOUN
fcis-28893	162	12	model	model	NOUN
fcis-28893	162	13	,	,	PUNCT
fcis-28893	162	14	which	which	PRON
fcis-28893	162	15	limits	limit	VERB
fcis-28893	162	16	the	the	DET
fcis-28893	162	17	crossmodal	crossmodal	ADJ
fcis-28893	162	18	learning	learning	NOUN
fcis-28893	162	19	ability	ability	NOUN
fcis-28893	162	20	of	of	ADP
fcis-28893	162	21	the	the	DET
fcis-28893	162	22	clip	clip	NOUN
fcis-28893	162	23	model	model	NOUN
fcis-28893	162	24	and	and	CCONJ
fcis-28893	162	25	fails	fail	VERB
fcis-28893	162	26	to	to	PART
fcis-28893	162	27	give	give	VERB
fcis-28893	162	28	full	full	ADJ
fcis-28893	162	29	play	play	NOUN
fcis-28893	162	30	to	to	ADP
fcis-28893	162	31	the	the	DET
fcis-28893	162	32	semantic	semantic	ADJ
fcis-28893	162	33	understanding	understanding	NOUN
fcis-28893	162	34	ability	ability	NOUN
fcis-28893	162	35	of	of	ADP
fcis-28893	162	36	the	the	DET
fcis-28893	162	37	clip	clip	NOUN
fcis-28893	162	38	model	model	NOUN
fcis-28893	162	39	through	through	ADP
fcis-28893	162	40	the	the	DET
fcis-28893	162	41	pre	pre	NOUN
fcis-28893	162	42	-	-	NOUN
fcis-28893	162	43	training	training	NOUN
fcis-28893	162	44	of	of	ADP
fcis-28893	162	45	hundreds	hundred	NOUN
fcis-28893	162	46	of	of	ADP
fcis-28893	162	47	millions	million	NOUN
fcis-28893	162	48	of	of	ADP
fcis-28893	162	49	graphic	graphic	ADJ
fcis-28893	162	50	datasets	dataset	NOUN
fcis-28893	162	51	.	.	PUNCT
fcis-28893	163	1	jiang	jiang	PROPN
fcis-28893	163	2	et	et	PROPN
fcis-28893	163	3	al.[17	al.[17	PROPN
fcis-28893	163	4	]	]	PUNCT
fcis-28893	163	5	pointed	point	VERB
fcis-28893	163	6	out	out	ADP
fcis-28893	163	7	two	two	NUM
fcis-28893	163	8	major	major	ADJ
fcis-28893	163	9	problems	problem	NOUN
fcis-28893	163	10	with	with	ADP
fcis-28893	163	11	previous	previous	ADJ
fcis-28893	163	12	approaches	approach	NOUN
fcis-28893	163	13	using	use	VERB
fcis-28893	163	14	the	the	DET
fcis-28893	163	15	vlp	vlp	PROPN
fcis-28893	163	16	model	model	NOUN
fcis-28893	163	17	:	:	PUNCT
fcis-28893	163	18	one	one	NUM
fcis-28893	163	19	,	,	PUNCT
fcis-28893	163	20	the	the	DET
fcis-28893	163	21	lack	lack	NOUN
fcis-28893	163	22	of	of	ADP
fcis-28893	163	23	necessary	necessary	ADJ
fcis-28893	163	24	underlying	underlying	ADJ
fcis-28893	163	25	alignment	alignment	NOUN
fcis-28893	163	26	capabilities	capability	NOUN
fcis-28893	163	27	,	,	PUNCT
fcis-28893	163	28	which	which	PRON
fcis-28893	163	29	leads	lead	VERB
fcis-28893	163	30	to	to	ADP
fcis-28893	163	31	insufficient	insufficient	ADJ
fcis-28893	163	32	efficiency	efficiency	NOUN
fcis-28893	163	33	in	in	ADP
fcis-28893	163	34	multi	multi	ADJ
fcis-28893	163	35	-	-	ADJ
fcis-28893	163	36	modal	modal	ADJ
fcis-28893	163	37	data	datum	NOUN
fcis-28893	163	38	matching	matching	NOUN
fcis-28893	163	39	;	;	PUNCT
fcis-28893	163	40	and	and	CCONJ
fcis-28893	163	41	two	two	NUM
fcis-28893	163	42	,	,	PUNCT
fcis-28893	163	43	the	the	DET
fcis-28893	163	44	reliance	reliance	NOUN
fcis-28893	163	45	on	on	ADP
fcis-28893	163	46	too	too	ADV
fcis-28893	163	47	much	much	ADJ
fcis-28893	163	48	a	a	DET
fcis-28893	163	49	priori	priori	ADJ
fcis-28893	163	50	knowledge	knowledge	NOUN
fcis-28893	163	51	to	to	PART
fcis-28893	163	52	explore	explore	VERB
fcis-28893	163	53	explicit	explicit	ADJ
fcis-28893	163	54	alignment	alignment	NOUN
fcis-28893	163	55	patterns	pattern	NOUN
fcis-28893	163	56	,	,	PUNCT
fcis-28893	163	57	which	which	PRON
fcis-28893	163	58	may	may	AUX
fcis-28893	163	59	trigger	trigger	VERB
fcis-28893	163	60	distortion	distortion	NOUN
fcis-28893	163	61	of	of	ADP
fcis-28893	163	62	intra	intra	ADJ
fcis-28893	163	63	-	-	ADJ
fcis-28893	163	64	modal	modal	ADJ
fcis-28893	163	65	information	information	NOUN
fcis-28893	163	66	.	.	PUNCT
fcis-28893	164	1	to	to	PART
fcis-28893	164	2	address	address	VERB
fcis-28893	164	3	these	these	DET
fcis-28893	164	4	issues	issue	NOUN
fcis-28893	164	5	,	,	PUNCT
fcis-28893	164	6	jiang	jiang	PROPN
fcis-28893	164	7	et	et	PROPN
fcis-28893	164	8	al	al	PROPN
fcis-28893	164	9	.	.	PROPN
fcis-28893	164	10	proposed	propose	VERB
fcis-28893	164	11	the	the	DET
fcis-28893	164	12	irra	irra	ADJ
fcis-28893	164	13	framework	framework	NOUN
fcis-28893	164	14	based	base	VERB
fcis-28893	164	15	on	on	ADP
fcis-28893	164	16	the	the	DET
fcis-28893	164	17	clip	clip	NOUN
fcis-28893	164	18	model	model	NOUN
fcis-28893	164	19	.	.	PUNCT
fcis-28893	165	1	unlike	unlike	ADP
fcis-28893	165	2	previous	previous	ADJ
fcis-28893	165	3	approaches	approach	NOUN
fcis-28893	165	4	,	,	PUNCT
fcis-28893	165	5	irra	irra	NOUN
fcis-28893	165	6	utilizes	utilize	VERB
fcis-28893	165	7	the	the	DET
fcis-28893	165	8	full	full	ADJ
fcis-28893	165	9	clip	clip	NOUN
fcis-28893	165	10	model	model	NOUN
fcis-28893	165	11	.	.	PUNCT
fcis-28893	166	1	the	the	DET
fcis-28893	166	2	irr	irr	NOUN
fcis-28893	166	3	module	module	NOUN
fcis-28893	166	4	employs	employ	VERB
fcis-28893	166	5	mlm	mlm	PROPN
fcis-28893	166	6	,	,	PUNCT
fcis-28893	166	7	a	a	DET
fcis-28893	166	8	text	text	NOUN
fcis-28893	166	9	masking	masking	NOUN
fcis-28893	166	10	strategy	strategy	NOUN
fcis-28893	166	11	widely	widely	ADV
fcis-28893	166	12	known	know	VERB
fcis-28893	166	13	for	for	ADP
fcis-28893	166	14	bert	bert	NOUN
fcis-28893	166	15	,	,	PUNCT
fcis-28893	166	16	to	to	PART
fcis-28893	166	17	optimize	optimize	VERB
fcis-28893	166	18	both	both	DET
fcis-28893	166	19	image	image	NOUN
fcis-28893	166	20	and	and	CCONJ
fcis-28893	166	21	context	context	NOUN
fcis-28893	166	22	feature	feature	NOUN
fcis-28893	166	23	vector	vector	NOUN
fcis-28893	166	24	alignment	alignment	NOUN
fcis-28893	166	25	and	and	CCONJ
fcis-28893	166	26	static	static	ADJ
fcis-28893	166	27	vector	vector	NOUN
fcis-28893	166	28	consistency	consistency	NOUN
fcis-28893	166	29	in	in	ADP
fcis-28893	166	30	the	the	DET
fcis-28893	166	31	joint	joint	ADJ
fcis-28893	166	32	embedding	embed	VERB
fcis-28893	166	33	space	space	NOUN
fcis-28893	166	34	through	through	ADP
fcis-28893	166	35	mlm	mlm	PROPN
fcis-28893	166	36	.	.	PUNCT
fcis-28893	167	1	to	to	PART
fcis-28893	167	2	realize	realize	VERB
fcis-28893	167	3	cross	cross	ADJ
fcis-28893	167	4	-	-	ADJ
fcis-28893	167	5	modal	modal	ADJ
fcis-28893	167	6	interaction	interaction	NOUN
fcis-28893	167	7	between	between	ADP
fcis-28893	167	8	images	image	NOUN
fcis-28893	167	9	and	and	CCONJ
fcis-28893	167	10	texts	text	NOUN
fcis-28893	167	11	,	,	PUNCT
fcis-28893	167	12	jiang	jiang	PROPN
fcis-28893	167	13	et	et	PROPN
fcis-28893	167	14	al	al	PROPN
fcis-28893	167	15	.	.	PROPN
fcis-28893	167	16	designed	design	VERB
fcis-28893	167	17	a	a	DET
fcis-28893	167	18	computationally	computationally	ADV
fcis-28893	167	19	more	more	ADV
fcis-28893	167	20	efficient	efficient	ADJ
fcis-28893	167	21	multimodal	multimodal	NOUN
fcis-28893	167	22	interaction	interaction	NOUN
fcis-28893	167	23	encoder	encoder	NOUN
fcis-28893	167	24	.	.	PUNCT
fcis-28893	168	1	the	the	DET
fcis-28893	168	2	encoder	encoder	NOUN
fcis-28893	168	3	consists	consist	VERB
fcis-28893	168	4	of	of	ADP
fcis-28893	168	5	a	a	DET
fcis-28893	168	6	multi	multi	ADJ
fcis-28893	168	7	-	-	ADJ
fcis-28893	168	8	head	head	ADJ
fcis-28893	168	9	cross	cross	NOUN
fcis-28893	168	10	attention	attention	NOUN
fcis-28893	168	11	layer	layer	NOUN
fcis-28893	168	12	(	(	PUNCT
fcis-28893	168	13	mca	mca	PROPN
fcis-28893	168	14	)	)	PUNCT
fcis-28893	168	15	with	with	ADP
fcis-28893	168	16	a	a	DET
fcis-28893	168	17	four	four	NUM
fcis-28893	168	18	-	-	PUNCT
fcis-28893	168	19	layer	layer	NOUN
fcis-28893	168	20	transformer	transformer	NOUN
fcis-28893	168	21	block	block	NOUN
fcis-28893	168	22	.	.	PUNCT
fcis-28893	169	1	for	for	ADP
fcis-28893	169	2	the	the	DET
fcis-28893	169	3	input	input	NOUN
fcis-28893	169	4	text	text	NOUN
fcis-28893	169	5	,	,	PUNCT
fcis-28893	169	6	the	the	DET
fcis-28893	169	7	text	text	NOUN
fcis-28893	169	8	encoding	encoding	NOUN
fcis-28893	169	9	features	feature	NOUN
fcis-28893	169	10	are	be	AUX
fcis-28893	169	11	masked	mask	VERB
fcis-28893	169	12	according	accord	VERB
fcis-28893	169	13	to	to	ADP
fcis-28893	169	14	the	the	DET
fcis-28893	169	15	bert	bert	PROPN
fcis-28893	169	16	's	's	PART
fcis-28893	169	17	strategy	strategy	NOUN
fcis-28893	169	18	mlm	mlm	PROPN
fcis-28893	169	19	,	,	PUNCT
fcis-28893	169	20	and	and	CCONJ
fcis-28893	169	21	then	then	ADV
fcis-28893	169	22	the	the	DET
fcis-28893	169	23	original	original	ADJ
fcis-28893	169	24	features	feature	NOUN
fcis-28893	169	25	of	of	ADP
fcis-28893	169	26	the	the	DET
fcis-28893	169	27	masked	mask	VERB
fcis-28893	169	28	parts	part	NOUN
fcis-28893	169	29	are	be	AUX
fcis-28893	169	30	predicted	predict	VERB
fcis-28893	169	31	by	by	ADP
fcis-28893	169	32	the	the	DET
fcis-28893	169	33	mlp	mlp	NOUN
fcis-28893	169	34	layer	layer	NOUN
fcis-28893	169	35	classifier	classifier	NOUN
fcis-28893	169	36	.	.	PUNCT
fcis-28893	170	1	through	through	ADP
fcis-28893	170	2	self	self	NOUN
fcis-28893	170	3	-	-	PUNCT
fcis-28893	170	4	attention	attention	NOUN
fcis-28893	170	5	and	and	CCONJ
fcis-28893	170	6	cross	cross	ADJ
fcis-28893	170	7	-	-	ADJ
fcis-28893	170	8	modal	modal	ADJ
fcis-28893	170	9	attention	attention	NOUN
fcis-28893	170	10	mechanisms	mechanism	NOUN
fcis-28893	170	11	,	,	PUNCT
fcis-28893	170	12	textual	textual	ADJ
fcis-28893	170	13	and	and	CCONJ
fcis-28893	170	14	visual	visual	ADJ
fcis-28893	170	15	features	feature	NOUN
fcis-28893	170	16	are	be	AUX
fcis-28893	170	17	first	first	ADV
fcis-28893	170	18	fused	fuse	VERB
fcis-28893	170	19	at	at	ADP
fcis-28893	170	20	the	the	DET
fcis-28893	170	21	cross	cross	ADJ
fcis-28893	170	22	-	-	ADJ
fcis-28893	170	23	attention	attention	ADJ
fcis-28893	170	24	layer	layer	NOUN
fcis-28893	170	25	before	before	ADP
fcis-28893	170	26	being	be	AUX
fcis-28893	170	27	input	input	NOUN
fcis-28893	170	28	into	into	ADP
fcis-28893	170	29	a	a	DET
fcis-28893	170	30	single	single	ADJ
fcis-28893	170	31	transformer	transformer	NOUN
fcis-28893	170	32	block	block	NOUN
fcis-28893	170	33	.	.	PUNCT
fcis-28893	171	1	the	the	DET
fcis-28893	171	2	multimodal	multimodal	ADJ
fcis-28893	171	3	interactive	interactive	ADJ
fcis-28893	171	4	encoder	encoder	NOUN
fcis-28893	171	5	establishes	establish	VERB
fcis-28893	171	6	the	the	DET
fcis-28893	171	7	link	link	NOUN
fcis-28893	171	8	between	between	ADP
fcis-28893	171	9	image	image	NOUN
fcis-28893	171	10	and	and	CCONJ
fcis-28893	171	11	text	text	NOUN
fcis-28893	171	12	without	without	ADP
fcis-28893	171	13	additional	additional	ADJ
fcis-28893	171	14	supervision	supervision	NOUN
fcis-28893	171	15	or	or	CCONJ
fcis-28893	171	16	inference	inference	NOUN
fcis-28893	171	17	overhead	overhead	ADV
fcis-28893	171	18	.	.	PUNCT
fcis-28893	172	1	in	in	ADP
fcis-28893	172	2	order	order	NOUN
fcis-28893	172	3	to	to	PART
fcis-28893	172	4	explore	explore	VERB
fcis-28893	172	5	more	more	ADV
fcis-28893	172	6	effective	effective	ADJ
fcis-28893	172	7	cross	cross	ADJ
fcis-28893	172	8	-	-	ADJ
fcis-28893	172	9	modal	modal	ADJ
fcis-28893	172	10	matching	matching	NOUN
fcis-28893	172	11	,	,	PUNCT
fcis-28893	172	12	jiang	jiang	PROPN
fcis-28893	172	13	et	et	PROPN
fcis-28893	172	14	al	al	PROPN
fcis-28893	172	15	.	.	PROPN
fcis-28893	172	16	designed	design	VERB
fcis-28893	172	17	the	the	DET
fcis-28893	172	18	sdm	sdm	PROPN
fcis-28893	172	19	loss	loss	NOUN
fcis-28893	172	20	loss	loss	NOUN
fcis-28893	172	21	function	function	NOUN
fcis-28893	172	22	to	to	PART
fcis-28893	172	23	combine	combine	VERB
fcis-28893	172	24	the	the	DET
fcis-28893	172	25	cosine	cosine	NOUN
fcis-28893	172	26	similarity	similarity	NOUN
fcis-28893	172	27	distributions	distribution	NOUN
fcis-28893	172	28	of	of	ADP
fcis-28893	172	29	image	image	NOUN
fcis-28893	172	30	-	-	PUNCT
fcis-28893	172	31	text	text	NOUN
fcis-28893	172	32	pair	pair	NOUN
fcis-28893	172	33	vectors	vector	NOUN
fcis-28893	172	34	into	into	ADP
fcis-28893	172	35	the	the	DET
fcis-28893	172	36	kl	kl	PROPN
fcis-28893	172	37	dispersion	dispersion	NOUN
fcis-28893	172	38	to	to	PART
fcis-28893	172	39	correlate	correlate	VERB
fcis-28893	172	40	the	the	DET
fcis-28893	172	41	representations	representation	NOUN
fcis-28893	172	42	of	of	ADP
fcis-28893	172	43	the	the	DET
fcis-28893	172	44	different	different	ADJ
fcis-28893	172	45	modalities	modality	NOUN
fcis-28893	172	46	,	,	PUNCT
fcis-28893	172	47	and	and	CCONJ
fcis-28893	172	48	into	into	ADP
fcis-28893	172	49	the	the	DET
fcis-28893	172	50	temperature	temperature	NOUN
fcis-28893	172	51	hyperparameter	hyperparameter	NOUN
fcis-28893	172	52	to	to	PART
fcis-28893	172	53	precisely	precisely	ADV
fcis-28893	172	54	control	control	VERB
fcis-28893	172	55	the	the	DET
fcis-28893	172	56	compactness	compactness	NOUN
fcis-28893	172	57	of	of	ADP
fcis-28893	172	58	the	the	DET
fcis-28893	172	59	similarity	similarity	NOUN
fcis-28893	172	60	distributions	distribution	NOUN
fcis-28893	172	61	,	,	PUNCT
fcis-28893	172	62	so	so	SCONJ
fcis-28893	172	63	that	that	SCONJ
fcis-28893	172	64	the	the	DET
fcis-28893	172	65	model	model	NOUN
fcis-28893	172	66	focuses	focus	VERB
fcis-28893	172	67	on	on	ADP
fcis-28893	172	68	the	the	DET
fcis-28893	172	69	difficult	difficult	ADJ
fcis-28893	172	70	negative	negative	ADJ
fcis-28893	172	71	samples	sample	NOUN
fcis-28893	172	72	,	,	PUNCT
fcis-28893	172	73	and	and	CCONJ
fcis-28893	172	74	at	at	ADP
fcis-28893	172	75	the	the	DET
fcis-28893	172	76	same	same	ADJ
fcis-28893	172	77	time	time	NOUN
fcis-28893	172	78	,	,	PUNCT
fcis-28893	172	79	expands	expand	VERB
fcis-28893	172	80	the	the	DET
fcis-28893	172	81	variance	variance	NOUN
fcis-28893	172	82	between	between	ADP
fcis-28893	172	83	non	non	ADJ
fcis-28893	172	84	-	-	ADJ
fcis-28893	172	85	matching	matching	ADJ
fcis-28893	172	86	pairs	pair	NOUN
fcis-28893	172	87	and	and	CCONJ
fcis-28893	172	88	the	the	DET
fcis-28893	172	89	correlation	correlation	NOUN
fcis-28893	172	90	between	between	ADP
fcis-28893	172	91	matching	match	VERB
fcis-28893	172	92	pairs	pair	NOUN
fcis-28893	172	93	.	.	PUNCT
fcis-28893	173	1	compared	compare	VERB
fcis-28893	173	2	to	to	PART
fcis-28893	173	3	cfine	cfine	VERB
fcis-28893	173	4	,	,	PUNCT
fcis-28893	173	5	irr	irr	PROPN
fcis-28893	173	6	uses	use	VERB
fcis-28893	173	7	the	the	DET
fcis-28893	173	8	full	full	ADJ
fcis-28893	173	9	clip	clip	NOUN
fcis-28893	173	10	,	,	PUNCT
fcis-28893	173	11	and	and	CCONJ
fcis-28893	173	12	the	the	DET
fcis-28893	173	13	introduction	introduction	NOUN
fcis-28893	173	14	of	of	ADP
fcis-28893	173	15	the	the	DET
fcis-28893	173	16	mlm	mlm	NOUN
fcis-28893	173	17	masking	masking	NOUN
fcis-28893	173	18	strategy	strategy	NOUN
fcis-28893	173	19	enhances	enhance	VERB
fcis-28893	173	20	irra	irra	VERB
fcis-28893	173	21	's	's	PART
fcis-28893	173	22	understanding	understanding	NOUN
fcis-28893	173	23	of	of	ADP
fcis-28893	173	24	contextual	contextual	ADJ
fcis-28893	173	25	semantic	semantic	ADJ
fcis-28893	173	26	information	information	NOUN
fcis-28893	173	27	.	.	PUNCT
fcis-28893	174	1	irra	irra	VERB
fcis-28893	174	2	significantly	significantly	ADV
fcis-28893	174	3	outperforms	outperform	VERB
fcis-28893	174	4	the	the	DET
fcis-28893	174	5	previous	previous	ADJ
fcis-28893	174	6	sota	sota	NOUN
fcis-28893	174	7	model	model	NOUN
fcis-28893	174	8	on	on	ADP
fcis-28893	174	9	several	several	ADJ
fcis-28893	174	10	datasets	dataset	NOUN
fcis-28893	174	11	,	,	PUNCT
fcis-28893	174	12	demonstrating	demonstrate	VERB
fcis-28893	174	13	superior	superior	ADJ
fcis-28893	174	14	performance	performance	NOUN
fcis-28893	174	15	.	.	PUNCT
fcis-28893	175	1	tan	tan	PROPN
fcis-28893	175	2	et	et	NOUN
fcis-28893	176	1	al.[18	al.[18	PROPN
fcis-28893	176	2	]	]	PUNCT
fcis-28893	176	3	pointed	point	VERB
fcis-28893	176	4	out	out	ADP
fcis-28893	176	5	that	that	SCONJ
fcis-28893	176	6	the	the	DET
fcis-28893	176	7	main	main	ADJ
fcis-28893	176	8	challenge	challenge	NOUN
fcis-28893	176	9	faced	face	VERB
fcis-28893	176	10	in	in	ADP
fcis-28893	176	11	the	the	DET
fcis-28893	176	12	field	field	NOUN
fcis-28893	176	13	is	be	AUX
fcis-28893	176	14	that	that	SCONJ
fcis-28893	176	15	the	the	DET
fcis-28893	176	16	annotation	annotation	NOUN
fcis-28893	176	17	process	process	NOUN
fcis-28893	176	18	is	be	AUX
fcis-28893	176	19	too	too	ADV
fcis-28893	176	20	difficult	difficult	ADJ
fcis-28893	176	21	,	,	PUNCT
fcis-28893	176	22	which	which	PRON
fcis-28893	176	23	leads	lead	VERB
fcis-28893	176	24	to	to	ADP
fcis-28893	176	25	the	the	DET
fcis-28893	176	26	small	small	ADJ
fcis-28893	176	27	size	size	NOUN
fcis-28893	176	28	of	of	ADP
fcis-28893	176	29	the	the	DET
fcis-28893	176	30	existing	exist	VERB
fcis-28893	176	31	dataset	dataset	NOUN
fcis-28893	176	32	and	and	CCONJ
fcis-28893	176	33	is	be	AUX
fcis-28893	176	34	not	not	PART
fcis-28893	176	35	conducive	conducive	ADJ
fcis-28893	176	36	to	to	ADP
fcis-28893	176	37	deep	deep	ADJ
fcis-28893	176	38	training	training	NOUN
fcis-28893	176	39	of	of	ADP
fcis-28893	176	40	models	model	NOUN
fcis-28893	176	41	.	.	PUNCT
fcis-28893	177	1	inspired	inspire	VERB
fcis-28893	177	2	by	by	ADP
fcis-28893	177	3	clip	clip	PROPN
fcis-28893	177	4	,	,	PUNCT
fcis-28893	177	5	tan	tan	PROPN
fcis-28893	177	6	et	et	PROPN
fcis-28893	177	7	al	al	PROPN
fcis-28893	177	8	.	.	PROPN
fcis-28893	177	9	investigated	investigate	VERB
fcis-28893	177	10	the	the	DET
fcis-28893	177	11	problem	problem	NOUN
fcis-28893	177	12	of	of	ADP
fcis-28893	177	13	reloadability	reloadability	NOUN
fcis-28893	177	14	through	through	ADP
fcis-28893	177	15	multimodal	multimodal	ADJ
fcis-28893	177	16	large	large	ADJ
fcis-28893	177	17	language	language	NOUN
fcis-28893	177	18	models	model	NOUN
fcis-28893	177	19	(	(	PUNCT
fcis-28893	177	20	mmlms	mmlm	NOUN
fcis-28893	177	21	)	)	PUNCT
fcis-28893	177	22	.	.	PUNCT
fcis-28893	178	1	first	first	ADV
fcis-28893	178	2	,	,	PUNCT
fcis-28893	178	3	tan	tan	PROPN
fcis-28893	178	4	et	et	PROPN
fcis-28893	178	5	al	al	PROPN
fcis-28893	178	6	.	.	PROPN
fcis-28893	178	7	prompted	prompt	VERB
fcis-28893	178	8	chatgpt	chatgpt	NOUN
fcis-28893	178	9	to	to	PART
fcis-28893	178	10	generate	generate	VERB
fcis-28893	178	11	multiple	multiple	ADJ
fcis-28893	178	12	description	description	NOUN
fcis-28893	178	13	templates	template	NOUN
fcis-28893	178	14	,	,	PUNCT
fcis-28893	178	15	which	which	PRON
fcis-28893	178	16	prompted	prompt	VERB
fcis-28893	178	17	the	the	DET
fcis-28893	178	18	mmlms	mmlm	NOUN
fcis-28893	178	19	model	model	NOUN
fcis-28893	178	20	to	to	PART
fcis-28893	178	21	generate	generate	VERB
fcis-28893	178	22	text	text	NOUN
fcis-28893	178	23	descriptions	description	NOUN
fcis-28893	178	24	with	with	ADP
fcis-28893	178	25	diversity	diversity	NOUN
fcis-28893	178	26	based	base	VERB
fcis-28893	178	27	on	on	ADP
fcis-28893	178	28	the	the	DET
fcis-28893	178	29	templates	template	NOUN
fcis-28893	178	30	,	,	PUNCT
fcis-28893	178	31	solving	solve	VERB
fcis-28893	178	32	the	the	DET
fcis-28893	178	33	overfitting	overfitte	VERB
fcis-28893	178	34	problem	problem	NOUN
fcis-28893	178	35	due	due	ADP
fcis-28893	178	36	to	to	ADP
fcis-28893	178	37	the	the	DET
fcis-28893	178	38	single	single	ADJ
fcis-28893	178	39	syntax	syntax	NOUN
fcis-28893	178	40	.	.	PUNCT
fcis-28893	179	1	despite	despite	SCONJ
fcis-28893	179	2	its	its	PRON
fcis-28893	179	3	power	power	NOUN
fcis-28893	179	4	,	,	PUNCT
fcis-28893	179	5	the	the	DET
fcis-28893	179	6	mmlms	mmlm	NOUN
fcis-28893	179	7	model	model	NOUN
fcis-28893	179	8	may	may	AUX
fcis-28893	179	9	still	still	ADV
fcis-28893	179	10	generate	generate	VERB
fcis-28893	179	11	mismatched	mismatch	VERB
fcis-28893	179	12	words	word	NOUN
fcis-28893	179	13	when	when	SCONJ
fcis-28893	179	14	performing	perform	VERB
fcis-28893	179	15	image	image	NOUN
fcis-28893	179	16	naming	naming	NOUN
fcis-28893	179	17	tasks	task	NOUN
fcis-28893	179	18	.	.	PUNCT
fcis-28893	180	1	therefore	therefore	ADV
fcis-28893	180	2	,	,	PUNCT
fcis-28893	180	3	tan	tan	PROPN
fcis-28893	180	4	et	et	PROPN
fcis-28893	180	5	al	al	PROPN
fcis-28893	180	6	.	.	PROPN
fcis-28893	180	7	designed	design	VERB
fcis-28893	180	8	the	the	DET
fcis-28893	180	9	noise	noise	NOUN
fcis-28893	180	10	-	-	PUNCT
fcis-28893	180	11	aware	aware	ADJ
fcis-28893	180	12	masking	masking	NOUN
fcis-28893	180	13	(	(	PUNCT
fcis-28893	180	14	nam	nam	NOUN
fcis-28893	180	15	)	)	PUNCT
fcis-28893	180	16	method	method	NOUN
fcis-28893	180	17	,	,	PUNCT
fcis-28893	180	18	which	which	PRON
fcis-28893	180	19	calculates	calculate	VERB
fcis-28893	180	20	the	the	DET
fcis-28893	180	21	similarity	similarity	NOUN
fcis-28893	180	22	between	between	ADP
fcis-28893	180	23	each	each	DET
fcis-28893	180	24	word	word	NOUN
fcis-28893	180	25	and	and	CCONJ
fcis-28893	180	26	the	the	DET
fcis-28893	180	27	image	image	NOUN
fcis-28893	180	28	region	region	NOUN
fcis-28893	180	29	,	,	PUNCT
fcis-28893	180	30	identifies	identify	VERB
fcis-28893	180	31	the	the	DET
fcis-28893	180	32	words	word	NOUN
fcis-28893	180	33	with	with	ADP
fcis-28893	180	34	too	too	ADV
fcis-28893	180	35	little	little	ADJ
fcis-28893	180	36	similarity	similarity	NOUN
fcis-28893	180	37	,	,	PUNCT
fcis-28893	180	38	and	and	CCONJ
fcis-28893	180	39	masks	mask	VERB
fcis-28893	180	40	them	they	PRON
fcis-28893	180	41	with	with	ADP
fcis-28893	180	42	a	a	DET
fcis-28893	180	43	higher	high	ADJ
fcis-28893	180	44	probability	probability	NOUN
fcis-28893	180	45	during	during	ADP
fcis-28893	180	46	training	training	NOUN
fcis-28893	180	47	,	,	PUNCT
fcis-28893	180	48	which	which	PRON
fcis-28893	180	49	effectively	effectively	ADV
fcis-28893	180	50	reduces	reduce	VERB
fcis-28893	180	51	the	the	DET
fcis-28893	180	52	noise	noise	NOUN
fcis-28893	180	53	generated	generate	VERB
fcis-28893	180	54	by	by	ADP
fcis-28893	180	55	the	the	DET
fcis-28893	180	56	automatic	automatic	ADJ
fcis-28893	180	57	generation	generation	NOUN
fcis-28893	180	58	of	of	ADP
fcis-28893	180	59	textual	textual	ADJ
fcis-28893	180	60	descriptions	description	NOUN
fcis-28893	180	61	by	by	ADP
fcis-28893	180	62	the	the	DET
fcis-28893	180	63	mmlms	mmlm	NOUN
fcis-28893	180	64	,	,	PUNCT
fcis-28893	180	65	while	while	SCONJ
fcis-28893	180	66	retaining	retain	VERB
fcis-28893	180	67	the	the	DET
fcis-28893	180	68	remaining	remain	VERB
fcis-28893	180	69	useful	useful	ADJ
fcis-28893	180	70	information	information	NOUN
fcis-28893	180	71	.	.	PUNCT
fcis-28893	181	1	unlike	unlike	ADP
fcis-28893	181	2	traditional	traditional	ADJ
fcis-28893	181	3	masked	mask	VERB
fcis-28893	181	4	language	language	NOUN
fcis-28893	181	5	model	model	NOUN
fcis-28893	181	6	mmlm	mmlm	NOUN
fcis-28893	181	7	methods	method	NOUN
fcis-28893	181	8	,	,	PUNCT
fcis-28893	181	9	nam	nam	NOUN
fcis-28893	181	10	performs	perform	VERB
fcis-28893	181	11	word	word	NOUN
fcis-28893	181	12	masking	masking	NOUN
fcis-28893	181	13	based	base	VERB
fcis-28893	181	14	on	on	ADP
fcis-28893	181	15	similarity	similarity	NOUN
fcis-28893	181	16	and	and	CCONJ
fcis-28893	181	17	does	do	AUX
fcis-28893	181	18	not	not	PART
fcis-28893	181	19	need	need	VERB
fcis-28893	181	20	to	to	PART
fcis-28893	181	21	re	re	VERB
fcis-28893	181	22	-	-	VERB
fcis-28893	181	23	predict	predict	VERB
fcis-28893	181	24	the	the	DET
fcis-28893	181	25	masked	masked	ADJ
fcis-28893	181	26	words	word	NOUN
fcis-28893	181	27	.	.	PUNCT
fcis-28893	182	1	based	base	VERB
fcis-28893	182	2	on	on	ADP
fcis-28893	182	3	the	the	DET
fcis-28893	182	4	above	above	ADJ
fcis-28893	182	5	approach	approach	NOUN
fcis-28893	182	6	,	,	PUNCT
fcis-28893	182	7	tan	tan	PROPN
fcis-28893	182	8	et	et	PROPN
fcis-28893	182	9	al	al	PROPN
fcis-28893	182	10	.	.	PROPN
fcis-28893	182	11	selected	select	VERB
fcis-28893	182	12	the	the	DET
fcis-28893	182	13	luperson	luperson	NOUN
fcis-28893	182	14	dataset	dataset	NOUN
fcis-28893	182	15	as	as	ADP
fcis-28893	182	16	the	the	DET
fcis-28893	182	17	image	image	NOUN
fcis-28893	182	18	source	source	NOUN
fcis-28893	182	19	to	to	PART
fcis-28893	182	20	generate	generate	VERB
fcis-28893	182	21	the	the	DET
fcis-28893	182	22	luperson	luperson	ADJ
fcis-28893	182	23	-	-	PUNCT
fcis-28893	182	24	mmlm	mmlm	NOUN
fcis-28893	182	25	dataset	dataset	NOUN
fcis-28893	182	26	containing	contain	VERB
fcis-28893	182	27	one	one	NUM
fcis-28893	182	28	million	million	NUM
fcis-28893	182	29	images	image	NOUN
fcis-28893	182	30	.	.	PUNCT
fcis-28893	183	1	notably	notably	ADV
fcis-28893	183	2	,	,	PUNCT
fcis-28893	183	3	using	use	VERB
fcis-28893	183	4	this	this	DET
fcis-28893	183	5	large	large	ADJ
fcis-28893	183	6	-	-	PUNCT
fcis-28893	183	7	scale	scale	NOUN
fcis-28893	183	8	dataset	dataset	NOUN
fcis-28893	183	9	as	as	ADV
fcis-28893	183	10	well	well	ADV
fcis-28893	183	11	as	as	ADP
fcis-28893	183	12	the	the	DET
fcis-28893	183	13	pretraining	pretraine	VERB
fcis-28893	183	14	parameters	parameter	NOUN
fcis-28893	183	15	of	of	ADP
fcis-28893	183	16	the	the	DET
fcis-28893	183	17	previous	previous	ADJ
fcis-28893	183	18	models	model	NOUN
fcis-28893	183	19	,	,	PUNCT
fcis-28893	183	20	the	the	DET
fcis-28893	183	21	authors	author	NOUN
fcis-28893	183	22	successfully	successfully	ADV
fcis-28893	183	23	reached	reach	VERB
fcis-28893	183	24	sota	sota	NOUN
fcis-28893	183	25	scores	score	NOUN
fcis-28893	183	26	on	on	ADP
fcis-28893	183	27	three	three	NUM
fcis-28893	183	28	benchmark	benchmark	NOUN
fcis-28893	183	29	125	125	NUM
fcis-28893	183	30	datasets	dataset	NOUN
fcis-28893	183	31	.	.	PUNCT
fcis-28893	184	1	tan	tan	PROPN
fcis-28893	184	2	et	et	PROPN
fcis-28893	184	3	al	al	PROPN
fcis-28893	184	4	.	.	PROPN
fcis-28893	184	5	's	's	PART
fcis-28893	184	6	idea	idea	NOUN
fcis-28893	184	7	of	of	ADP
fcis-28893	184	8	generating	generate	VERB
fcis-28893	184	9	a	a	DET
fcis-28893	184	10	large	large	ADJ
fcis-28893	184	11	-	-	PUNCT
fcis-28893	184	12	scale	scale	NOUN
fcis-28893	184	13	dataset	dataset	NOUN
fcis-28893	184	14	using	use	VERB
fcis-28893	184	15	the	the	DET
fcis-28893	184	16	mmlms	mmlm	NOUN
fcis-28893	184	17	large	large	ADJ
fcis-28893	184	18	language	language	NOUN
fcis-28893	184	19	model	model	NOUN
fcis-28893	184	20	remedies	remedy	VERB
fcis-28893	184	21	the	the	DET
fcis-28893	184	22	problem	problem	NOUN
fcis-28893	184	23	of	of	ADP
fcis-28893	184	24	missing	miss	VERB
fcis-28893	184	25	datasets	dataset	NOUN
fcis-28893	184	26	in	in	ADP
fcis-28893	184	27	the	the	DET
fcis-28893	184	28	field	field	NOUN
fcis-28893	184	29	and	and	CCONJ
fcis-28893	184	30	significantly	significantly	ADV
fcis-28893	184	31	advances	advance	VERB
fcis-28893	184	32	the	the	DET
fcis-28893	184	33	field	field	NOUN
fcis-28893	184	34	.	.	PUNCT
fcis-28893	185	1	5	5	X
fcis-28893	185	2	.	.	X
fcis-28893	185	3	conclusion	conclusion	VERB
fcis-28893	185	4	various	various	ADJ
fcis-28893	185	5	types	type	NOUN
fcis-28893	185	6	of	of	ADP
fcis-28893	185	7	text	text	NOUN
fcis-28893	185	8	-	-	PUNCT
fcis-28893	185	9	based	base	VERB
fcis-28893	185	10	pedestrian	pedestrian	NOUN
fcis-28893	185	11	retrieval	retrieval	NOUN
fcis-28893	185	12	methods	method	NOUN
fcis-28893	185	13	have	have	VERB
fcis-28893	185	14	their	their	PRON
fcis-28893	185	15	own	own	ADJ
fcis-28893	185	16	advantages	advantage	NOUN
fcis-28893	185	17	and	and	CCONJ
fcis-28893	185	18	disadvantages	disadvantage	NOUN
fcis-28893	185	19	.	.	PUNCT
fcis-28893	186	1	the	the	DET
fcis-28893	186	2	advantages	advantage	NOUN
fcis-28893	186	3	and	and	CCONJ
fcis-28893	186	4	disadvantages	disadvantage	NOUN
fcis-28893	186	5	of	of	ADP
fcis-28893	186	6	each	each	DET
fcis-28893	186	7	type	type	NOUN
fcis-28893	186	8	of	of	ADP
fcis-28893	186	9	methods	method	NOUN
fcis-28893	186	10	are	be	AUX
fcis-28893	186	11	demonstrated	demonstrate	VERB
fcis-28893	186	12	and	and	CCONJ
fcis-28893	186	13	compared	compare	VERB
fcis-28893	186	14	through	through	ADP
fcis-28893	186	15	table	table	NOUN
fcis-28893	186	16	3	3	NUM
fcis-28893	186	17	:	:	PUNCT
fcis-28893	186	18	table	table	NOUN
fcis-28893	186	19	3	3	NUM
fcis-28893	186	20	.	.	PUNCT
fcis-28893	186	21	comparison	comparison	NOUN
fcis-28893	186	22	of	of	ADP
fcis-28893	186	23	advantages	advantage	NOUN
fcis-28893	186	24	and	and	CCONJ
fcis-28893	186	25	disadvantages	disadvantage	NOUN
fcis-28893	186	26	of	of	ADP
fcis-28893	186	27	various	various	ADJ
fcis-28893	186	28	methods	method	NOUN
fcis-28893	186	29	of	of	ADP
fcis-28893	186	30	text	text	NOUN
fcis-28893	186	31	-	-	PUNCT
fcis-28893	186	32	based	base	VERB
fcis-28893	186	33	pedestrian	pedestrian	NOUN
fcis-28893	186	34	retrieval	retrieval	NOUN
fcis-28893	186	35	techniques	technique	NOUN
fcis-28893	186	36	type	type	NOUN
fcis-28893	186	37	advantage	advantage	NOUN
fcis-28893	186	38	drawbacks	drawback	NOUN
fcis-28893	186	39	global	global	ADJ
fcis-28893	186	40	feature	feature	NOUN
fcis-28893	186	41	matching	match	VERB
fcis-28893	186	42	relatively	relatively	ADV
fcis-28893	186	43	simple	simple	ADJ
fcis-28893	186	44	model	model	NOUN
fcis-28893	186	45	structure	structure	NOUN
fcis-28893	186	46	,	,	PUNCT
fcis-28893	186	47	low	low	ADJ
fcis-28893	186	48	computational	computational	ADJ
fcis-28893	186	49	complexity	complexity	NOUN
fcis-28893	186	50	,	,	PUNCT
fcis-28893	186	51	low	low	ADJ
fcis-28893	186	52	retrieval	retrieval	NOUN
fcis-28893	186	53	overhead	overhead	NOUN
fcis-28893	186	54	,	,	PUNCT
fcis-28893	186	55	suitable	suitable	ADJ
fcis-28893	186	56	for	for	ADP
fcis-28893	186	57	end	end	NOUN
fcis-28893	186	58	-	-	PUNCT
fcis-28893	186	59	to	to	ADP
fcis-28893	186	60	-	-	PUNCT
fcis-28893	186	61	end	end	NOUN
fcis-28893	186	62	applications	application	NOUN
fcis-28893	186	63	focus	focus	VERB
fcis-28893	186	64	only	only	ADV
fcis-28893	186	65	on	on	ADP
fcis-28893	186	66	the	the	DET
fcis-28893	186	67	global	global	ADJ
fcis-28893	186	68	features	feature	NOUN
fcis-28893	186	69	in	in	ADP
fcis-28893	186	70	the	the	DET
fcis-28893	186	71	graphic	graphic	ADJ
fcis-28893	186	72	samples	sample	NOUN
fcis-28893	186	73	,	,	PUNCT
fcis-28893	186	74	not	not	PART
fcis-28893	186	75	enough	enough	ADJ
fcis-28893	186	76	mining	mining	NOUN
fcis-28893	186	77	of	of	ADP
fcis-28893	186	78	finegrained	finegraine	VERB
fcis-28893	186	79	local	local	ADJ
fcis-28893	186	80	information	information	NOUN
fcis-28893	186	81	,	,	PUNCT
fcis-28893	186	82	relatively	relatively	ADV
fcis-28893	186	83	low	low	ADJ
fcis-28893	186	84	retrieval	retrieval	NOUN
fcis-28893	186	85	accuracy	accuracy	NOUN
fcis-28893	186	86	multigranularity	multigranularity	NOUN
fcis-28893	186	87	information	information	NOUN
fcis-28893	186	88	capable	capable	ADJ
fcis-28893	186	89	of	of	ADP
fcis-28893	186	90	capturing	capture	VERB
fcis-28893	186	91	multi	multi	ADJ
fcis-28893	186	92	-	-	ADJ
fcis-28893	186	93	granularity	granularity	NOUN
fcis-28893	186	94	detail	detail	NOUN
fcis-28893	186	95	information	information	NOUN
fcis-28893	186	96	,	,	PUNCT
fcis-28893	186	97	with	with	ADP
fcis-28893	186	98	strong	strong	ADJ
fcis-28893	186	99	detail	detail	NOUN
fcis-28893	186	100	learning	learn	VERB
fcis-28893	186	101	capability	capability	NOUN
fcis-28893	186	102	,	,	PUNCT
fcis-28893	186	103	it	it	PRON
fcis-28893	186	104	can	can	AUX
fcis-28893	186	105	effectively	effectively	ADV
fcis-28893	186	106	improve	improve	VERB
fcis-28893	186	107	the	the	DET
fcis-28893	186	108	retrieval	retrieval	NOUN
fcis-28893	186	109	accuracy	accuracy	NOUN
fcis-28893	186	110	.	.	PUNCT
fcis-28893	187	1	additional	additional	ADJ
fcis-28893	187	2	modules	module	NOUN
fcis-28893	187	3	are	be	AUX
fcis-28893	187	4	usually	usually	ADV
fcis-28893	187	5	required	require	VERB
fcis-28893	187	6	to	to	PART
fcis-28893	187	7	align	align	VERB
fcis-28893	187	8	information	information	NOUN
fcis-28893	187	9	at	at	ADP
fcis-28893	187	10	different	different	ADJ
fcis-28893	187	11	granularities	granularity	NOUN
fcis-28893	187	12	,	,	PUNCT
fcis-28893	187	13	and	and	CCONJ
fcis-28893	187	14	the	the	DET
fcis-28893	187	15	model	model	NOUN
fcis-28893	187	16	structure	structure	NOUN
fcis-28893	187	17	is	be	AUX
fcis-28893	187	18	more	more	ADV
fcis-28893	187	19	complex	complex	ADJ
fcis-28893	187	20	,	,	PUNCT
fcis-28893	187	21	making	make	VERB
fcis-28893	187	22	it	it	PRON
fcis-28893	187	23	difficult	difficult	ADJ
fcis-28893	187	24	to	to	PART
fcis-28893	187	25	realize	realize	VERB
fcis-28893	187	26	end	end	VERB
fcis-28893	187	27	-	-	PUNCT
fcis-28893	187	28	to	to	ADP
fcis-28893	187	29	-	-	PUNCT
fcis-28893	187	30	end	end	NOUN
fcis-28893	187	31	training	training	NOUN
fcis-28893	187	32	and	and	CCONJ
fcis-28893	187	33	realtime	realtime	NOUN
fcis-28893	187	34	retrieval	retrieval	NOUN
fcis-28893	187	35	.	.	PUNCT
fcis-28893	188	1	antagonistic	antagonistic	ADJ
fcis-28893	188	2	thought	think	VERB
fcis-28893	188	3	competitive	competitive	ADJ
fcis-28893	188	4	retrieval	retrieval	NOUN
fcis-28893	188	5	accuracy	accuracy	NOUN
fcis-28893	188	6	can	can	AUX
fcis-28893	188	7	be	be	AUX
fcis-28893	188	8	achieved	achieve	VERB
fcis-28893	188	9	while	while	SCONJ
fcis-28893	188	10	keeping	keep	VERB
fcis-28893	188	11	the	the	DET
fcis-28893	188	12	model	model	NOUN
fcis-28893	188	13	structure	structure	NOUN
fcis-28893	188	14	relatively	relatively	ADV
fcis-28893	188	15	simple	simple	ADJ
fcis-28893	188	16	consistent	consistent	ADJ
fcis-28893	188	17	with	with	ADP
fcis-28893	188	18	gan	gan	PROPN
fcis-28893	188	19	,	,	PUNCT
fcis-28893	188	20	a	a	DET
fcis-28893	188	21	balance	balance	NOUN
fcis-28893	188	22	between	between	ADP
fcis-28893	188	23	generator	generator	NOUN
fcis-28893	188	24	and	and	CCONJ
fcis-28893	188	25	discriminator	discriminator	NOUN
fcis-28893	188	26	is	be	AUX
fcis-28893	188	27	required	require	VERB
fcis-28893	188	28	,	,	PUNCT
fcis-28893	188	29	and	and	CCONJ
fcis-28893	188	30	definitional	definitional	ADJ
fcis-28893	188	31	difficulties	difficulty	NOUN
fcis-28893	188	32	lead	lead	VERB
fcis-28893	188	33	to	to	ADP
fcis-28893	188	34	difficulty	difficulty	NOUN
fcis-28893	188	35	in	in	ADP
fcis-28893	188	36	converging	converge	VERB
fcis-28893	188	37	the	the	DET
fcis-28893	188	38	model	model	NOUN
fcis-28893	188	39	cross	cross	ADJ
fcis-28893	188	40	-	-	ADJ
fcis-28893	188	41	modal	modal	ADJ
fcis-28893	188	42	attention	attention	NOUN
fcis-28893	188	43	mechanisms	mechanism	NOUN
fcis-28893	188	44	establish	establish	VERB
fcis-28893	188	45	the	the	DET
fcis-28893	188	46	connection	connection	NOUN
fcis-28893	188	47	between	between	ADP
fcis-28893	188	48	each	each	DET
fcis-28893	188	49	word	word	NOUN
fcis-28893	188	50	and	and	CCONJ
fcis-28893	188	51	phrase	phrase	NOUN
fcis-28893	188	52	and	and	CCONJ
fcis-28893	188	53	pixel	pixel	NOUN
fcis-28893	188	54	and	and	CCONJ
fcis-28893	188	55	image	image	NOUN
fcis-28893	188	56	block	block	NOUN
fcis-28893	188	57	to	to	PART
fcis-28893	188	58	realize	realize	VERB
fcis-28893	188	59	deep	deep	ADJ
fcis-28893	188	60	cross	cross	ADJ
fcis-28893	188	61	-	-	ADJ
fcis-28893	188	62	modal	modal	ADJ
fcis-28893	188	63	alignment	alignment	NOUN
fcis-28893	188	64	and	and	CCONJ
fcis-28893	188	65	significantly	significantly	ADV
fcis-28893	188	66	improve	improve	VERB
fcis-28893	188	67	the	the	DET
fcis-28893	188	68	model	model	NOUN
fcis-28893	188	69	's	's	PART
fcis-28893	188	70	cross	cross	ADJ
fcis-28893	188	71	-	-	ADJ
fcis-28893	188	72	modal	modal	ADJ
fcis-28893	188	73	semantic	semantic	ADJ
fcis-28893	188	74	understanding	understanding	NOUN
fcis-28893	188	75	and	and	CCONJ
fcis-28893	188	76	retrieval	retrieval	NOUN
fcis-28893	188	77	accuracy	accuracy	NOUN
fcis-28893	188	78	the	the	DET
fcis-28893	188	79	increase	increase	NOUN
fcis-28893	188	80	in	in	ADP
fcis-28893	188	81	computational	computational	ADJ
fcis-28893	188	82	complexity	complexity	NOUN
fcis-28893	188	83	and	and	CCONJ
fcis-28893	188	84	retrieval	retrieval	NOUN
fcis-28893	188	85	time	time	NOUN
fcis-28893	188	86	makes	make	VERB
fcis-28893	188	87	it	it	PRON
fcis-28893	188	88	difficult	difficult	ADJ
fcis-28893	188	89	for	for	ADP
fcis-28893	188	90	practical	practical	ADJ
fcis-28893	188	91	applications	application	NOUN
fcis-28893	188	92	to	to	PART
fcis-28893	188	93	bear	bear	VERB
fcis-28893	188	94	such	such	DET
fcis-28893	188	95	a	a	DET
fcis-28893	188	96	heavy	heavy	ADJ
fcis-28893	188	97	computational	computational	ADJ
fcis-28893	188	98	burden	burden	NOUN
fcis-28893	188	99	vlp	vlp	PROPN
fcis-28893	188	100	with	with	ADP
fcis-28893	188	101	the	the	DET
fcis-28893	188	102	powerful	powerful	ADJ
fcis-28893	188	103	cross	cross	ADJ
fcis-28893	188	104	-	-	ADJ
fcis-28893	188	105	modal	modal	ADJ
fcis-28893	188	106	learning	learning	NOUN
fcis-28893	188	107	and	and	CCONJ
fcis-28893	188	108	semantic	semantic	ADJ
fcis-28893	188	109	understanding	understanding	NOUN
fcis-28893	188	110	capability	capability	NOUN
fcis-28893	188	111	of	of	ADP
fcis-28893	188	112	vlp	vlp	PROPN
fcis-28893	188	113	model	model	NOUN
fcis-28893	188	114	,	,	PUNCT
fcis-28893	188	115	it	it	PRON
fcis-28893	188	116	realizes	realize	VERB
fcis-28893	188	117	the	the	DET
fcis-28893	188	118	significant	significant	ADJ
fcis-28893	188	119	improvement	improvement	NOUN
fcis-28893	188	120	of	of	ADP
fcis-28893	188	121	retrieval	retrieval	NOUN
fcis-28893	188	122	accuracy	accuracy	NOUN
fcis-28893	188	123	vlp	vlp	PROPN
fcis-28893	188	124	models	model	NOUN
fcis-28893	188	125	require	require	VERB
fcis-28893	188	126	large	large	ADJ
fcis-28893	188	127	amounts	amount	NOUN
fcis-28893	188	128	of	of	ADP
fcis-28893	188	129	data	datum	NOUN
fcis-28893	188	130	and	and	CCONJ
fcis-28893	188	131	resources	resource	NOUN
fcis-28893	188	132	for	for	ADP
fcis-28893	188	133	training	training	NOUN
fcis-28893	188	134	and	and	CCONJ
fcis-28893	188	135	fine	fine	ADV
fcis-28893	188	136	-	-	PUNCT
fcis-28893	188	137	tuning	tuning	NOUN
fcis-28893	188	138	,	,	PUNCT
fcis-28893	188	139	and	and	CCONJ
fcis-28893	188	140	are	be	AUX
fcis-28893	188	141	difficult	difficult	ADJ
fcis-28893	188	142	and	and	CCONJ
fcis-28893	188	143	expensive	expensive	ADJ
fcis-28893	188	144	to	to	PART
fcis-28893	188	145	use	use	VERB
fcis-28893	188	146	and	and	CCONJ
fcis-28893	188	147	reproduce	reproduce	VERB
fcis-28893	188	148	the	the	DET
fcis-28893	188	149	performance	performance	NOUN
fcis-28893	188	150	of	of	ADP
fcis-28893	188	151	the	the	DET
fcis-28893	188	152	various	various	ADJ
fcis-28893	188	153	methods	method	NOUN
fcis-28893	188	154	on	on	ADP
fcis-28893	188	155	the	the	DET
fcis-28893	188	156	cuhkpedes	cuhkpede	NOUN
fcis-28893	188	157	,	,	PUNCT
fcis-28893	188	158	igfg	igfg	NOUN
fcis-28893	188	159	-	-	PUNCT
fcis-28893	188	160	pedes	pes	NOUN
fcis-28893	188	161	and	and	CCONJ
fcis-28893	188	162	rstpreid	rstpreid	NOUN
fcis-28893	188	163	datasets	dataset	NOUN
fcis-28893	188	164	are	be	AUX
fcis-28893	188	165	compared	compare	VERB
fcis-28893	188	166	in	in	ADP
fcis-28893	188	167	table	table	NOUN
fcis-28893	188	168	4	4	NUM
fcis-28893	188	169	.	.	X
fcis-28893	188	170	text	text	NOUN
fcis-28893	188	171	-	-	PUNCT
fcis-28893	188	172	based	base	VERB
fcis-28893	188	173	pedestrian	pedestrian	NOUN
fcis-28893	188	174	retrieval	retrieval	NOUN
fcis-28893	188	175	techniques	technique	NOUN
fcis-28893	188	176	are	be	AUX
fcis-28893	188	177	an	an	DET
fcis-28893	188	178	emerging	emerge	VERB
fcis-28893	188	179	field	field	NOUN
fcis-28893	188	180	that	that	PRON
fcis-28893	188	181	still	still	ADV
fcis-28893	188	182	faces	face	VERB
fcis-28893	188	183	many	many	ADJ
fcis-28893	188	184	challenges	challenge	NOUN
fcis-28893	188	185	:	:	PUNCT
fcis-28893	188	186	1	1	X
fcis-28893	188	187	)	)	PUNCT
fcis-28893	188	188	cross	cross	ADJ
fcis-28893	188	189	-	-	ADJ
fcis-28893	188	190	modal	modal	ADJ
fcis-28893	188	191	feature	feature	NOUN
fcis-28893	188	192	alignment	alignment	NOUN
fcis-28893	188	193	:	:	PUNCT
fcis-28893	188	194	cross	cross	ADJ
fcis-28893	188	195	-	-	ADJ
fcis-28893	188	196	modal	modal	ADJ
fcis-28893	188	197	feature	feature	NOUN
fcis-28893	188	198	alignment	alignment	NOUN
fcis-28893	188	199	is	be	AUX
fcis-28893	188	200	still	still	ADV
fcis-28893	188	201	a	a	DET
fcis-28893	188	202	core	core	NOUN
fcis-28893	188	203	problem	problem	NOUN
fcis-28893	188	204	in	in	ADP
fcis-28893	188	205	the	the	DET
fcis-28893	188	206	field	field	NOUN
fcis-28893	188	207	.	.	PUNCT
fcis-28893	189	1	it	it	PRON
fcis-28893	189	2	is	be	AUX
fcis-28893	189	3	difficult	difficult	ADJ
fcis-28893	189	4	for	for	SCONJ
fcis-28893	189	5	existing	exist	VERB
fcis-28893	189	6	methods	method	NOUN
fcis-28893	189	7	to	to	PART
fcis-28893	189	8	improve	improve	VERB
fcis-28893	189	9	the	the	DET
fcis-28893	189	10	retrieval	retrieval	NOUN
fcis-28893	189	11	accuracy	accuracy	NOUN
fcis-28893	189	12	while	while	SCONJ
fcis-28893	189	13	considering	consider	VERB
fcis-28893	189	14	the	the	DET
fcis-28893	189	15	computational	computational	ADJ
fcis-28893	189	16	efficiency	efficiency	NOUN
fcis-28893	189	17	and	and	CCONJ
fcis-28893	189	18	speed	speed	NOUN
fcis-28893	189	19	,	,	PUNCT
fcis-28893	189	20	and	and	CCONJ
fcis-28893	189	21	the	the	DET
fcis-28893	189	22	balance	balance	NOUN
fcis-28893	189	23	between	between	ADP
fcis-28893	189	24	accuracy	accuracy	NOUN
fcis-28893	189	25	and	and	CCONJ
fcis-28893	189	26	speed	speed	NOUN
fcis-28893	189	27	has	have	AUX
fcis-28893	189	28	not	not	PART
fcis-28893	189	29	been	be	AUX
fcis-28893	189	30	effectively	effectively	ADV
fcis-28893	189	31	solved	solve	VERB
fcis-28893	189	32	in	in	ADP
fcis-28893	189	33	large	large	ADJ
fcis-28893	189	34	-	-	PUNCT
fcis-28893	189	35	scale	scale	NOUN
fcis-28893	189	36	data	datum	NOUN
fcis-28893	189	37	retrieval	retrieval	NOUN
fcis-28893	189	38	tasks	task	NOUN
fcis-28893	189	39	.	.	PUNCT
fcis-28893	190	1	2	2	X
fcis-28893	190	2	)	)	PUNCT
fcis-28893	190	3	pedestrian	pedestrian	NOUN
fcis-28893	190	4	occlusion	occlusion	NOUN
fcis-28893	190	5	problem	problem	NOUN
fcis-28893	190	6	:	:	PUNCT
fcis-28893	190	7	pedestrian	pedestrian	NOUN
fcis-28893	190	8	occlusion	occlusion	NOUN
fcis-28893	190	9	is	be	AUX
fcis-28893	190	10	common	common	ADJ
fcis-28893	190	11	in	in	ADP
fcis-28893	190	12	practical	practical	ADJ
fcis-28893	190	13	applications	application	NOUN
fcis-28893	190	14	,	,	PUNCT
fcis-28893	190	15	and	and	CCONJ
fcis-28893	190	16	the	the	DET
fcis-28893	190	17	occlusion	occlusion	NOUN
fcis-28893	190	18	problem	problem	NOUN
fcis-28893	190	19	has	have	AUX
fcis-28893	190	20	not	not	PART
fcis-28893	190	21	been	be	AUX
fcis-28893	190	22	fully	fully	ADV
fcis-28893	190	23	considered	consider	VERB
fcis-28893	190	24	in	in	ADP
fcis-28893	190	25	the	the	DET
fcis-28893	190	26	current	current	ADJ
fcis-28893	190	27	domain	domain	NOUN
fcis-28893	190	28	.	.	PUNCT
fcis-28893	191	1	the	the	DET
fcis-28893	191	2	lack	lack	NOUN
fcis-28893	191	3	of	of	ADP
fcis-28893	191	4	occlusion	occlusion	NOUN
fcis-28893	191	5	-	-	PUNCT
fcis-28893	191	6	related	relate	VERB
fcis-28893	191	7	samples	sample	NOUN
fcis-28893	191	8	and	and	CCONJ
fcis-28893	191	9	datasets	dataset	NOUN
fcis-28893	191	10	leads	lead	VERB
fcis-28893	191	11	to	to	ADP
fcis-28893	191	12	the	the	DET
fcis-28893	191	13	lack	lack	NOUN
fcis-28893	191	14	of	of	ADP
fcis-28893	191	15	reliability	reliability	NOUN
fcis-28893	191	16	of	of	ADP
fcis-28893	191	17	the	the	DET
fcis-28893	191	18	main	main	ADJ
fcis-28893	191	19	methods	method	NOUN
fcis-28893	191	20	in	in	ADP
fcis-28893	191	21	occlusion	occlusion	NOUN
fcis-28893	191	22	scenarios	scenario	NOUN
fcis-28893	191	23	.	.	PUNCT
fcis-28893	192	1	3	3	X
fcis-28893	192	2	)	)	PUNCT
fcis-28893	192	3	the	the	DET
fcis-28893	192	4	problem	problem	NOUN
fcis-28893	192	5	of	of	ADP
fcis-28893	192	6	the	the	DET
fcis-28893	192	7	number	number	NOUN
fcis-28893	192	8	of	of	ADP
fcis-28893	192	9	model	model	NOUN
fcis-28893	192	10	parameters	parameter	NOUN
fcis-28893	192	11	:	:	PUNCT
fcis-28893	192	12	model	model	PROPN
fcis-28893	192	13	lightweighting	lightweighting	NOUN
fcis-28893	192	14	is	be	AUX
fcis-28893	192	15	another	another	DET
fcis-28893	192	16	challenge	challenge	NOUN
fcis-28893	192	17	in	in	ADP
fcis-28893	192	18	the	the	DET
fcis-28893	192	19	field	field	NOUN
fcis-28893	192	20	.	.	PUNCT
fcis-28893	193	1	existing	exist	VERB
fcis-28893	193	2	methods	method	NOUN
fcis-28893	193	3	based	base	VERB
fcis-28893	193	4	on	on	ADP
fcis-28893	193	5	cross	cross	ADJ
fcis-28893	193	6	-	-	ADJ
fcis-28893	193	7	modal	modal	ADJ
fcis-28893	193	8	attention	attention	NOUN
fcis-28893	193	9	mechanism	mechanism	NOUN
fcis-28893	193	10	and	and	CCONJ
fcis-28893	193	11	visual	visual	ADJ
fcis-28893	193	12	language	language	NOUN
fcis-28893	193	13	pre	pre	ADJ
fcis-28893	193	14	-	-	ADJ
fcis-28893	193	15	training	training	ADJ
fcis-28893	193	16	models	model	NOUN
fcis-28893	193	17	have	have	VERB
fcis-28893	193	18	significant	significant	ADJ
fcis-28893	193	19	improvement	improvement	NOUN
fcis-28893	193	20	in	in	ADP
fcis-28893	193	21	accuracy	accuracy	NOUN
fcis-28893	193	22	,	,	PUNCT
fcis-28893	193	23	but	but	CCONJ
fcis-28893	193	24	their	their	PRON
fcis-28893	193	25	huge	huge	ADJ
fcis-28893	193	26	number	number	NOUN
fcis-28893	193	27	of	of	ADP
fcis-28893	193	28	model	model	NOUN
fcis-28893	193	29	parameters	parameter	NOUN
fcis-28893	193	30	and	and	CCONJ
fcis-28893	193	31	extremely	extremely	ADV
fcis-28893	193	32	high	high	ADJ
fcis-28893	193	33	requirements	requirement	NOUN
fcis-28893	193	34	on	on	ADP
fcis-28893	193	35	hardware	hardware	NOUN
fcis-28893	193	36	equipment	equipment	NOUN
fcis-28893	193	37	make	make	VERB
fcis-28893	193	38	it	it	PRON
fcis-28893	193	39	difficult	difficult	ADJ
fcis-28893	193	40	to	to	PART
fcis-28893	193	41	retrieve	retrieve	VERB
fcis-28893	193	42	them	they	PRON
fcis-28893	193	43	in	in	ADP
fcis-28893	193	44	real	real	ADJ
fcis-28893	193	45	time	time	NOUN
fcis-28893	193	46	in	in	ADP
fcis-28893	193	47	practical	practical	ADJ
fcis-28893	193	48	applications	application	NOUN
fcis-28893	193	49	,	,	PUNCT
fcis-28893	193	50	which	which	PRON
fcis-28893	193	51	limits	limit	VERB
fcis-28893	193	52	their	their	PRON
fcis-28893	193	53	practical	practical	ADJ
fcis-28893	193	54	deployment	deployment	NOUN
fcis-28893	193	55	.	.	PUNCT
fcis-28893	194	1	4	4	NUM
fcis-28893	194	2	)	)	PUNCT
fcis-28893	194	3	dataset	dataset	NOUN
fcis-28893	194	4	problem	problem	NOUN
fcis-28893	194	5	:	:	PUNCT
fcis-28893	194	6	in	in	ADP
fcis-28893	194	7	the	the	DET
fcis-28893	194	8	privacy	privacy	NOUN
fcis-28893	194	9	protection	protection	NOUN
fcis-28893	194	10	and	and	CCONJ
fcis-28893	194	11	other	other	ADJ
fcis-28893	194	12	factors	factor	NOUN
fcis-28893	194	13	,	,	PUNCT
fcis-28893	194	14	it	it	PRON
fcis-28893	194	15	is	be	AUX
fcis-28893	194	16	very	very	ADV
fcis-28893	194	17	difficult	difficult	ADJ
fcis-28893	194	18	to	to	PART
fcis-28893	194	19	construct	construct	VERB
fcis-28893	194	20	high	high	ADJ
fcis-28893	194	21	-	-	PUNCT
fcis-28893	194	22	quality	quality	NOUN
fcis-28893	194	23	and	and	CCONJ
fcis-28893	194	24	largescale	largescale	NOUN
fcis-28893	194	25	datasets	dataset	NOUN
fcis-28893	194	26	.	.	PUNCT
fcis-28893	195	1	currently	currently	ADV
fcis-28893	195	2	,	,	PUNCT
fcis-28893	195	3	it	it	PRON
fcis-28893	195	4	mainly	mainly	ADV
fcis-28893	195	5	relies	rely	VERB
fcis-28893	195	6	on	on	ADP
fcis-28893	195	7	three	three	NUM
fcis-28893	195	8	datasets	dataset	NOUN
fcis-28893	195	9	for	for	ADP
fcis-28893	195	10	model	model	NOUN
fcis-28893	195	11	training	training	NOUN
fcis-28893	195	12	and	and	CCONJ
fcis-28893	195	13	performance	performance	NOUN
fcis-28893	195	14	evaluation	evaluation	NOUN
fcis-28893	195	15	,	,	PUNCT
fcis-28893	195	16	which	which	PRON
fcis-28893	195	17	limits	limit	VERB
fcis-28893	195	18	the	the	DET
fcis-28893	195	19	diversity	diversity	NOUN
fcis-28893	195	20	and	and	CCONJ
fcis-28893	195	21	innovation	innovation	NOUN
fcis-28893	195	22	of	of	ADP
fcis-28893	195	23	algorithms	algorithm	NOUN
fcis-28893	195	24	in	in	ADP
fcis-28893	195	25	the	the	DET
fcis-28893	195	26	field	field	NOUN
fcis-28893	195	27	to	to	ADP
fcis-28893	195	28	some	some	DET
fcis-28893	195	29	extent	extent	NOUN
fcis-28893	195	30	.	.	PUNCT
fcis-28893	196	1	in	in	ADP
fcis-28893	196	2	summary	summary	NOUN
fcis-28893	196	3	,	,	PUNCT
fcis-28893	196	4	although	although	SCONJ
fcis-28893	196	5	text	text	NOUN
fcis-28893	196	6	-	-	PUNCT
fcis-28893	196	7	based	base	VERB
fcis-28893	196	8	pedestrian	pedestrian	NOUN
fcis-28893	196	9	retrieval	retrieval	NOUN
fcis-28893	196	10	technology	technology	NOUN
fcis-28893	196	11	has	have	AUX
fcis-28893	196	12	made	make	VERB
fcis-28893	196	13	some	some	DET
fcis-28893	196	14	progress	progress	NOUN
fcis-28893	196	15	in	in	ADP
fcis-28893	196	16	the	the	DET
fcis-28893	196	17	past	past	ADJ
fcis-28893	196	18	few	few	ADJ
fcis-28893	196	19	years	year	NOUN
fcis-28893	196	20	,	,	PUNCT
fcis-28893	196	21	it	it	PRON
fcis-28893	196	22	still	still	ADV
fcis-28893	196	23	faces	face	VERB
fcis-28893	196	24	the	the	DET
fcis-28893	196	25	problems	problem	NOUN
fcis-28893	196	26	of	of	ADP
fcis-28893	196	27	cross	cross	ADJ
fcis-28893	196	28	-	-	ADJ
fcis-28893	196	29	modal	modal	ADJ
fcis-28893	196	30	feature	feature	NOUN
fcis-28893	196	31	pairs	pair	NOUN
fcis-28893	196	32	alignment	alignment	NOUN
fcis-28893	196	33	,	,	PUNCT
fcis-28893	196	34	pedestrian	pedestrian	NOUN
fcis-28893	196	35	occlusion	occlusion	NOUN
fcis-28893	196	36	,	,	PUNCT
fcis-28893	196	37	number	number	NOUN
fcis-28893	196	38	of	of	ADP
fcis-28893	196	39	model	model	NOUN
fcis-28893	196	40	parameters	parameter	NOUN
fcis-28893	196	41	and	and	CCONJ
fcis-28893	196	42	datasets	dataset	NOUN
fcis-28893	196	43	.	.	PUNCT
fcis-28893	197	1	combining	combine	VERB
fcis-28893	197	2	the	the	DET
fcis-28893	197	3	existing	exist	VERB
fcis-28893	197	4	research	research	NOUN
fcis-28893	197	5	status	status	NOUN
fcis-28893	197	6	and	and	CCONJ
fcis-28893	197	7	domestic	domestic	ADJ
fcis-28893	197	8	and	and	CCONJ
fcis-28893	197	9	international	international	ADJ
fcis-28893	197	10	research	research	NOUN
fcis-28893	197	11	hotspots	hotspot	NOUN
fcis-28893	197	12	,	,	PUNCT
fcis-28893	197	13	this	this	DET
fcis-28893	197	14	paper	paper	NOUN
fcis-28893	197	15	looks	look	VERB
fcis-28893	197	16	forward	forward	ADV
fcis-28893	197	17	to	to	ADP
fcis-28893	197	18	the	the	DET
fcis-28893	197	19	future	future	ADJ
fcis-28893	197	20	development	development	NOUN
fcis-28893	197	21	direction	direction	NOUN
fcis-28893	197	22	of	of	ADP
fcis-28893	197	23	text	text	NOUN
fcis-28893	197	24	-	-	PUNCT
fcis-28893	197	25	based	base	VERB
fcis-28893	197	26	pedestrian	pedestrian	NOUN
fcis-28893	197	27	retrieval	retrieval	NOUN
fcis-28893	197	28	technology	technology	NOUN
fcis-28893	197	29	and	and	CCONJ
fcis-28893	197	30	explores	explore	VERB
fcis-28893	197	31	possible	possible	ADJ
fcis-28893	197	32	breakthrough	breakthrough	NOUN
fcis-28893	197	33	paths	path	NOUN
fcis-28893	197	34	.	.	PUNCT
fcis-28893	198	1	1	1	X
fcis-28893	198	2	)	)	PUNCT
fcis-28893	198	3	cross	cross	ADJ
fcis-28893	198	4	-	-	ADJ
fcis-28893	198	5	modal	modal	ADJ
fcis-28893	198	6	alignment	alignment	NOUN
fcis-28893	198	7	with	with	ADP
fcis-28893	198	8	the	the	DET
fcis-28893	198	9	help	help	NOUN
fcis-28893	198	10	of	of	ADP
fcis-28893	198	11	additional	additional	ADJ
fcis-28893	198	12	information	information	NOUN
fcis-28893	198	13	and	and	CCONJ
fcis-28893	198	14	modules	module	NOUN
fcis-28893	198	15	.	.	PUNCT
fcis-28893	199	1	to	to	PART
fcis-28893	199	2	reduce	reduce	VERB
fcis-28893	199	3	the	the	DET
fcis-28893	199	4	inherent	inherent	ADJ
fcis-28893	199	5	differences	difference	NOUN
fcis-28893	199	6	between	between	ADP
fcis-28893	199	7	image	image	NOUN
fcis-28893	199	8	and	and	CCONJ
fcis-28893	199	9	text	text	NOUN
fcis-28893	199	10	modalities	modality	NOUN
fcis-28893	199	11	,	,	PUNCT
fcis-28893	199	12	the	the	DET
fcis-28893	199	13	cross	cross	ADJ
fcis-28893	199	14	-	-	ADJ
fcis-28893	199	15	modal	modal	ADJ
fcis-28893	199	16	learning	learning	NOUN
fcis-28893	199	17	ability	ability	NOUN
fcis-28893	199	18	of	of	ADP
fcis-28893	199	19	the	the	DET
fcis-28893	199	20	model	model	NOUN
fcis-28893	199	21	can	can	AUX
fcis-28893	199	22	be	be	AUX
fcis-28893	199	23	enhanced	enhance	VERB
fcis-28893	199	24	with	with	ADP
fcis-28893	199	25	the	the	DET
fcis-28893	199	26	help	help	NOUN
fcis-28893	199	27	of	of	ADP
fcis-28893	199	28	additional	additional	ADJ
fcis-28893	199	29	information	information	NOUN
fcis-28893	199	30	to	to	PART
fcis-28893	199	31	better	well	ADV
fcis-28893	199	32	achieve	achieve	VERB
fcis-28893	199	33	feature	feature	NOUN
fcis-28893	199	34	alignment	alignment	NOUN
fcis-28893	199	35	can	can	AUX
fcis-28893	199	36	be	be	AUX
fcis-28893	199	37	explored	explore	VERB
fcis-28893	199	38	.	.	PUNCT
fcis-28893	200	1	for	for	ADP
fcis-28893	200	2	example	example	NOUN
fcis-28893	200	3	,	,	PUNCT
fcis-28893	200	4	the	the	DET
fcis-28893	200	5	lapscore	lapscore	PROPN
fcis-28893	200	6	method	method	NOUN
fcis-28893	200	7	proposed	propose	VERB
fcis-28893	200	8	by	by	ADP
fcis-28893	200	9	wu	wu	PROPN
fcis-28893	200	10	et	et	PROPN
fcis-28893	200	11	al[19	al[19	PROPN
fcis-28893	200	12	]	]	PUNCT
fcis-28893	200	13	to	to	PART
fcis-28893	200	14	design	design	VERB
fcis-28893	200	15	a	a	DET
fcis-28893	200	16	color	color	NOUN
fcis-28893	200	17	inference	inference	NOUN
fcis-28893	200	18	module	module	NOUN
fcis-28893	200	19	can	can	AUX
fcis-28893	200	20	be	be	AUX
fcis-28893	200	21	directly	directly	ADV
fcis-28893	200	22	attached	attach	VERB
fcis-28893	200	23	to	to	ADP
fcis-28893	200	24	the	the	DET
fcis-28893	200	25	mainstream	mainstream	NOUN
fcis-28893	200	26	baseline	baseline	PROPN
fcis-28893	200	27	model	model	NOUN
fcis-28893	200	28	,	,	PUNCT
fcis-28893	200	29	enabling	enable	VERB
fcis-28893	200	30	the	the	DET
fcis-28893	200	31	model	model	NOUN
fcis-28893	200	32	to	to	PART
fcis-28893	200	33	make	make	VERB
fcis-28893	200	34	full	full	ADJ
fcis-28893	200	35	use	use	NOUN
fcis-28893	200	36	of	of	ADP
fcis-28893	200	37	the	the	DET
fcis-28893	200	38	color	color	NOUN
fcis-28893	200	39	features	feature	NOUN
fcis-28893	200	40	of	of	ADP
fcis-28893	200	41	the	the	DET
fcis-28893	200	42	pedestrian	pedestrian	NOUN
fcis-28893	200	43	's	's	PART
fcis-28893	200	44	clothing	clothing	NOUN
fcis-28893	200	45	to	to	PART
fcis-28893	200	46	improve	improve	VERB
fcis-28893	200	47	the	the	DET
fcis-28893	200	48	accuracy	accuracy	NOUN
fcis-28893	200	49	and	and	CCONJ
fcis-28893	200	50	efficiency	efficiency	NOUN
fcis-28893	200	51	of	of	ADP
fcis-28893	200	52	retrieval	retrieval	NOUN
fcis-28893	200	53	.	.	PUNCT
fcis-28893	201	1	in	in	ADP
fcis-28893	201	2	addition	addition	NOUN
fcis-28893	201	3	,	,	PUNCT
fcis-28893	201	4	behavioral	behavioral	ADJ
fcis-28893	201	5	and	and	CCONJ
fcis-28893	201	6	pose	pose	VERB
fcis-28893	201	7	information	information	NOUN
fcis-28893	201	8	can	can	AUX
fcis-28893	201	9	also	also	ADV
fcis-28893	201	10	provide	provide	VERB
fcis-28893	201	11	strong	strong	ADJ
fcis-28893	201	12	support	support	NOUN
fcis-28893	201	13	to	to	ADP
fcis-28893	201	14	the	the	DET
fcis-28893	201	15	model	model	NOUN
fcis-28893	201	16	,	,	PUNCT
fcis-28893	201	17	especially	especially	ADV
fcis-28893	201	18	in	in	ADP
fcis-28893	201	19	scenarios	scenario	NOUN
fcis-28893	201	20	where	where	SCONJ
fcis-28893	201	21	the	the	DET
fcis-28893	201	22	target	target	NOUN
fcis-28893	201	23	pedestrians	pedestrian	NOUN
fcis-28893	201	24	are	be	AUX
fcis-28893	201	25	occluded	occlude	VERB
fcis-28893	201	26	by	by	ADP
fcis-28893	201	27	obstacles	obstacle	NOUN
fcis-28893	201	28	,	,	PUNCT
fcis-28893	201	29	and	and	CCONJ
fcis-28893	201	30	this	this	DET
fcis-28893	201	31	information	information	NOUN
fcis-28893	201	32	can	can	AUX
fcis-28893	201	33	significantly	significantly	ADV
fcis-28893	201	34	improve	improve	VERB
fcis-28893	201	35	the	the	DET
fcis-28893	201	36	robustness	robustness	NOUN
fcis-28893	201	37	of	of	ADP
fcis-28893	201	38	the	the	DET
fcis-28893	201	39	model	model	NOUN
fcis-28893	201	40	.	.	PUNCT
fcis-28893	202	1	combining	combine	VERB
fcis-28893	202	2	these	these	DET
fcis-28893	202	3	additional	additional	ADJ
fcis-28893	202	4	information	information	NOUN
fcis-28893	202	5	,	,	PUNCT
fcis-28893	202	6	text	text	NOUN
fcis-28893	202	7	-	-	PUNCT
fcis-28893	202	8	based	base	VERB
fcis-28893	202	9	pedestrian	pedestrian	NOUN
fcis-28893	202	10	retrieval	retrieval	NOUN
fcis-28893	202	11	is	be	AUX
fcis-28893	202	12	expected	expect	VERB
fcis-28893	202	13	to	to	PART
fcis-28893	202	14	be	be	AUX
fcis-28893	202	15	a	a	DET
fcis-28893	202	16	very	very	ADV
fcis-28893	202	17	promising	promising	ADJ
fcis-28893	202	18	development	development	NOUN
fcis-28893	202	19	direction	direction	NOUN
fcis-28893	202	20	.	.	PUNCT
fcis-28893	203	1	2	2	X
fcis-28893	203	2	)	)	PUNCT
fcis-28893	203	3	constructing	construct	VERB
fcis-28893	203	4	large	large	ADJ
fcis-28893	203	5	-	-	PUNCT
fcis-28893	203	6	scale	scale	NOUN
fcis-28893	203	7	datasets	dataset	NOUN
fcis-28893	203	8	.	.	PUNCT
fcis-28893	204	1	compared	compare	VERB
fcis-28893	204	2	with	with	ADP
fcis-28893	204	3	other	other	ADJ
fcis-28893	204	4	popular	popular	ADJ
fcis-28893	204	5	fields	field	NOUN
fcis-28893	204	6	,	,	PUNCT
fcis-28893	204	7	the	the	DET
fcis-28893	204	8	number	number	NOUN
fcis-28893	204	9	of	of	ADP
fcis-28893	204	10	text	text	NOUN
fcis-28893	204	11	-	-	PUNCT
fcis-28893	204	12	based	base	VERB
fcis-28893	204	13	pedestrian	pedestrian	NOUN
fcis-28893	204	14	retrieval	retrieval	NOUN
fcis-28893	204	15	datasets	dataset	NOUN
fcis-28893	204	16	is	be	AUX
fcis-28893	204	17	small	small	ADJ
fcis-28893	204	18	,	,	PUNCT
fcis-28893	204	19	which	which	PRON
fcis-28893	204	20	limits	limit	VERB
fcis-28893	204	21	the	the	DET
fcis-28893	204	22	further	further	ADJ
fcis-28893	204	23	development	development	NOUN
fcis-28893	204	24	of	of	ADP
fcis-28893	204	25	the	the	DET
fcis-28893	204	26	field	field	NOUN
fcis-28893	204	27	.	.	PUNCT
fcis-28893	205	1	tan	tan	PROPN
fcis-28893	205	2	et	et	NOUN
fcis-28893	206	1	al.[18	al.[18	PROPN
fcis-28893	206	2	]	]	PUNCT
fcis-28893	206	3	utilize	utilize	VERB
fcis-28893	206	4	the	the	DET
fcis-28893	206	5	mmlms	mmlm	NOUN
fcis-28893	206	6	large	large	ADJ
fcis-28893	206	7	language	language	NOUN
fcis-28893	206	8	model	model	NOUN
fcis-28893	206	9	to	to	PART
fcis-28893	206	10	automatically	automatically	ADV
fcis-28893	206	11	annotate	annotate	VERB
fcis-28893	206	12	the	the	DET
fcis-28893	206	13	images	image	NOUN
fcis-28893	206	14	to	to	PART
fcis-28893	206	15	form	form	VERB
fcis-28893	206	16	a	a	DET
fcis-28893	206	17	large	large	ADJ
fcis-28893	206	18	-	-	PUNCT
fcis-28893	206	19	scale	scale	NOUN
fcis-28893	206	20	dataset	dataset	NOUN
fcis-28893	206	21	;	;	PUNCT
fcis-28893	206	22	wu	wu	PROPN
fcis-28893	206	23	et	et	NOUN
fcis-28893	206	24	al.[20	al.[20	NOUN
fcis-28893	206	25	]	]	PUNCT
fcis-28893	206	26	construct	construct	VERB
fcis-28893	206	27	datasets	dataset	NOUN
fcis-28893	206	28	more	more	ADV
fcis-28893	206	29	in	in	ADP
fcis-28893	206	30	line	line	NOUN
fcis-28893	206	31	with	with	ADP
fcis-28893	206	32	the	the	DET
fcis-28893	206	33	existence	existence	NOUN
fcis-28893	206	34	of	of	ADP
fcis-28893	206	35	occlusion	occlusion	NOUN
fcis-28893	206	36	problems	problem	NOUN
fcis-28893	206	37	in	in	ADP
fcis-28893	206	38	real	real	ADJ
fcis-28893	206	39	scenes	scene	NOUN
fcis-28893	206	40	by	by	ADP
fcis-28893	206	41	designing	design	VERB
fcis-28893	206	42	an	an	DET
fcis-28893	206	43	occlusion	occlusion	NOUN
fcis-28893	206	44	module	module	NOUN
fcis-28893	206	45	.	.	PUNCT
fcis-28893	207	1	referring	refer	VERB
fcis-28893	207	2	to	to	ADP
fcis-28893	207	3	previous	previous	ADJ
fcis-28893	207	4	studies	study	NOUN
fcis-28893	207	5	,	,	PUNCT
fcis-28893	207	6	collecting	collect	VERB
fcis-28893	207	7	relevant	relevant	ADJ
fcis-28893	207	8	images	image	NOUN
fcis-28893	207	9	from	from	ADP
fcis-28893	207	10	pedestrian	pedestrian	NOUN
fcis-28893	207	11	re	re	NOUN
fcis-28893	207	12	-	-	NOUN
fcis-28893	207	13	identification	identification	NOUN
fcis-28893	207	14	and	and	CCONJ
fcis-28893	207	15	video	video	NOUN
fcis-28893	207	16	126	126	NUM
fcis-28893	207	17	pedestrian	pedestrian	NOUN
fcis-28893	207	18	retrieval	retrieval	NOUN
fcis-28893	207	19	datasets	dataset	NOUN
fcis-28893	207	20	is	be	AUX
fcis-28893	207	21	a	a	DET
fcis-28893	207	22	feasible	feasible	ADJ
fcis-28893	207	23	expansion	expansion	NOUN
fcis-28893	207	24	method	method	NOUN
fcis-28893	207	25	that	that	PRON
fcis-28893	207	26	can	can	AUX
fcis-28893	207	27	further	far	ADV
fcis-28893	207	28	promote	promote	VERB
fcis-28893	207	29	the	the	DET
fcis-28893	207	30	progress	progress	NOUN
fcis-28893	207	31	of	of	ADP
fcis-28893	207	32	text	text	NOUN
fcis-28893	207	33	-	-	PUNCT
fcis-28893	207	34	based	base	VERB
fcis-28893	207	35	pedestrian	pedestrian	NOUN
fcis-28893	207	36	retrieval	retrieval	NOUN
fcis-28893	207	37	technology	technology	NOUN
fcis-28893	207	38	.	.	PUNCT
fcis-28893	208	1	3	3	X
fcis-28893	208	2	)	)	PUNCT
fcis-28893	208	3	lightweight	lightweight	ADJ
fcis-28893	208	4	design	design	NOUN
fcis-28893	208	5	and	and	CCONJ
fcis-28893	208	6	improvement	improvement	NOUN
fcis-28893	208	7	.	.	PUNCT
fcis-28893	209	1	the	the	DET
fcis-28893	209	2	current	current	ADJ
fcis-28893	209	3	mainstream	mainstream	NOUN
fcis-28893	209	4	methods	method	NOUN
fcis-28893	209	5	,	,	PUNCT
fcis-28893	209	6	with	with	ADP
fcis-28893	209	7	huge	huge	ADJ
fcis-28893	209	8	computational	computational	ADJ
fcis-28893	209	9	overheads	overhead	NOUN
fcis-28893	209	10	,	,	PUNCT
fcis-28893	209	11	are	be	AUX
fcis-28893	209	12	difficult	difficult	ADJ
fcis-28893	209	13	to	to	PART
fcis-28893	209	14	be	be	AUX
fcis-28893	209	15	directly	directly	ADV
fcis-28893	209	16	applied	apply	VERB
fcis-28893	209	17	to	to	ADP
fcis-28893	209	18	real	real	ADJ
fcis-28893	209	19	-	-	PUNCT
fcis-28893	209	20	world	world	NOUN
fcis-28893	209	21	scenarios	scenario	NOUN
fcis-28893	209	22	.	.	PUNCT
fcis-28893	210	1	future	future	ADJ
fcis-28893	210	2	research	research	NOUN
fcis-28893	210	3	should	should	AUX
fcis-28893	210	4	focus	focus	VERB
fcis-28893	210	5	on	on	ADP
fcis-28893	210	6	the	the	DET
fcis-28893	210	7	lightweight	lightweight	ADJ
fcis-28893	210	8	improvement	improvement	NOUN
fcis-28893	210	9	of	of	ADP
fcis-28893	210	10	the	the	DET
fcis-28893	210	11	model	model	NOUN
fcis-28893	210	12	or	or	CCONJ
fcis-28893	210	13	designing	design	VERB
fcis-28893	210	14	new	new	ADJ
fcis-28893	210	15	lightweight	lightweight	ADJ
fcis-28893	210	16	models	model	NOUN
fcis-28893	210	17	to	to	PART
fcis-28893	210	18	reduce	reduce	VERB
fcis-28893	210	19	the	the	DET
fcis-28893	210	20	demand	demand	NOUN
fcis-28893	210	21	for	for	ADP
fcis-28893	210	22	computational	computational	ADJ
fcis-28893	210	23	resources	resource	NOUN
fcis-28893	210	24	and	and	CCONJ
fcis-28893	210	25	achieve	achieve	VERB
fcis-28893	210	26	more	more	ADV
fcis-28893	210	27	efficient	efficient	ADJ
fcis-28893	210	28	deployment	deployment	NOUN
fcis-28893	210	29	and	and	CCONJ
fcis-28893	210	30	retrieval	retrieval	NOUN
fcis-28893	210	31	.	.	PUNCT
fcis-28893	211	1	4	4	X
fcis-28893	211	2	)	)	PUNCT
fcis-28893	211	3	focus	focus	NOUN
fcis-28893	211	4	on	on	ADP
fcis-28893	211	5	the	the	DET
fcis-28893	211	6	research	research	NOUN
fcis-28893	211	7	hotspots	hotspot	NOUN
fcis-28893	211	8	in	in	ADP
fcis-28893	211	9	related	related	ADJ
fcis-28893	211	10	fields	field	NOUN
fcis-28893	211	11	.	.	PUNCT
fcis-28893	212	1	at	at	ADP
fcis-28893	212	2	present	present	ADJ
fcis-28893	212	3	,	,	PUNCT
fcis-28893	212	4	many	many	ADJ
fcis-28893	212	5	mainstream	mainstream	ADJ
fcis-28893	212	6	methods	method	NOUN
fcis-28893	212	7	draw	draw	VERB
fcis-28893	212	8	on	on	ADP
fcis-28893	212	9	the	the	DET
fcis-28893	212	10	technology	technology	NOUN
fcis-28893	212	11	and	and	CCONJ
fcis-28893	212	12	experience	experience	NOUN
fcis-28893	212	13	in	in	ADP
fcis-28893	212	14	the	the	DET
fcis-28893	212	15	field	field	NOUN
fcis-28893	212	16	of	of	ADP
fcis-28893	212	17	graphic	graphic	ADJ
fcis-28893	212	18	matching	matching	NOUN
fcis-28893	212	19	or	or	CCONJ
fcis-28893	212	20	pedestrian	pedestrian	NOUN
fcis-28893	212	21	re	re	NOUN
fcis-28893	212	22	-	-	NOUN
fcis-28893	212	23	identification	identification	NOUN
fcis-28893	212	24	.	.	PUNCT
fcis-28893	213	1	it	it	PRON
fcis-28893	213	2	should	should	AUX
fcis-28893	213	3	continuously	continuously	ADV
fcis-28893	213	4	pay	pay	VERB
fcis-28893	213	5	attention	attention	NOUN
fcis-28893	213	6	to	to	ADP
fcis-28893	213	7	the	the	DET
fcis-28893	213	8	progress	progress	NOUN
fcis-28893	213	9	in	in	ADP
fcis-28893	213	10	related	related	ADJ
fcis-28893	213	11	fields	field	NOUN
fcis-28893	213	12	to	to	PART
fcis-28893	213	13	discover	discover	VERB
fcis-28893	213	14	new	new	ADJ
fcis-28893	213	15	methods	method	NOUN
fcis-28893	213	16	and	and	CCONJ
fcis-28893	213	17	ideas	idea	NOUN
fcis-28893	213	18	.	.	PUNCT
fcis-28893	214	1	vlp	vlp	PROPN
fcis-28893	214	2	model	model	PROPN
fcis-28893	214	3	,	,	PUNCT
fcis-28893	214	4	as	as	ADP
fcis-28893	214	5	a	a	DET
fcis-28893	214	6	research	research	NOUN
fcis-28893	214	7	hotspot	hotspot	NOUN
fcis-28893	214	8	in	in	ADP
fcis-28893	214	9	recent	recent	ADJ
fcis-28893	214	10	years	year	NOUN
fcis-28893	214	11	,	,	PUNCT
fcis-28893	214	12	plays	play	VERB
fcis-28893	214	13	an	an	DET
fcis-28893	214	14	important	important	ADJ
fcis-28893	214	15	role	role	NOUN
fcis-28893	214	16	in	in	ADP
fcis-28893	214	17	text	text	NOUN
fcis-28893	214	18	-	-	PUNCT
fcis-28893	214	19	based	base	VERB
fcis-28893	214	20	pedestrian	pedestrian	NOUN
fcis-28893	214	21	retrieval	retrieval	NOUN
fcis-28893	214	22	.	.	PUNCT
fcis-28893	215	1	currently	currently	ADV
fcis-28893	215	2	,	,	PUNCT
fcis-28893	215	3	clip	clip	NOUN
fcis-28893	215	4	model	model	NOUN
fcis-28893	215	5	is	be	AUX
fcis-28893	215	6	mainly	mainly	ADV
fcis-28893	215	7	utilized	utilize	VERB
fcis-28893	215	8	as	as	ADP
fcis-28893	215	9	the	the	DET
fcis-28893	215	10	backbone	backbone	NOUN
fcis-28893	215	11	network	network	NOUN
fcis-28893	215	12	in	in	ADP
fcis-28893	215	13	the	the	DET
fcis-28893	215	14	field	field	NOUN
fcis-28893	215	15	,	,	PUNCT
fcis-28893	215	16	which	which	PRON
fcis-28893	215	17	can	can	AUX
fcis-28893	215	18	be	be	AUX
fcis-28893	215	19	mined	mine	VERB
fcis-28893	215	20	to	to	PART
fcis-28893	215	21	be	be	AUX
fcis-28893	215	22	more	more	ADV
fcis-28893	215	23	suitable	suitable	ADJ
fcis-28893	215	24	for	for	ADP
fcis-28893	215	25	the	the	DET
fcis-28893	215	26	field	field	NOUN
fcis-28893	215	27	of	of	ADP
fcis-28893	215	28	vlp	vlp	PROPN
fcis-28893	215	29	model	model	NOUN
fcis-28893	215	30	,	,	PUNCT
fcis-28893	215	31	e.g.	e.g.	ADV
fcis-28893	215	32	,	,	PUNCT
fcis-28893	215	33	bai	bai	PROPN
fcis-28893	215	34	et	et	PROPN
fcis-28893	215	35	al.[21	al.[21	PROPN
fcis-28893	215	36	]	]	PUNCT
fcis-28893	215	37	based	base	VERB
fcis-28893	215	38	on	on	ADP
fcis-28893	215	39	the	the	DET
fcis-28893	215	40	albef	albef	NOUN
fcis-28893	215	41	model	model	NOUN
fcis-28893	215	42	proposed	propose	VERB
fcis-28893	215	43	the	the	DET
fcis-28893	215	44	rasa	rasa	PROPN
fcis-28893	215	45	model	model	NOUN
fcis-28893	215	46	to	to	PART
fcis-28893	215	47	further	far	ADV
fcis-28893	215	48	advance	advance	VERB
fcis-28893	215	49	the	the	DET
fcis-28893	215	50	domain	domain	NOUN
fcis-28893	215	51	development	development	NOUN
fcis-28893	215	52	.	.	PUNCT
fcis-28893	216	1	therefore	therefore	ADV
fcis-28893	216	2	,	,	PUNCT
fcis-28893	216	3	in	in	ADP
fcis-28893	216	4	-	-	PUNCT
fcis-28893	216	5	depth	depth	NOUN
fcis-28893	216	6	exploration	exploration	NOUN
fcis-28893	216	7	of	of	ADP
fcis-28893	216	8	the	the	DET
fcis-28893	216	9	application	application	NOUN
fcis-28893	216	10	and	and	CCONJ
fcis-28893	216	11	optimization	optimization	NOUN
fcis-28893	216	12	of	of	ADP
fcis-28893	216	13	vlp	vlp	PROPN
fcis-28893	216	14	models	model	NOUN
fcis-28893	216	15	is	be	AUX
fcis-28893	216	16	an	an	DET
fcis-28893	216	17	important	important	ADJ
fcis-28893	216	18	direction	direction	NOUN
fcis-28893	216	19	for	for	ADP
fcis-28893	216	20	future	future	ADJ
fcis-28893	216	21	research	research	NOUN
fcis-28893	216	22	.	.	PUNCT
fcis-28893	217	1	table	table	NOUN
fcis-28893	217	2	4	4	NUM
fcis-28893	217	3	.	.	PUNCT
fcis-28893	217	4	comparison	comparison	NOUN
fcis-28893	217	5	of	of	ADP
fcis-28893	217	6	scores	score	NOUN
fcis-28893	217	7	on	on	ADP
fcis-28893	217	8	the	the	DET
fcis-28893	217	9	chuk	chuk	PROPN
fcis-28893	217	10	-	-	PUNCT
fcis-28893	217	11	pedes	pes	NOUN
fcis-28893	217	12	,	,	PUNCT
fcis-28893	217	13	icfg	icfg	NOUN
fcis-28893	217	14	-	-	NOUN
fcis-28893	217	15	pedes	pes	NOUN
fcis-28893	217	16	,	,	PUNCT
fcis-28893	217	17	and	and	CCONJ
fcis-28893	217	18	rstpreid	rstpreid	NOUN
fcis-28893	217	19	datasets	dataset	NOUN
fcis-28893	217	20	category	category	NOUN
fcis-28893	217	21	method	method	VERB
fcis-28893	217	22	image	image	NOUN
fcis-28893	217	23	encoder	encoder	NOUN
fcis-28893	217	24	text	text	NOUN
fcis-28893	217	25	encoder	encoder	NOUN
fcis-28893	217	26	rank-1	rank-1	ADP
fcis-28893	217	27	rank-5	rank-5	CCONJ
fcis-28893	217	28	rank-10	rank-10	PROPN
fcis-28893	217	29	chukpedes	chukpede	NOUN
fcis-28893	217	30	icfgpedes	icfgpede	VERB
fcis-28893	217	31	rstpreid	rstpreid	PROPN
fcis-28893	217	32	chukpedes	chukpede	VERB
fcis-28893	217	33	icfgpedes	icfgpede	NOUN
fcis-28893	217	34	rstpreid	rstpreid	PROPN
fcis-28893	217	35	chukpedes	chukpede	VERB
fcis-28893	217	36	icfgpedes	icfgpede	VERB
fcis-28893	217	37	rstpreid	rstpreid	PROPN
fcis-28893	217	38	global	global	ADJ
fcis-28893	217	39	feature	feature	NOUN
fcis-28893	217	40	gna	gna	PROPN
fcis-28893	217	41	rnn	rnn	VERB
fcis-28893	217	42	vgg16	vgg16	PROPN
fcis-28893	217	43	lstm	lstm	PROPN
fcis-28893	217	44	19.05	19.05	NUM
fcis-28893	217	45	n	n	NOUN
fcis-28893	217	46	/	/	SYM
fcis-28893	217	47	a	a	DET
fcis-28893	217	48	n	n	NOUN
fcis-28893	217	49	/	/	SYM
fcis-28893	217	50	a	a	DET
fcis-28893	217	51	n	n	NOUN
fcis-28893	217	52	/	/	SYM
fcis-28893	217	53	a	a	DET
fcis-28893	217	54	n	n	NOUN
fcis-28893	217	55	/	/	SYM
fcis-28893	217	56	a	a	DET
fcis-28893	217	57	n	n	NOUN
fcis-28893	217	58	/	/	SYM
fcis-28893	217	59	a	a	DET
fcis-28893	217	60	53.64	53.64	NUM
fcis-28893	217	61	n	n	NOUN
fcis-28893	217	62	/	/	SYM
fcis-28893	217	63	a	a	PRON
fcis-28893	217	64	n	n	NOUN
fcis-28893	217	65	/	/	SYM
fcis-28893	217	66	a	a	DET
fcis-28893	217	67	iatv	iatv	PROPN
fcis-28893	217	68	vgg16	vgg16	NOUN
fcis-28893	217	69	lstm	lstm	NOUN
fcis-28893	217	70	25.94	25.94	NUM
fcis-28893	217	71	n	n	CCONJ
fcis-28893	217	72	/	/	SYM
fcis-28893	217	73	a	a	DET
fcis-28893	217	74	n	n	NOUN
fcis-28893	217	75	/	/	SYM
fcis-28893	217	76	a	a	DET
fcis-28893	217	77	n	n	NOUN
fcis-28893	217	78	/	/	SYM
fcis-28893	217	79	a	a	DET
fcis-28893	217	80	n	n	NOUN
fcis-28893	217	81	/	/	SYM
fcis-28893	217	82	a	a	DET
fcis-28893	217	83	n	n	NOUN
fcis-28893	217	84	/	/	SYM
fcis-28893	217	85	a	a	DET
fcis-28893	217	86	60.48	60.48	NUM
fcis-28893	217	87	n	n	CCONJ
fcis-28893	217	88	/	/	SYM
fcis-28893	217	89	a	a	DET
fcis-28893	217	90	n	n	CCONJ
fcis-28893	217	91	/	/	SYM
fcis-28893	217	92	a	a	DET
fcis-28893	217	93	dual	dual	ADJ
fcis-28893	217	94	-	-	PUNCT
fcis-28893	217	95	path	path	NOUN
fcis-28893	217	96	resnet-50	resnet-50	PROPN
fcis-28893	217	97	resnet-50	resnet-50	PROPN
fcis-28893	217	98	n	n	CCONJ
fcis-28893	217	99	/	/	SYM
fcis-28893	217	100	a	a	DET
fcis-28893	217	101	38.99	38.99	NUM
fcis-28893	217	102	n	n	NOUN
fcis-28893	217	103	/	/	SYM
fcis-28893	217	104	a	a	DET
fcis-28893	217	105	n	n	NOUN
fcis-28893	217	106	/	/	SYM
fcis-28893	217	107	a	a	DET
fcis-28893	217	108	59.44	59.44	NUM
fcis-28893	217	109	n	n	NOUN
fcis-28893	217	110	/	/	SYM
fcis-28893	217	111	a	a	DET
fcis-28893	217	112	n	n	NOUN
fcis-28893	217	113	/	/	SYM
fcis-28893	217	114	a	a	DET
fcis-28893	217	115	68.41	68.41	NUM
fcis-28893	217	116	n	n	CCONJ
fcis-28893	217	117	/	/	SYM
fcis-28893	217	118	a	a	DET
fcis-28893	217	119	pwm	pwm	NOUN
fcis-28893	217	120	vgg16	vgg16	NOUN
fcis-28893	217	121	lstm	lstm	NOUN
fcis-28893	217	122	27.14	27.14	NUM
fcis-28893	217	123	n	n	NOUN
fcis-28893	217	124	/	/	SYM
fcis-28893	217	125	a	a	DET
fcis-28893	217	126	n	n	NOUN
fcis-28893	217	127	/	/	SYM
fcis-28893	217	128	a	a	DET
fcis-28893	217	129	49.45	49.45	NUM
fcis-28893	217	130	n	n	CCONJ
fcis-28893	217	131	/	/	SYM
fcis-28893	217	132	a	a	DET
fcis-28893	217	133	n	n	NOUN
fcis-28893	217	134	/	/	SYM
fcis-28893	217	135	a	a	DET
fcis-28893	217	136	61.02	61.02	NUM
fcis-28893	217	137	n	n	CCONJ
fcis-28893	217	138	/	/	SYM
fcis-28893	217	139	a	a	PRON
fcis-28893	217	140	n	n	NOUN
fcis-28893	217	141	/	/	SYM
fcis-28893	217	142	a	a	DET
fcis-28893	217	143	cmpm+cmpc	cmpm+cmpc	PROPN
fcis-28893	217	144	mobilenet	mobilenet	NOUN
fcis-28893	217	145	lstm	lstm	VERB
fcis-28893	217	146	49.37	49.37	NUM
fcis-28893	217	147	43.51	43.51	NUM
fcis-28893	217	148	n	n	CCONJ
fcis-28893	217	149	/	/	SYM
fcis-28893	217	150	a	a	DET
fcis-28893	217	151	n	n	NOUN
fcis-28893	217	152	/	/	SYM
fcis-28893	217	153	a	a	DET
fcis-28893	217	154	65.44	65.44	NUM
fcis-28893	217	155	n	n	CCONJ
fcis-28893	217	156	/	/	SYM
fcis-28893	217	157	a	a	DET
fcis-28893	217	158	79.27	79.27	NUM
fcis-28893	217	159	74.26	74.26	NUM
fcis-28893	217	160	n	n	CCONJ
fcis-28893	217	161	/	/	SYM
fcis-28893	217	162	a	a	DET
fcis-28893	217	163	multigranularity	multigranularity	NOUN
fcis-28893	217	164	pma	pma	NOUN
fcis-28893	217	165	vgg16/	vgg16/	PROPN
fcis-28893	217	166	resnet-50	resnet-50	PROPN
fcis-28893	217	167	lstm	lstm	NOUN
fcis-28893	217	168	53.81	53.81	NUM
fcis-28893	217	169	n	n	NOUN
fcis-28893	217	170	/	/	SYM
fcis-28893	218	1	a	a	DET
fcis-28893	218	2	n	n	NOUN
fcis-28893	218	3	/	/	SYM
fcis-28893	218	4	a	a	DET
fcis-28893	218	5	73.54	73.54	NUM
fcis-28893	218	6	n	n	CCONJ
fcis-28893	218	7	/	/	SYM
fcis-28893	218	8	a	a	DET
fcis-28893	218	9	n	n	NOUN
fcis-28893	218	10	/	/	SYM
fcis-28893	218	11	a	a	PRON
fcis-28893	218	12	81.23	81.23	NUM
fcis-28893	218	13	n	n	CCONJ
fcis-28893	218	14	/	/	SYM
fcis-28893	218	15	a	a	PRON
fcis-28893	218	16	n	n	CCONJ
fcis-28893	218	17	/	/	SYM
fcis-28893	218	18	a	a	DET
fcis-28893	218	19	tvfr	tvfr	ADJ
fcis-28893	218	20	mobilenet	mobilenet	NOUN
fcis-28893	218	21	lstm	lstm	PROPN
fcis-28893	218	22	53.87	53.87	NUM
fcis-28893	218	23	n	n	CCONJ
fcis-28893	218	24	/	/	SYM
fcis-28893	218	25	a	a	DET
fcis-28893	218	26	n	n	NOUN
fcis-28893	218	27	/	/	SYM
fcis-28893	218	28	a	a	DET
fcis-28893	218	29	75.25	75.25	NUM
fcis-28893	218	30	n	n	NOUN
fcis-28893	218	31	/	/	SYM
fcis-28893	218	32	a	a	DET
fcis-28893	218	33	n	n	NOUN
fcis-28893	218	34	/	/	SYM
fcis-28893	218	35	a	a	DET
fcis-28893	218	36	83.47	83.47	NUM
fcis-28893	218	37	n	n	CCONJ
fcis-28893	218	38	/	/	SYM
fcis-28893	218	39	a	a	DET
fcis-28893	218	40	n	n	NOUN
fcis-28893	218	41	/	/	SYM
fcis-28893	218	42	a	a	DET
fcis-28893	218	43	mia	mia	PROPN
fcis-28893	218	44	resnet-50	resnet-50	PROPN
fcis-28893	218	45	bi	bi	PROPN
fcis-28893	218	46	-	-	NOUN
fcis-28893	218	47	gru	gru	NOUN
fcis-28893	218	48	53.10	53.10	NUM
fcis-28893	218	49	46.49	46.49	NUM
fcis-28893	218	50	n	n	X
fcis-28893	218	51	/	/	SYM
fcis-28893	218	52	a	a	PRON
fcis-28893	218	53	75.00	75.00	NUM
fcis-28893	218	54	67.14	67.14	NUM
fcis-28893	218	55	n	n	CCONJ
fcis-28893	218	56	/	/	SYM
fcis-28893	218	57	a	a	DET
fcis-28893	218	58	82.90	82.90	NUM
fcis-28893	218	59	75.18	75.18	NUM
fcis-28893	218	60	n	n	CCONJ
fcis-28893	218	61	/	/	SYM
fcis-28893	218	62	a	a	DET
fcis-28893	218	63	vitaa	vitaa	PROPN
fcis-28893	218	64	resnet-50	resnet-50	PROPN
fcis-28893	218	65	lstm	lstm	NOUN
fcis-28893	218	66	55.97	55.97	NUM
fcis-28893	218	67	50.98	50.98	NUM
fcis-28893	218	68	n	n	NOUN
fcis-28893	218	69	/	/	SYM
fcis-28893	218	70	a	a	DET
fcis-28893	218	71	75.84	75.84	NUM
fcis-28893	218	72	68.79	68.79	NUM
fcis-28893	218	73	n	n	CCONJ
fcis-28893	218	74	/	/	SYM
fcis-28893	218	75	a	a	DET
fcis-28893	218	76	83.52	83.52	NUM
fcis-28893	218	77	75.78	75.78	NUM
fcis-28893	218	78	n	n	CCONJ
fcis-28893	218	79	/	/	SYM
fcis-28893	218	80	a	a	DET
fcis-28893	218	81	tipcb	tipcb	NOUN
fcis-28893	218	82	resnet-50	resnet-50	PROPN
fcis-28893	218	83	bert	bert	PROPN
fcis-28893	218	84	63.63	63.63	NUM
fcis-28893	218	85	54.96	54.96	NUM
fcis-28893	218	86	n	n	CCONJ
fcis-28893	218	87	/	/	SYM
fcis-28893	218	88	a	a	DET
fcis-28893	218	89	82.82	82.82	NUM
fcis-28893	218	90	74.72	74.72	NUM
fcis-28893	218	91	n	n	NOUN
fcis-28893	218	92	/	/	SYM
fcis-28893	218	93	a	a	PRON
fcis-28893	218	94	89.01	89.01	NUM
fcis-28893	218	95	81.89	81.89	NUM
fcis-28893	218	96	n	n	NOUN
fcis-28893	218	97	/	/	SYM
fcis-28893	218	98	a	a	DET
fcis-28893	218	99	caibc	caibc	NOUN
fcis-28893	218	100	resnet-50	resnet-50	PROPN
fcis-28893	218	101	bi	bi	PROPN
fcis-28893	218	102	-	-	NOUN
fcis-28893	218	103	gru	gru	NOUN
fcis-28893	218	104	64.43	64.43	NUM
fcis-28893	218	105	n	n	NOUN
fcis-28893	218	106	/	/	SYM
fcis-28893	218	107	a	a	DET
fcis-28893	218	108	47.35	47.35	NUM
fcis-28893	218	109	82.87	82.87	NUM
fcis-28893	218	110	n	n	NOUN
fcis-28893	218	111	/	/	SYM
fcis-28893	218	112	a	a	DET
fcis-28893	218	113	69.55	69.55	NUM
fcis-28893	218	114	88.37	88.37	NUM
fcis-28893	218	115	n	n	CCONJ
fcis-28893	218	116	/	/	SYM
fcis-28893	218	117	a	a	DET
fcis-28893	218	118	79.00	79.00	NUM
fcis-28893	218	119	manet	manet	NOUN
fcis-28893	218	120	resnet-50	resnet-50	PROPN
fcis-28893	218	121	bert	bert	PROPN
fcis-28893	218	122	65.64	65.64	NUM
fcis-28893	218	123	59.44	59.44	NUM
fcis-28893	218	124	n	n	CCONJ
fcis-28893	218	125	/	/	SYM
fcis-28893	218	126	a	a	DET
fcis-28893	218	127	83.01	83.01	NUM
fcis-28893	218	128	76.80	76.80	NUM
fcis-28893	218	129	n	n	NOUN
fcis-28893	218	130	/	/	SYM
fcis-28893	218	131	a	a	DET
fcis-28893	218	132	88.78	88.78	NUM
fcis-28893	218	133	82.75	82.75	NUM
fcis-28893	218	134	n	n	CCONJ
fcis-28893	218	135	/	/	SYM
fcis-28893	218	136	a	a	DET
fcis-28893	218	137	ssan	ssan	NOUN
fcis-28893	218	138	resnet-50	resnet-50	PROPN
fcis-28893	218	139	lstm	lstm	NOUN
fcis-28893	218	140	61.37	61.37	NUM
fcis-28893	218	141	54.23	54.23	NUM
fcis-28893	218	142	43.50	43.50	NUM
fcis-28893	218	143	80.15	80.15	NUM
fcis-28893	218	144	72.63	72.63	NUM
fcis-28893	218	145	67.80	67.80	NUM
fcis-28893	218	146	86.73	86.73	NUM
fcis-28893	218	147	79.53	79.53	NUM
fcis-28893	218	148	77.15	77.15	NUM
fcis-28893	218	149	adversarial	adversarial	ADJ
fcis-28893	218	150	learning	learn	VERB
fcis-28893	218	151	a	a	DET
fcis-28893	218	152	-	-	PUNCT
fcis-28893	218	153	ganet	ganet	NOUN
fcis-28893	218	154	resnet-50	resnet-50	PROPN
fcis-28893	218	155	lstm	lstm	NOUN
fcis-28893	218	156	39.52	39.52	NUM
fcis-28893	218	157	n	n	NOUN
fcis-28893	218	158	/	/	SYM
fcis-28893	218	159	a	a	DET
fcis-28893	218	160	n	n	NOUN
fcis-28893	218	161	/	/	SYM
fcis-28893	218	162	a	a	PRON
fcis-28893	218	163	69.91	69.91	NUM
fcis-28893	218	164	n	n	CCONJ
fcis-28893	218	165	/	/	SYM
fcis-28893	218	166	a	a	DET
fcis-28893	218	167	n	n	NOUN
fcis-28893	218	168	/	/	SYM
fcis-28893	218	169	a	a	DET
fcis-28893	218	170	80.91	80.91	NUM
fcis-28893	218	171	n	n	NOUN
fcis-28893	218	172	/	/	SYM
fcis-28893	218	173	a	a	PRON
fcis-28893	218	174	n	n	CCONJ
fcis-28893	218	175	/	/	SYM
fcis-28893	218	176	a	a	DET
fcis-28893	218	177	timam	timam	ADJ
fcis-28893	218	178	resnet-101	resnet-101	NOUN
fcis-28893	218	179	bert	bert	PROPN
fcis-28893	218	180	54.51	54.51	NUM
fcis-28893	218	181	n	n	NOUN
fcis-28893	218	182	/	/	SYM
fcis-28893	218	183	a	a	DET
fcis-28893	218	184	n	n	NOUN
fcis-28893	218	185	/	/	SYM
fcis-28893	218	186	a	a	DET
fcis-28893	218	187	77.56	77.56	NUM
fcis-28893	218	188	n	n	CCONJ
fcis-28893	218	189	/	/	SYM
fcis-28893	218	190	a	a	DET
fcis-28893	218	191	n	n	NOUN
fcis-28893	218	192	/	/	SYM
fcis-28893	218	193	a	a	DET
fcis-28893	218	194	84.78	84.78	NUM
fcis-28893	218	195	n	n	CCONJ
fcis-28893	218	196	/	/	SYM
fcis-28893	218	197	a	a	DET
fcis-28893	218	198	n	n	NOUN
fcis-28893	218	199	/	/	SYM
fcis-28893	218	200	a	a	DET
fcis-28893	218	201	crossmodal	crossmodal	ADJ
fcis-28893	218	202	attention	attention	NOUN
fcis-28893	218	203	dssl	dssl	VERB
fcis-28893	218	204	resnet-50	resnet-50	PROPN
fcis-28893	218	205	bi	bi	PROPN
fcis-28893	218	206	-	-	NOUN
fcis-28893	218	207	gru	gru	NOUN
fcis-28893	218	208	62.33	62.33	NUM
fcis-28893	218	209	n	n	CCONJ
fcis-28893	218	210	/	/	SYM
fcis-28893	218	211	a	a	DET
fcis-28893	218	212	39.05	39.05	NUM
fcis-28893	218	213	82.11	82.11	NUM
fcis-28893	218	214	n	n	CCONJ
fcis-28893	218	215	/	/	SYM
fcis-28893	218	216	a	a	DET
fcis-28893	218	217	62.60	62.60	NUM
fcis-28893	218	218	88.01	88.01	NUM
fcis-28893	218	219	n	n	CCONJ
fcis-28893	218	220	/	/	SYM
fcis-28893	218	221	a	a	DET
fcis-28893	218	222	73.95	73.95	NUM
fcis-28893	218	223	nafs	nafs	PROPN
fcis-28893	218	224	resnet-50	resnet-50	PROPN
fcis-28893	218	225	bert	bert	PROPN
fcis-28893	218	226	61.50	61.50	NUM
fcis-28893	218	227	n	n	NOUN
fcis-28893	218	228	/	/	SYM
fcis-28893	218	229	a	a	DET
fcis-28893	218	230	n	n	NOUN
fcis-28893	218	231	/	/	SYM
fcis-28893	218	232	a	a	DET
fcis-28893	218	233	81.19	81.19	NUM
fcis-28893	218	234	n	n	CCONJ
fcis-28893	218	235	/	/	SYM
fcis-28893	218	236	a	a	DET
fcis-28893	218	237	n	n	NOUN
fcis-28893	218	238	/	/	SYM
fcis-28893	218	239	a	a	PRON
fcis-28893	218	240	87.51	87.51	NUM
fcis-28893	218	241	n	n	NOUN
fcis-28893	218	242	/	/	SYM
fcis-28893	218	243	a	a	DET
fcis-28893	218	244	n	n	NOUN
fcis-28893	218	245	/	/	SYM
fcis-28893	218	246	a	a	DET
fcis-28893	218	247	acsa	acsa	NOUN
fcis-28893	218	248	swin	swin	PROPN
fcis-28893	218	249	transformer	transformer	PROPN
fcis-28893	218	250	bert	bert	PROPN
fcis-28893	218	251	68.67	68.67	NUM
fcis-28893	218	252	n	n	CCONJ
fcis-28893	218	253	/	/	SYM
fcis-28893	218	254	a	a	DET
fcis-28893	218	255	n	n	NOUN
fcis-28893	218	256	/	/	SYM
fcis-28893	218	257	a	a	DET
fcis-28893	218	258	85.61	85.61	NUM
fcis-28893	218	259	n	n	CCONJ
fcis-28893	218	260	/	/	SYM
fcis-28893	218	261	a	a	DET
fcis-28893	218	262	n	n	NOUN
fcis-28893	218	263	/	/	SYM
fcis-28893	218	264	a	a	DET
fcis-28893	218	265	90.66	90.66	NUM
fcis-28893	218	266	n	n	CCONJ
fcis-28893	218	267	/	/	SYM
fcis-28893	218	268	a	a	PRON
fcis-28893	218	269	n	n	NOUN
fcis-28893	218	270	/	/	SYM
fcis-28893	218	271	a	a	DET
fcis-28893	218	272	ccmf	ccmf	ADJ
fcis-28893	218	273	vit	vit	NOUN
fcis-28893	218	274	bert	bert	PROPN
fcis-28893	218	275	73.81	73.81	NUM
fcis-28893	218	276	n	n	CCONJ
fcis-28893	218	277	/	/	SYM
fcis-28893	218	278	a	a	DET
fcis-28893	218	279	57.35	57.35	NUM
fcis-28893	218	280	88.89	88.89	NUM
fcis-28893	218	281	n	n	CCONJ
fcis-28893	218	282	/	/	SYM
fcis-28893	218	283	a	a	DET
fcis-28893	218	284	77.50	77.50	NUM
fcis-28893	218	285	92.77	92.77	NUM
fcis-28893	218	286	n	n	CCONJ
fcis-28893	218	287	/	/	SYM
fcis-28893	218	288	a	a	DET
fcis-28893	218	289	85.50	85.50	NUM
fcis-28893	218	290	vlp	vlp	PROPN
fcis-28893	218	291	psld	psld	PROPN
fcis-28893	218	292	resnet-50	resnet-50	PROPN
fcis-28893	218	293	clip	clip	NOUN
fcis-28893	218	294	64.08	64.08	NUM
fcis-28893	218	295	n	n	CCONJ
fcis-28893	218	296	/	/	SYM
fcis-28893	218	297	a	a	DET
fcis-28893	218	298	n	n	NOUN
fcis-28893	218	299	/	/	SYM
fcis-28893	218	300	a	a	DET
fcis-28893	218	301	81.73	81.73	NUM
fcis-28893	218	302	n	n	NOUN
fcis-28893	218	303	/	/	SYM
fcis-28893	218	304	a	a	DET
fcis-28893	218	305	n	n	NOUN
fcis-28893	218	306	/	/	SYM
fcis-28893	218	307	a	a	DET
fcis-28893	218	308	88.19	88.19	NUM
fcis-28893	218	309	n	n	CCONJ
fcis-28893	218	310	/	/	SYM
fcis-28893	218	311	a	a	DET
fcis-28893	218	312	n	n	NOUN
fcis-28893	218	313	/	/	SYM
fcis-28893	218	314	a	a	DET
fcis-28893	218	315	cfine	cfine	ADJ
fcis-28893	218	316	clip	clip	NOUN
fcis-28893	218	317	bert	bert	PROPN
fcis-28893	218	318	69.57	69.57	NUM
fcis-28893	218	319	60.83	60.83	NUM
fcis-28893	218	320	50.55	50.55	NUM
fcis-28893	218	321	85.93	85.93	NUM
fcis-28893	218	322	76.55	76.55	NUM
fcis-28893	218	323	72.50	72.50	NUM
fcis-28893	218	324	91.15	91.15	NUM
fcis-28893	218	325	82.42	82.42	NUM
fcis-28893	218	326	81.60	81.60	NUM
fcis-28893	218	327	cskt	cskt	PROPN
fcis-28893	218	328	clip	clip	NOUN
fcis-28893	218	329	clip	clip	VERB
fcis-28893	218	330	69.70	69.70	NUM
fcis-28893	218	331	58.90	58.90	NUM
fcis-28893	218	332	57.75	57.75	NUM
fcis-28893	218	333	86.92	86.92	NUM
fcis-28893	218	334	77.31	77.31	NUM
fcis-28893	218	335	81.30	81.30	NUM
fcis-28893	218	336	91.80	91.80	NUM
fcis-28893	218	337	83.56	83.56	NUM
fcis-28893	218	338	88.35	88.35	NUM
fcis-28893	218	339	irra	irra	ADJ
fcis-28893	218	340	clip	clip	NOUN
fcis-28893	218	341	clip	clip	NOUN
fcis-28893	218	342	73.38	73.38	NUM
fcis-28893	218	343	63.46	63.46	NUM
fcis-28893	218	344	60.20	60.20	NUM
fcis-28893	218	345	89.93	89.93	NUM
fcis-28893	218	346	80.25	80.25	NUM
fcis-28893	218	347	81.30	81.30	NUM
fcis-28893	218	348	93.71	93.71	NUM
fcis-28893	218	349	85.82	85.82	NUM
fcis-28893	218	350	88.20	88.20	NUM
fcis-28893	218	351	rasa	rasa	PROPN
fcis-28893	218	352	albef	albef	PROPN
fcis-28893	218	353	albef	albef	PROPN
fcis-28893	218	354	76.51	76.51	NUM
fcis-28893	218	355	65.28	65.28	NUM
fcis-28893	218	356	66.90	66.90	NUM
fcis-28893	218	357	90.29	90.29	NUM
fcis-28893	218	358	80.40	80.40	NUM
fcis-28893	218	359	86.50	86.50	NUM
fcis-28893	218	360	94.25	94.25	NUM
fcis-28893	218	361	85.12	85.12	NUM
fcis-28893	218	362	91.35	91.35	NUM
fcis-28893	218	363	aul	aul	PROPN
fcis-28893	218	364	swin	swin	PROPN
fcis-28893	218	365	transformer	transformer	PROPN
fcis-28893	218	366	bert	bert	PROPN
fcis-28893	218	367	77.23	77.23	NUM
fcis-28893	218	368	69.16	69.16	NUM
fcis-28893	218	369	71.65	71.65	NUM
fcis-28893	218	370	90.43	90.43	NUM
fcis-28893	218	371	83.32	83.32	NUM
fcis-28893	218	372	87.55	87.55	NUM
fcis-28893	218	373	94.41	94.41	NUM
fcis-28893	218	374	88.37	88.37	NUM
fcis-28893	218	375	92.05	92.05	NUM
fcis-28893	218	376	sap	sap	PROPN
fcis-28893	218	377	-	-	PUNCT
fcis-28893	218	378	sam	sam	PROPN
fcis-28893	218	379	sam	sam	PROPN
fcis-28893	218	380	-	-	PUNCT
fcis-28893	218	381	vit	vit	PROPN
fcis-28893	218	382	bert	bert	PROPN
fcis-28893	218	383	75.05	75.05	NUM
fcis-28893	218	384	63.97	63.97	NUM
fcis-28893	218	385	62.85	62.85	NUM
fcis-28893	218	386	89.93	89.93	NUM
fcis-28893	218	387	80.84	80.84	NUM
fcis-28893	218	388	82.65	82.65	NUM
fcis-28893	218	389	93.73	93.73	NUM
fcis-28893	218	390	86.17	86.17	NUM
fcis-28893	218	391	89.85	89.85	NUM
fcis-28893	218	392	note	note	NOUN
fcis-28893	218	393	:	:	PUNCT
fcis-28893	218	394	bolded	bolde	VERB
fcis-28893	218	395	text	text	NOUN
fcis-28893	218	396	indicates	indicate	VERB
fcis-28893	218	397	optimal	optimal	ADJ
fcis-28893	218	398	values	value	NOUN
fcis-28893	218	399	,	,	PUNCT
fcis-28893	218	400	"	"	PUNCT
fcis-28893	218	401	n	n	CCONJ
fcis-28893	218	402	/	/	SYM
fcis-28893	218	403	a	a	PRON
fcis-28893	218	404	"	"	PUNCT
fcis-28893	218	405	indicates	indicate	VERB
fcis-28893	218	406	that	that	SCONJ
fcis-28893	218	407	the	the	DET
fcis-28893	218	408	relevant	relevant	ADJ
fcis-28893	218	409	data	datum	NOUN
fcis-28893	218	410	are	be	AUX
fcis-28893	218	411	not	not	PART
fcis-28893	218	412	available	available	ADJ
fcis-28893	218	413	.	.	PUNCT
fcis-28893	219	1	acknowledgments	acknowledgment	NOUN
fcis-28893	219	2	this	this	DET
fcis-28893	219	3	work	work	NOUN
fcis-28893	219	4	is	be	AUX
fcis-28893	219	5	supported	support	VERB
fcis-28893	219	6	by	by	ADP
fcis-28893	219	7	the	the	DET
fcis-28893	219	8	double	double	ADJ
fcis-28893	219	9	first	first	ADJ
fcis-28893	219	10	-	-	PUNCT
fcis-28893	219	11	class	class	NOUN
fcis-28893	219	12	special	special	ADJ
fcis-28893	219	13	project	project	NOUN
fcis-28893	219	14	of	of	ADP
fcis-28893	219	15	security	security	NOUN
fcis-28893	219	16	and	and	CCONJ
fcis-28893	219	17	prevention	prevention	NOUN
fcis-28893	219	18	project	project	NOUN
fcis-28893	219	19	of	of	ADP
fcis-28893	219	20	people	people	NOUN
fcis-28893	219	21	's	's	PART
fcis-28893	219	22	public	public	ADJ
fcis-28893	219	23	security	security	NOUN
fcis-28893	219	24	university	university	PROPN
fcis-28893	219	25	of	of	ADP
fcis-28893	219	26	china	china	PROPN
fcis-28893	219	27	(	(	PUNCT
fcis-28893	219	28	2023syl08	2023syl08	NUM
fcis-28893	219	29	)	)	PUNCT
fcis-28893	219	30	.	.	PUNCT
fcis-28893	220	1	references	reference	NOUN
fcis-28893	220	2	[	[	X
fcis-28893	220	3	1	1	X
fcis-28893	220	4	]	]	PUNCT
fcis-28893	220	5	s.	s.	PROPN
fcis-28893	220	6	li	li	PROPN
fcis-28893	220	7	,	,	PUNCT
fcis-28893	220	8	t.	t.	PROPN
fcis-28893	220	9	xiao	xiao	PROPN
fcis-28893	220	10	,	,	PUNCT
fcis-28893	220	11	h.	h.	PROPN
fcis-28893	220	12	li	li	PROPN
fcis-28893	220	13	,	,	PUNCT
fcis-28893	220	14	b.	b.	PROPN
fcis-28893	220	15	zhou	zhou	PROPN
fcis-28893	220	16	,	,	PUNCT
fcis-28893	220	17	d.	d.	PROPN
fcis-28893	220	18	yue	yue	PROPN
fcis-28893	220	19	,	,	PUNCT
fcis-28893	220	20	and	and	CCONJ
fcis-28893	220	21	x.	x.	PROPN
fcis-28893	220	22	wang	wang	PROPN
fcis-28893	220	23	,	,	PUNCT
fcis-28893	220	24	"	"	PUNCT
fcis-28893	220	25	person	person	NOUN
fcis-28893	220	26	search	search	VERB
fcis-28893	220	27	with	with	ADP
fcis-28893	220	28	natural	natural	ADJ
fcis-28893	220	29	language	language	NOUN
fcis-28893	220	30	description	description	NOUN
fcis-28893	220	31	,	,	PUNCT
fcis-28893	220	32	"	"	PUNCT
fcis-28893	220	33	in	in	ADP
fcis-28893	220	34	2017	2017	NUM
fcis-28893	220	35	ieee	ieee	NOUN
fcis-28893	220	36	conference	conference	NOUN
fcis-28893	220	37	on	on	ADP
fcis-28893	220	38	computer	computer	NOUN
fcis-28893	220	39	vision	vision	NOUN
fcis-28893	220	40	and	and	CCONJ
fcis-28893	220	41	pattern	pattern	NOUN
fcis-28893	220	42	recognition	recognition	NOUN
fcis-28893	220	43	(	(	PUNCT
fcis-28893	220	44	cvpr	cvpr	NOUN
fcis-28893	220	45	)	)	PUNCT
fcis-28893	220	46	,	,	PUNCT
fcis-28893	220	47	21	21	NUM
fcis-28893	220	48	-	-	SYM
fcis-28893	220	49	26	26	NUM
fcis-28893	220	50	july	july	PROPN
fcis-28893	220	51	2017	2017	NUM
fcis-28893	220	52	2017	2017	NUM
fcis-28893	220	53	:	:	PUNCT
fcis-28893	220	54	ieee	ieee	NOUN
fcis-28893	220	55	,	,	PUNCT
fcis-28893	220	56	pp	pp	ADJ
fcis-28893	220	57	.	.	PUNCT
fcis-28893	221	1	5187	5187	NUM
fcis-28893	221	2	-	-	SYM
fcis-28893	221	3	5196	5196	NUM
fcis-28893	221	4	,	,	PUNCT
fcis-28893	221	5	doi	doi	NOUN
fcis-28893	221	6	:	:	PUNCT
fcis-28893	221	7	10.1109	10.1109	NUM
fcis-28893	221	8	/	/	SYM
fcis-28893	221	9	cvpr.2017.551	cvpr.2017.551	PROPN
fcis-28893	221	10	.	.	PUNCT
fcis-28893	222	1	127	127	NUM
fcis-28893	223	1	[	[	X
fcis-28893	223	2	2	2	NUM
fcis-28893	223	3	]	]	PUNCT
fcis-28893	223	4	z.	z.	PROPN
fcis-28893	223	5	ding	ding	PROPN
fcis-28893	223	6	,	,	PUNCT
fcis-28893	223	7	c.	c.	PROPN
fcis-28893	223	8	ding	ding	PROPN
fcis-28893	223	9	,	,	PUNCT
fcis-28893	223	10	z.	z.	PROPN
fcis-28893	223	11	shao	shao	PROPN
fcis-28893	223	12	,	,	PUNCT
fcis-28893	223	13	and	and	CCONJ
fcis-28893	223	14	d.	d.	PROPN
fcis-28893	223	15	tao	tao	PROPN
fcis-28893	223	16	,	,	PUNCT
fcis-28893	223	17	"	"	PUNCT
fcis-28893	223	18	semantically	semantically	ADV
fcis-28893	223	19	selfaligned	selfaligne	VERB
fcis-28893	223	20	network	network	NOUN
fcis-28893	223	21	for	for	ADP
fcis-28893	223	22	text	text	NOUN
fcis-28893	223	23	-	-	PUNCT
fcis-28893	223	24	to	to	ADP
fcis-28893	223	25	-	-	PUNCT
fcis-28893	223	26	image	image	NOUN
fcis-28893	223	27	part	part	NOUN
fcis-28893	223	28	-	-	PUNCT
fcis-28893	223	29	aware	aware	ADJ
fcis-28893	223	30	person	person	NOUN
fcis-28893	223	31	reidentification	reidentification	NOUN
fcis-28893	223	32	,	,	PUNCT
fcis-28893	223	33	"	"	PUNCT
fcis-28893	223	34	p.	p.	NOUN
fcis-28893	223	35	arxiv:2107.12666doi	arxiv:2107.12666doi	PROPN
fcis-28893	223	36	:	:	PUNCT
fcis-28893	223	37	10.48550	10.48550	NUM
fcis-28893	223	38	/	/	SYM
fcis-28893	223	39	arxiv	arxiv	PROPN
fcis-28893	223	40	.	.	PUNCT
fcis-28893	224	1	2107	2107	NUM
fcis-28893	224	2	.	.	PUNCT
fcis-28893	225	1	12666	12666	NUM
fcis-28893	225	2	.	.	PUNCT
fcis-28893	226	1	[	[	X
fcis-28893	226	2	3	3	X
fcis-28893	226	3	]	]	PUNCT
fcis-28893	226	4	a.	a.	NOUN
fcis-28893	226	5	zhu	zhu	PROPN
fcis-28893	226	6	et	et	PROPN
fcis-28893	226	7	al	al	PROPN
fcis-28893	226	8	.	.	PROPN
fcis-28893	226	9	,	,	PUNCT
fcis-28893	226	10	"	"	PUNCT
fcis-28893	226	11	dssl	dssl	PROPN
fcis-28893	226	12	:	:	PUNCT
fcis-28893	226	13	deep	deep	ADJ
fcis-28893	226	14	surroundings	surroundings	ADJ
fcis-28893	226	15	-	-	PUNCT
fcis-28893	226	16	person	person	NOUN
fcis-28893	226	17	separation	separation	NOUN
fcis-28893	226	18	learning	learn	VERB
fcis-28893	226	19	for	for	ADP
fcis-28893	226	20	text	text	NOUN
fcis-28893	226	21	-	-	PUNCT
fcis-28893	226	22	based	base	VERB
fcis-28893	226	23	person	person	NOUN
fcis-28893	226	24	retrieval	retrieval	NOUN
fcis-28893	226	25	,	,	PUNCT
fcis-28893	226	26	"	"	PUNCT
fcis-28893	226	27	p.	p.	NOUN
fcis-28893	226	28	arxiv	arxiv	PROPN
fcis-28893	226	29	:	:	PUNCT
fcis-28893	226	30	2109	2109	NUM
fcis-28893	226	31	.	.	PUNCT
fcis-28893	227	1	05534	05534	NUM
fcis-28893	227	2	doi	doi	NOUN
fcis-28893	227	3	:	:	PUNCT
fcis-28893	227	4	10.48550	10.48550	NUM
fcis-28893	227	5	/	/	SYM
fcis-28893	227	6	arxiv.2109.05534	arxiv.2109.05534	NOUN
fcis-28893	227	7	.	.	PUNCT
fcis-28893	228	1	[	[	X
fcis-28893	228	2	4	4	X
fcis-28893	228	3	]	]	PUNCT
fcis-28893	228	4	s.	s.	PROPN
fcis-28893	228	5	hochreiter	hochreiter	PROPN
fcis-28893	228	6	and	and	CCONJ
fcis-28893	228	7	j.	j.	PROPN
fcis-28893	228	8	schmidhuber	schmidhuber	PROPN
fcis-28893	228	9	,	,	PUNCT
fcis-28893	228	10	"	"	PUNCT
fcis-28893	228	11	long	long	ADJ
fcis-28893	228	12	short	short	ADJ
fcis-28893	228	13	-	-	PUNCT
fcis-28893	228	14	term	term	NOUN
fcis-28893	228	15	memory	memory	NOUN
fcis-28893	228	16	,	,	PUNCT
fcis-28893	228	17	"	"	PUNCT
fcis-28893	228	18	neural	neural	ADJ
fcis-28893	228	19	computation	computation	NOUN
fcis-28893	228	20	,	,	PUNCT
fcis-28893	228	21	vol	vol	NOUN
fcis-28893	228	22	.	.	PROPN
fcis-28893	228	23	9	9	NUM
fcis-28893	228	24	,	,	PUNCT
fcis-28893	228	25	no	no	INTJ
fcis-28893	228	26	.	.	NOUN
fcis-28893	228	27	8	8	NUM
fcis-28893	228	28	,	,	PUNCT
fcis-28893	228	29	pp	pp	ADJ
fcis-28893	228	30	.	.	PUNCT
fcis-28893	229	1	1735	1735	NUM
fcis-28893	229	2	-	-	SYM
fcis-28893	229	3	1780	1780	NUM
fcis-28893	229	4	,	,	PUNCT
fcis-28893	229	5	1997	1997	NUM
fcis-28893	229	6	,	,	PUNCT
fcis-28893	229	7	doi	doi	NOUN
fcis-28893	229	8	:	:	PUNCT
fcis-28893	229	9	10.1162	10.1162	NUM
fcis-28893	229	10	/	/	SYM
fcis-28893	229	11	neco.1997.9.8.1735	neco.1997.9.8.1735	NOUN
fcis-28893	229	12	.	.	PUNCT
fcis-28893	230	1	[	[	X
fcis-28893	230	2	5	5	X
fcis-28893	230	3	]	]	PUNCT
fcis-28893	230	4	s.	s.	PROPN
fcis-28893	230	5	li	li	PROPN
fcis-28893	230	6	,	,	PUNCT
fcis-28893	230	7	t.	t.	PROPN
fcis-28893	230	8	xiao	xiao	PROPN
fcis-28893	230	9	,	,	PUNCT
fcis-28893	230	10	h.	h.	PROPN
fcis-28893	230	11	li	li	PROPN
fcis-28893	230	12	,	,	PUNCT
fcis-28893	230	13	w.	w.	PROPN
fcis-28893	230	14	yang	yang	PROPN
fcis-28893	230	15	,	,	PUNCT
fcis-28893	230	16	and	and	CCONJ
fcis-28893	230	17	x.	x.	PROPN
fcis-28893	230	18	wang	wang	PROPN
fcis-28893	230	19	,	,	PUNCT
fcis-28893	230	20	"	"	PUNCT
fcis-28893	230	21	identity	identity	NOUN
fcis-28893	230	22	-	-	PUNCT
fcis-28893	230	23	aware	aware	ADJ
fcis-28893	230	24	textual	textual	ADJ
fcis-28893	230	25	-	-	PUNCT
fcis-28893	230	26	visual	visual	ADJ
fcis-28893	230	27	matching	matching	NOUN
fcis-28893	230	28	with	with	ADP
fcis-28893	230	29	latent	latent	NOUN
fcis-28893	230	30	co	co	NOUN
fcis-28893	230	31	-	-	NOUN
fcis-28893	230	32	attention	attention	NOUN
fcis-28893	230	33	,	,	PUNCT
fcis-28893	230	34	"	"	PUNCT
fcis-28893	230	35	p.	p.	NOUN
fcis-28893	230	36	arxiv:1708.01988doi	arxiv:1708.01988doi	NOUN
fcis-28893	230	37	:	:	PUNCT
fcis-28893	230	38	10.48550	10.48550	NUM
fcis-28893	230	39	/	/	SYM
fcis-28893	230	40	arxiv.1708.01988	arxiv.1708.01988	NOUN
fcis-28893	230	41	.	.	PUNCT
fcis-28893	231	1	[	[	X
fcis-28893	231	2	6	6	NUM
fcis-28893	231	3	]	]	PUNCT
fcis-28893	231	4	t.	t.	PROPN
fcis-28893	231	5	chen	chen	PROPN
fcis-28893	231	6	,	,	PUNCT
fcis-28893	231	7	c.	c.	PROPN
fcis-28893	231	8	xu	xu	PROPN
fcis-28893	231	9	,	,	PUNCT
fcis-28893	231	10	and	and	CCONJ
fcis-28893	231	11	j.	j.	PROPN
fcis-28893	231	12	luo	luo	PROPN
fcis-28893	231	13	,	,	PUNCT
fcis-28893	231	14	"	"	PUNCT
fcis-28893	231	15	improving	improve	VERB
fcis-28893	231	16	text	text	NOUN
fcis-28893	231	17	-	-	PUNCT
fcis-28893	231	18	based	base	VERB
fcis-28893	231	19	person	person	NOUN
fcis-28893	231	20	search	search	NOUN
fcis-28893	231	21	by	by	ADP
fcis-28893	231	22	spatial	spatial	ADJ
fcis-28893	231	23	matching	matching	NOUN
fcis-28893	231	24	and	and	CCONJ
fcis-28893	231	25	adaptive	adaptive	ADJ
fcis-28893	231	26	threshold	threshold	NOUN
fcis-28893	231	27	,	,	PUNCT
fcis-28893	231	28	"	"	PUNCT
fcis-28893	231	29	in	in	ADP
fcis-28893	231	30	2018	2018	NUM
fcis-28893	231	31	ieee	ieee	NOUN
fcis-28893	231	32	winter	winter	NOUN
fcis-28893	231	33	conference	conference	NOUN
fcis-28893	231	34	on	on	ADP
fcis-28893	231	35	applications	application	NOUN
fcis-28893	231	36	of	of	ADP
fcis-28893	231	37	computer	computer	NOUN
fcis-28893	231	38	vision	vision	NOUN
fcis-28893	231	39	(	(	PUNCT
fcis-28893	231	40	wacv	wacv	NOUN
fcis-28893	231	41	)	)	PUNCT
fcis-28893	231	42	,	,	PUNCT
fcis-28893	231	43	12	12	NUM
fcis-28893	231	44	-	-	SYM
fcis-28893	231	45	15	15	NUM
fcis-28893	231	46	march	march	NOUN
fcis-28893	231	47	2018	2018	NUM
fcis-28893	231	48	2018	2018	NUM
fcis-28893	231	49	:	:	PUNCT
fcis-28893	231	50	ieee	ieee	NOUN
fcis-28893	231	51	,	,	PUNCT
fcis-28893	231	52	pp	pp	ADJ
fcis-28893	231	53	.	.	PUNCT
fcis-28893	231	54	1879	1879	NUM
fcis-28893	231	55	-	-	SYM
fcis-28893	231	56	1887	1887	NUM
fcis-28893	231	57	,	,	PUNCT
fcis-28893	231	58	doi	doi	NOUN
fcis-28893	231	59	:	:	PUNCT
fcis-28893	231	60	10.1109	10.1109	NUM
fcis-28893	231	61	/	/	SYM
fcis-28893	231	62	wacv.2018.00208	wacv.2018.00208	NOUN
fcis-28893	231	63	.	.	PUNCT
fcis-28893	232	1	[	[	X
fcis-28893	232	2	7	7	X
fcis-28893	232	3	]	]	X
fcis-28893	232	4	y.	y.	PROPN
fcis-28893	232	5	zhang	zhang	PROPN
fcis-28893	232	6	and	and	CCONJ
fcis-28893	232	7	h.	h.	PROPN
fcis-28893	232	8	lu	lu	PROPN
fcis-28893	232	9	,	,	PUNCT
fcis-28893	232	10	"	"	PUNCT
fcis-28893	232	11	deep	deep	ADJ
fcis-28893	232	12	cross	cross	ADJ
fcis-28893	232	13	-	-	ADJ
fcis-28893	232	14	modal	modal	ADJ
fcis-28893	232	15	projection	projection	NOUN
fcis-28893	232	16	learning	learning	NOUN
fcis-28893	232	17	for	for	ADP
fcis-28893	232	18	image	image	NOUN
fcis-28893	232	19	-	-	PUNCT
fcis-28893	232	20	text	text	NOUN
fcis-28893	232	21	matching	matching	NOUN
fcis-28893	232	22	,	,	PUNCT
fcis-28893	232	23	"	"	PUNCT
fcis-28893	232	24	in	in	ADP
fcis-28893	232	25	european	european	ADJ
fcis-28893	232	26	conference	conference	NOUN
fcis-28893	232	27	on	on	ADP
fcis-28893	232	28	computer	computer	NOUN
fcis-28893	232	29	vision	vision	NOUN
fcis-28893	232	30	,	,	PUNCT
fcis-28893	232	31	cham	cham	NOUN
fcis-28893	232	32	,	,	PUNCT
fcis-28893	232	33	2018	2018	NUM
fcis-28893	232	34	:	:	PUNCT
fcis-28893	232	35	springer	springer	NOUN
fcis-28893	232	36	international	international	ADJ
fcis-28893	232	37	publishing	publishing	NOUN
fcis-28893	232	38	,	,	PUNCT
fcis-28893	232	39	in	in	ADP
fcis-28893	232	40	computer	computer	NOUN
fcis-28893	232	41	vision	vision	NOUN
fcis-28893	232	42	–	–	PUNCT
fcis-28893	232	43	eccv	eccv	ADV
fcis-28893	232	44	2018	2018	NUM
fcis-28893	232	45	,	,	PUNCT
fcis-28893	232	46	pp	pp	ADJ
fcis-28893	232	47	.	.	PUNCT
fcis-28893	233	1	707	707	NUM
fcis-28893	233	2	-	-	SYM
fcis-28893	233	3	723	723	NUM
fcis-28893	233	4	,	,	PUNCT
fcis-28893	233	5	doi	doi	NOUN
fcis-28893	233	6	:	:	PUNCT
fcis-28893	233	7	10.1007/978	10.1007/978	NUM
fcis-28893	233	8	-	-	SYM
fcis-28893	233	9	3	3	NUM
fcis-28893	233	10	-	-	PUNCT
fcis-28893	233	11	030	030	NUM
fcis-28893	233	12	-	-	PUNCT
fcis-28893	233	13	01246	01246	NUM
fcis-28893	233	14	-	-	PUNCT
fcis-28893	233	15	5_42	5_42	NUM
fcis-28893	233	16	.	.	PUNCT
fcis-28893	234	1	[	[	X
fcis-28893	234	2	8	8	NUM
fcis-28893	234	3	]	]	X
fcis-28893	234	4	y.	y.	PROPN
fcis-28893	234	5	jing	jing	PROPN
fcis-28893	234	6	,	,	PUNCT
fcis-28893	234	7	c.	c.	PROPN
fcis-28893	234	8	si	si	PROPN
fcis-28893	234	9	,	,	PUNCT
fcis-28893	234	10	j.	j.	PROPN
fcis-28893	234	11	wang	wang	PROPN
fcis-28893	234	12	,	,	PUNCT
fcis-28893	234	13	w.	w.	PROPN
fcis-28893	234	14	wang	wang	PROPN
fcis-28893	234	15	,	,	PUNCT
fcis-28893	234	16	l.	l.	PROPN
fcis-28893	234	17	wang	wang	PROPN
fcis-28893	234	18	,	,	PUNCT
fcis-28893	234	19	and	and	CCONJ
fcis-28893	234	20	t.	t.	PROPN
fcis-28893	234	21	tan	tan	PROPN
fcis-28893	234	22	,	,	PUNCT
fcis-28893	234	23	"	"	PUNCT
fcis-28893	234	24	poseguided	poseguide	VERB
fcis-28893	234	25	multi	multi	ADJ
fcis-28893	234	26	-	-	ADJ
fcis-28893	234	27	granularity	granularity	NOUN
fcis-28893	234	28	attention	attention	NOUN
fcis-28893	234	29	network	network	NOUN
fcis-28893	234	30	for	for	ADP
fcis-28893	234	31	text	text	NOUN
fcis-28893	234	32	-	-	PUNCT
fcis-28893	234	33	based	base	VERB
fcis-28893	234	34	person	person	NOUN
fcis-28893	234	35	search	search	NOUN
fcis-28893	234	36	,	,	PUNCT
fcis-28893	234	37	"	"	PUNCT
fcis-28893	234	38	proceedings	proceeding	NOUN
fcis-28893	234	39	of	of	ADP
fcis-28893	234	40	the	the	DET
fcis-28893	234	41	aaai	aaai	PROPN
fcis-28893	234	42	conference	conference	NOUN
fcis-28893	234	43	on	on	ADP
fcis-28893	234	44	artificial	artificial	ADJ
fcis-28893	234	45	intelligence	intelligence	NOUN
fcis-28893	234	46	,	,	PUNCT
fcis-28893	234	47	vol	vol	NOUN
fcis-28893	234	48	.	.	PROPN
fcis-28893	235	1	34	34	NUM
fcis-28893	235	2	,	,	PUNCT
fcis-28893	235	3	no	no	INTJ
fcis-28893	235	4	.	.	NOUN
fcis-28893	235	5	07	07	NUM
fcis-28893	235	6	,	,	PUNCT
fcis-28893	235	7	pp	pp	ADJ
fcis-28893	235	8	.	.	PUNCT
fcis-28893	236	1	11189	11189	NUM
fcis-28893	236	2	-	-	SYM
fcis-28893	236	3	11196	11196	NUM
fcis-28893	236	4	,	,	PUNCT
fcis-28893	236	5	04/03	04/03	NUM
fcis-28893	236	6	2020	2020	NUM
fcis-28893	236	7	,	,	PUNCT
fcis-28893	236	8	doi	doi	NOUN
fcis-28893	236	9	:	:	PUNCT
fcis-28893	236	10	10.1609	10.1609	NUM
fcis-28893	236	11	/	/	SYM
fcis-28893	236	12	aaai	aaai	PROPN
fcis-28893	236	13	.	.	PUNCT
fcis-28893	237	1	v34i07.6777	v34i07.6777	PROPN
fcis-28893	237	2	.	.	PUNCT
fcis-28893	238	1	[	[	X
fcis-28893	238	2	9	9	NUM
fcis-28893	238	3	]	]	PUNCT
fcis-28893	238	4	k.	k.	PROPN
fcis-28893	238	5	niu	niu	PROPN
fcis-28893	238	6	,	,	PUNCT
fcis-28893	238	7	y.	y.	PROPN
fcis-28893	238	8	huang	huang	PROPN
fcis-28893	238	9	,	,	PUNCT
fcis-28893	238	10	w.	w.	PROPN
fcis-28893	238	11	ouyang	ouyang	PROPN
fcis-28893	238	12	,	,	PUNCT
fcis-28893	238	13	and	and	CCONJ
fcis-28893	238	14	l.	l.	PROPN
fcis-28893	238	15	wang	wang	PROPN
fcis-28893	238	16	,	,	PUNCT
fcis-28893	238	17	"	"	PUNCT
fcis-28893	238	18	improving	improve	VERB
fcis-28893	238	19	description	description	NOUN
fcis-28893	238	20	-	-	PUNCT
fcis-28893	238	21	based	base	VERB
fcis-28893	238	22	person	person	NOUN
fcis-28893	238	23	re	re	NOUN
fcis-28893	238	24	-	-	NOUN
fcis-28893	238	25	identification	identification	NOUN
fcis-28893	238	26	by	by	ADP
fcis-28893	238	27	multigranularity	multigranularity	NOUN
fcis-28893	238	28	image	image	NOUN
fcis-28893	238	29	-	-	PUNCT
fcis-28893	238	30	text	text	NOUN
fcis-28893	238	31	alignments	alignment	NOUN
fcis-28893	238	32	,	,	PUNCT
fcis-28893	238	33	"	"	PUNCT
fcis-28893	238	34	p.	p.	NOUN
fcis-28893	238	35	arxiv:1906.09610doi	arxiv:1906.09610doi	NOUN
fcis-28893	238	36	:	:	PUNCT
fcis-28893	238	37	10	10	NUM
fcis-28893	238	38	.	.	X
fcis-28893	238	39	48550	48550	NUM
fcis-28893	238	40	/	/	SYM
fcis-28893	238	41	arxiv.1906.09610	arxiv.1906.09610	NUM
fcis-28893	238	42	.	.	PUNCT
fcis-28893	239	1	[	[	X
fcis-28893	239	2	10	10	NUM
fcis-28893	239	3	]	]	X
fcis-28893	239	4	y.	y.	PROPN
fcis-28893	239	5	sun	sun	PROPN
fcis-28893	239	6	,	,	PUNCT
fcis-28893	239	7	l.	l.	PROPN
fcis-28893	239	8	zheng	zheng	PROPN
fcis-28893	239	9	,	,	PUNCT
fcis-28893	239	10	y.	y.	PROPN
fcis-28893	239	11	yang	yang	PROPN
fcis-28893	239	12	,	,	PUNCT
fcis-28893	239	13	q.	q.	PROPN
fcis-28893	239	14	tian	tian	PROPN
fcis-28893	239	15	,	,	PUNCT
fcis-28893	239	16	and	and	CCONJ
fcis-28893	239	17	s.	s.	PROPN
fcis-28893	239	18	wang	wang	PROPN
fcis-28893	239	19	,	,	PUNCT
fcis-28893	239	20	"	"	PUNCT
fcis-28893	239	21	beyond	beyond	ADP
fcis-28893	239	22	part	part	NOUN
fcis-28893	239	23	models	model	NOUN
fcis-28893	239	24	:	:	PUNCT
fcis-28893	239	25	person	person	NOUN
fcis-28893	239	26	retrieval	retrieval	NOUN
fcis-28893	239	27	with	with	ADP
fcis-28893	239	28	refined	refined	ADJ
fcis-28893	239	29	part	part	NOUN
fcis-28893	239	30	pooling	pool	VERB
fcis-28893	239	31	(	(	PUNCT
fcis-28893	239	32	and	and	CCONJ
fcis-28893	239	33	a	a	DET
fcis-28893	239	34	strong	strong	ADJ
fcis-28893	239	35	convolutional	convolutional	ADJ
fcis-28893	239	36	baseline	baseline	NOUN
fcis-28893	239	37	)	)	PUNCT
fcis-28893	239	38	,	,	PUNCT
fcis-28893	239	39	"	"	PUNCT
fcis-28893	239	40	p.	p.	NOUN
fcis-28893	239	41	arxiv:1711.09349doi	arxiv:1711.09349doi	NOUN
fcis-28893	239	42	:	:	PUNCT
fcis-28893	239	43	10.48550	10.48550	NUM
fcis-28893	239	44	/	/	SYM
fcis-28893	239	45	arxiv.1711.09349	arxiv.1711.09349	NOUN
fcis-28893	239	46	.	.	PUNCT
fcis-28893	240	1	[	[	X
fcis-28893	240	2	11	11	NUM
fcis-28893	240	3	]	]	PUNCT
fcis-28893	240	4	z.	z.	PROPN
fcis-28893	240	5	wang	wang	PROPN
fcis-28893	240	6	,	,	PUNCT
fcis-28893	240	7	z.	z.	PROPN
fcis-28893	240	8	fang	fang	PROPN
fcis-28893	240	9	,	,	PUNCT
fcis-28893	240	10	j.	j.	PROPN
fcis-28893	240	11	wang	wang	PROPN
fcis-28893	240	12	,	,	PUNCT
fcis-28893	240	13	and	and	CCONJ
fcis-28893	240	14	y.	y.	PROPN
fcis-28893	240	15	yang	yang	PROPN
fcis-28893	240	16	,	,	PUNCT
fcis-28893	240	17	"	"	PUNCT
fcis-28893	240	18	vitaa	vitaa	ADJ
fcis-28893	240	19	:	:	PUNCT
fcis-28893	240	20	visualtextual	visualtextual	ADJ
fcis-28893	240	21	attributes	attribute	NOUN
fcis-28893	240	22	alignment	alignment	NOUN
fcis-28893	240	23	in	in	ADP
fcis-28893	240	24	person	person	NOUN
fcis-28893	240	25	search	search	NOUN
fcis-28893	240	26	by	by	ADP
fcis-28893	240	27	natural	natural	ADJ
fcis-28893	240	28	language	language	NOUN
fcis-28893	240	29	,	,	PUNCT
fcis-28893	240	30	"	"	PUNCT
fcis-28893	240	31	p.	p.	NOUN
fcis-28893	240	32	arxiv:2005.07327doi	arxiv:2005.07327doi	NOUN
fcis-28893	240	33	:	:	PUNCT
fcis-28893	240	34	10.48550	10.48550	NUM
fcis-28893	240	35	/	/	SYM
fcis-28893	240	36	arxiv	arxiv	PROPN
fcis-28893	240	37	.	.	PUNCT
fcis-28893	241	1	2005	2005	NUM
fcis-28893	241	2	.	.	PUNCT
fcis-28893	242	1	07327	07327	NUM
fcis-28893	242	2	.	.	PUNCT
fcis-28893	243	1	[	[	X
fcis-28893	243	2	12	12	NUM
fcis-28893	243	3	]	]	PUNCT
fcis-28893	243	4	j.	j.	PROPN
fcis-28893	243	5	liu	liu	PROPN
fcis-28893	243	6	,	,	PUNCT
fcis-28893	243	7	z.-j	z.-j	PROPN
fcis-28893	243	8	.	.	PUNCT
fcis-28893	243	9	zha	zha	PROPN
fcis-28893	243	10	,	,	PUNCT
fcis-28893	243	11	r.	r.	PROPN
fcis-28893	243	12	hong	hong	PROPN
fcis-28893	243	13	,	,	PUNCT
fcis-28893	243	14	m.	m.	PROPN
fcis-28893	243	15	wang	wang	PROPN
fcis-28893	243	16	,	,	PUNCT
fcis-28893	243	17	and	and	CCONJ
fcis-28893	243	18	y.	y.	PROPN
fcis-28893	243	19	zhang	zhang	PROPN
fcis-28893	243	20	,	,	PUNCT
fcis-28893	243	21	"	"	PUNCT
fcis-28893	243	22	deep	deep	ADJ
fcis-28893	243	23	adversarial	adversarial	ADJ
fcis-28893	243	24	graph	graph	NOUN
fcis-28893	243	25	attention	attention	NOUN
fcis-28893	243	26	convolution	convolution	NOUN
fcis-28893	243	27	network	network	NOUN
fcis-28893	243	28	for	for	ADP
fcis-28893	243	29	textbased	textbase	VERB
fcis-28893	243	30	person	person	NOUN
fcis-28893	243	31	search	search	NOUN
fcis-28893	243	32	,	,	PUNCT
fcis-28893	243	33	"	"	PUNCT
fcis-28893	243	34	presented	present	VERB
fcis-28893	243	35	at	at	ADP
fcis-28893	243	36	the	the	DET
fcis-28893	243	37	proceedings	proceeding	NOUN
fcis-28893	243	38	of	of	ADP
fcis-28893	243	39	the	the	DET
fcis-28893	243	40	27th	27th	ADJ
fcis-28893	243	41	acm	acm	PROPN
fcis-28893	243	42	international	international	ADJ
fcis-28893	243	43	conference	conference	NOUN
fcis-28893	243	44	on	on	ADP
fcis-28893	243	45	multimedia	multimedia	NOUN
fcis-28893	243	46	,	,	PUNCT
fcis-28893	243	47	nice	nice	ADJ
fcis-28893	243	48	,	,	PUNCT
fcis-28893	243	49	france	france	PROPN
fcis-28893	243	50	,	,	PUNCT
fcis-28893	243	51	2019	2019	NUM
fcis-28893	243	52	.	.	PUNCT
fcis-28893	244	1	[	[	X
fcis-28893	244	2	online	online	X
fcis-28893	244	3	]	]	X
fcis-28893	244	4	.	.	PUNCT
fcis-28893	245	1	available	available	ADJ
fcis-28893	245	2	:	:	PUNCT
fcis-28893	246	1	https://doi.org/10	https://doi.org/10	PROPN
fcis-28893	246	2	.	.	PUNCT
fcis-28893	247	1	1145/	1145/	NUM
fcis-28893	247	2	3343031	3343031	NUM
fcis-28893	247	3	.	.	PUNCT
fcis-28893	248	1	3350991	3350991	NUM
fcis-28893	248	2	.	.	PUNCT
fcis-28893	249	1	[	[	X
fcis-28893	249	2	13	13	NUM
fcis-28893	249	3	]	]	X
fcis-28893	249	4	n.	n.	NOUN
fcis-28893	249	5	sarafianos	sarafianos	PROPN
fcis-28893	249	6	,	,	PUNCT
fcis-28893	249	7	x.	x.	PROPN
fcis-28893	249	8	xu	xu	PROPN
fcis-28893	249	9	,	,	PUNCT
fcis-28893	249	10	and	and	CCONJ
fcis-28893	249	11	i.	i.	PROPN
fcis-28893	249	12	a.	a.	PROPN
fcis-28893	249	13	kakadiaris	kakadiaris	PROPN
fcis-28893	249	14	,	,	PUNCT
fcis-28893	249	15	"	"	PUNCT
fcis-28893	249	16	adversarial	adversarial	ADJ
fcis-28893	249	17	representation	representation	NOUN
fcis-28893	249	18	learning	learn	VERB
fcis-28893	249	19	for	for	ADP
fcis-28893	249	20	text	text	NOUN
fcis-28893	249	21	-	-	PUNCT
fcis-28893	249	22	to	to	ADP
fcis-28893	249	23	-	-	PUNCT
fcis-28893	249	24	image	image	NOUN
fcis-28893	249	25	matching	matching	NOUN
fcis-28893	249	26	,	,	PUNCT
fcis-28893	249	27	"	"	PUNCT
fcis-28893	249	28	p.	p.	NOUN
fcis-28893	249	29	arxiv:1908.10534doi	arxiv:1908.10534doi	NOUN
fcis-28893	249	30	:	:	PUNCT
fcis-28893	249	31	10.48550	10.48550	NUM
fcis-28893	249	32	/	/	SYM
fcis-28893	249	33	arxiv.1908.10534	arxiv.1908.10534	NUM
fcis-28893	249	34	.	.	PUNCT
fcis-28893	250	1	[	[	X
fcis-28893	250	2	14	14	NUM
fcis-28893	250	3	]	]	PUNCT
fcis-28893	250	4	j.	j.	PROPN
fcis-28893	250	5	devlin	devlin	PROPN
fcis-28893	250	6	,	,	PUNCT
fcis-28893	250	7	m.-w	m.-w	PROPN
fcis-28893	250	8	.	.	PUNCT
fcis-28893	251	1	chang	chang	PROPN
fcis-28893	251	2	,	,	PUNCT
fcis-28893	251	3	k.	k.	PROPN
fcis-28893	251	4	lee	lee	PROPN
fcis-28893	251	5	,	,	PUNCT
fcis-28893	251	6	and	and	CCONJ
fcis-28893	251	7	k.	k.	PROPN
fcis-28893	251	8	toutanova	toutanova	PROPN
fcis-28893	251	9	,	,	PUNCT
fcis-28893	251	10	"	"	PUNCT
fcis-28893	251	11	bert	bert	NOUN
fcis-28893	251	12	:	:	PUNCT
fcis-28893	251	13	pre	pre	ADJ
fcis-28893	251	14	-	-	NOUN
fcis-28893	251	15	training	training	NOUN
fcis-28893	251	16	of	of	ADP
fcis-28893	251	17	deep	deep	ADJ
fcis-28893	251	18	bidirectional	bidirectional	ADJ
fcis-28893	251	19	transformers	transformer	NOUN
fcis-28893	251	20	for	for	ADP
fcis-28893	251	21	language	language	NOUN
fcis-28893	251	22	understanding	understanding	NOUN
fcis-28893	251	23	,	,	PUNCT
fcis-28893	251	24	"	"	PUNCT
fcis-28893	251	25	p.	p.	NOUN
fcis-28893	251	26	arxiv:1810.04805doi	arxiv:1810.04805doi	NOUN
fcis-28893	251	27	:	:	PUNCT
fcis-28893	251	28	10.48550	10.48550	NUM
fcis-28893	251	29	/	/	SYM
fcis-28893	251	30	arxiv	arxiv	PROPN
fcis-28893	251	31	.	.	PUNCT
fcis-28893	252	1	1810	1810	NUM
fcis-28893	252	2	.	.	PUNCT
fcis-28893	253	1	04805	04805	NUM
fcis-28893	253	2	.	.	PUNCT
fcis-28893	254	1	[	[	X
fcis-28893	254	2	15	15	NUM
fcis-28893	254	3	]	]	X
fcis-28893	254	4	s.	s.	PROPN
fcis-28893	254	5	yan	yan	PROPN
fcis-28893	254	6	,	,	PUNCT
fcis-28893	254	7	n.	n.	PROPN
fcis-28893	254	8	dong	dong	PROPN
fcis-28893	254	9	,	,	PUNCT
fcis-28893	254	10	l.	l.	PROPN
fcis-28893	254	11	zhang	zhang	PROPN
fcis-28893	254	12	,	,	PUNCT
fcis-28893	254	13	and	and	CCONJ
fcis-28893	254	14	j.	j.	PROPN
fcis-28893	254	15	tang	tang	PROPN
fcis-28893	254	16	,	,	PUNCT
fcis-28893	254	17	"	"	PUNCT
fcis-28893	254	18	clip	clip	NOUN
fcis-28893	254	19	-	-	PUNCT
fcis-28893	254	20	driven	drive	VERB
fcis-28893	254	21	finegrained	finegraine	VERB
fcis-28893	254	22	text	text	NOUN
fcis-28893	254	23	-	-	PUNCT
fcis-28893	254	24	image	image	NOUN
fcis-28893	254	25	person	person	NOUN
fcis-28893	254	26	re	re	NOUN
fcis-28893	254	27	-	-	NOUN
fcis-28893	254	28	identification	identification	NOUN
fcis-28893	254	29	,	,	PUNCT
fcis-28893	254	30	"	"	PUNCT
fcis-28893	254	31	p.	p.	NOUN
fcis-28893	254	32	arxiv:2210	arxiv:2210	NOUN
fcis-28893	254	33	.	.	PUNCT
fcis-28893	255	1	10276doi	10276doi	NUM
fcis-28893	255	2	:	:	PUNCT
fcis-28893	255	3	10.48550	10.48550	NUM
fcis-28893	255	4	/	/	SYM
fcis-28893	255	5	arxiv.2210.10276	arxiv.2210.10276	NOUN
fcis-28893	255	6	.	.	PUNCT
fcis-28893	256	1	[	[	X
fcis-28893	256	2	16	16	NUM
fcis-28893	256	3	]	]	PUNCT
fcis-28893	256	4	a.	a.	PROPN
fcis-28893	256	5	radford	radford	PROPN
fcis-28893	256	6	et	et	PROPN
fcis-28893	256	7	al	al	PROPN
fcis-28893	256	8	.	.	PROPN
fcis-28893	256	9	,	,	PUNCT
fcis-28893	256	10	"	"	PUNCT
fcis-28893	256	11	learning	learn	VERB
fcis-28893	256	12	transferable	transferable	ADJ
fcis-28893	256	13	visual	visual	ADJ
fcis-28893	256	14	models	model	NOUN
fcis-28893	256	15	from	from	ADP
fcis-28893	256	16	natural	natural	ADJ
fcis-28893	256	17	language	language	NOUN
fcis-28893	256	18	supervision	supervision	NOUN
fcis-28893	256	19	,	,	PUNCT
fcis-28893	256	20	"	"	PUNCT
fcis-28893	256	21	p.	p.	NOUN
fcis-28893	256	22	arxiv:2103.00020doi	arxiv:2103.00020doi	NOUN
fcis-28893	256	23	:	:	PUNCT
fcis-28893	256	24	10	10	NUM
fcis-28893	256	25	.	.	X
fcis-28893	256	26	48550/	48550/	NUM
fcis-28893	256	27	arxiv.2103.00020	arxiv.2103.00020	NOUN
fcis-28893	256	28	.	.	PUNCT
fcis-28893	257	1	[	[	X
fcis-28893	257	2	17	17	NUM
fcis-28893	257	3	]	]	X
fcis-28893	257	4	d.	d.	PROPN
fcis-28893	257	5	jiang	jiang	PROPN
fcis-28893	257	6	and	and	CCONJ
fcis-28893	257	7	m.	m.	NOUN
fcis-28893	257	8	ye	ye	PROPN
fcis-28893	257	9	,	,	PUNCT
fcis-28893	257	10	"	"	PUNCT
fcis-28893	257	11	cross	cross	ADJ
fcis-28893	257	12	-	-	ADJ
fcis-28893	257	13	modal	modal	ADJ
fcis-28893	257	14	implicit	implicit	ADJ
fcis-28893	257	15	relation	relation	NOUN
fcis-28893	257	16	reasoning	reasoning	NOUN
fcis-28893	257	17	and	and	CCONJ
fcis-28893	257	18	aligning	align	VERB
fcis-28893	257	19	for	for	ADP
fcis-28893	257	20	text	text	NOUN
fcis-28893	257	21	-	-	PUNCT
fcis-28893	257	22	to	to	ADP
fcis-28893	257	23	-	-	PUNCT
fcis-28893	257	24	image	image	NOUN
fcis-28893	257	25	person	person	NOUN
fcis-28893	257	26	retrieval	retrieval	NOUN
fcis-28893	257	27	,	,	PUNCT
fcis-28893	257	28	"	"	PUNCT
fcis-28893	257	29	p.	p.	NOUN
fcis-28893	257	30	arxiv	arxiv	PROPN
fcis-28893	257	31	:	:	PUNCT
fcis-28893	257	32	2303.12501doi	2303.12501doi	NUM
fcis-28893	257	33	:	:	PUNCT
fcis-28893	257	34	10.48550	10.48550	NUM
fcis-28893	257	35	/	/	SYM
fcis-28893	257	36	arxiv.2303.12501	arxiv.2303.12501	PROPN
fcis-28893	257	37	.	.	PUNCT
fcis-28893	258	1	[	[	X
fcis-28893	258	2	18	18	NUM
fcis-28893	258	3	]	]	X
fcis-28893	258	4	w.	w.	PROPN
fcis-28893	258	5	tan	tan	PROPN
fcis-28893	258	6	,	,	PUNCT
fcis-28893	258	7	c.	c.	PROPN
fcis-28893	258	8	ding	ding	PROPN
fcis-28893	258	9	,	,	PUNCT
fcis-28893	258	10	j.	j.	PROPN
fcis-28893	258	11	jiang	jiang	PROPN
fcis-28893	258	12	,	,	PUNCT
fcis-28893	258	13	f.	f.	PROPN
fcis-28893	258	14	wang	wang	PROPN
fcis-28893	258	15	,	,	PUNCT
fcis-28893	258	16	y.	y.	PROPN
fcis-28893	258	17	zhan	zhan	PROPN
fcis-28893	258	18	,	,	PUNCT
fcis-28893	258	19	and	and	CCONJ
fcis-28893	258	20	d.	d.	PROPN
fcis-28893	258	21	tao	tao	PROPN
fcis-28893	258	22	,	,	PUNCT
fcis-28893	258	23	"	"	PUNCT
fcis-28893	258	24	harnessing	harness	VERB
fcis-28893	258	25	the	the	DET
fcis-28893	258	26	power	power	NOUN
fcis-28893	258	27	of	of	ADP
fcis-28893	258	28	mllms	mllm	NOUN
fcis-28893	258	29	for	for	ADP
fcis-28893	258	30	transferable	transferable	ADJ
fcis-28893	258	31	text	text	NOUN
fcis-28893	258	32	-	-	PUNCT
fcis-28893	258	33	toimage	toimage	NOUN
fcis-28893	258	34	person	person	NOUN
fcis-28893	258	35	reid	reid	NOUN
fcis-28893	258	36	,	,	PUNCT
fcis-28893	258	37	"	"	PUNCT
fcis-28893	258	38	p.	p.	NOUN
fcis-28893	258	39	arxiv:2405.04940doi	arxiv:2405.04940doi	PROPN
fcis-28893	258	40	:	:	PUNCT
fcis-28893	258	41	10.48550	10.48550	NUM
fcis-28893	258	42	/	/	SYM
fcis-28893	258	43	arxiv	arxiv	NOUN
fcis-28893	258	44	.	.	PUNCT
fcis-28893	259	1	2405.04940	2405.04940	NOUN
fcis-28893	259	2	.	.	PUNCT
fcis-28893	260	1	[	[	X
fcis-28893	260	2	19	19	NUM
fcis-28893	260	3	]	]	X
fcis-28893	260	4	y.	y.	PROPN
fcis-28893	260	5	wu	wu	PROPN
fcis-28893	260	6	,	,	PUNCT
fcis-28893	260	7	z.	z.	PROPN
fcis-28893	260	8	yan	yan	PROPN
fcis-28893	260	9	,	,	PUNCT
fcis-28893	260	10	x.	x.	PROPN
fcis-28893	260	11	han	han	PROPN
fcis-28893	260	12	,	,	PUNCT
fcis-28893	260	13	g.	g.	PROPN
fcis-28893	260	14	li	li	PROPN
fcis-28893	260	15	,	,	PUNCT
fcis-28893	260	16	c.	c.	PROPN
fcis-28893	260	17	zou	zou	PROPN
fcis-28893	260	18	,	,	PUNCT
fcis-28893	260	19	and	and	CCONJ
fcis-28893	260	20	s.	s.	PROPN
fcis-28893	260	21	cui	cui	PROPN
fcis-28893	260	22	,	,	PUNCT
fcis-28893	260	23	"	"	PUNCT
fcis-28893	260	24	lapscore	lapscore	ADJ
fcis-28893	260	25	:	:	PUNCT
fcis-28893	260	26	language	language	NOUN
fcis-28893	260	27	-	-	PUNCT
fcis-28893	260	28	guided	guide	VERB
fcis-28893	260	29	person	person	NOUN
fcis-28893	260	30	search	search	NOUN
fcis-28893	260	31	via	via	ADP
fcis-28893	260	32	color	color	NOUN
fcis-28893	260	33	reasoning	reasoning	NOUN
fcis-28893	260	34	,	,	PUNCT
fcis-28893	260	35	"	"	PUNCT
fcis-28893	260	36	in	in	ADP
fcis-28893	260	37	2021	2021	NUM
fcis-28893	260	38	ieee	ieee	NOUN
fcis-28893	260	39	/	/	SYM
fcis-28893	260	40	cvf	cvf	NOUN
fcis-28893	260	41	international	international	ADJ
fcis-28893	260	42	conference	conference	NOUN
fcis-28893	260	43	on	on	ADP
fcis-28893	260	44	computer	computer	NOUN
fcis-28893	260	45	vision	vision	NOUN
fcis-28893	260	46	(	(	PUNCT
fcis-28893	260	47	iccv	iccv	PROPN
fcis-28893	260	48	)	)	PUNCT
fcis-28893	260	49	,	,	PUNCT
fcis-28893	260	50	10	10	NUM
fcis-28893	260	51	-	-	SYM
fcis-28893	260	52	17	17	NUM
fcis-28893	260	53	oct	oct	NOUN
fcis-28893	260	54	.	.	PROPN
fcis-28893	260	55	2021	2021	NUM
fcis-28893	260	56	2021	2021	NUM
fcis-28893	260	57	:	:	PUNCT
fcis-28893	260	58	ieee	ieee	NOUN
fcis-28893	260	59	,	,	PUNCT
fcis-28893	260	60	pp	pp	ADJ
fcis-28893	260	61	.	.	PUNCT
fcis-28893	261	1	1604	1604	NUM
fcis-28893	261	2	-	-	SYM
fcis-28893	261	3	1613	1613	NUM
fcis-28893	261	4	,	,	PUNCT
fcis-28893	261	5	doi	doi	NOUN
fcis-28893	261	6	:	:	PUNCT
fcis-28893	261	7	10.1109/	10.1109/	NUM
fcis-28893	261	8	iccv48922.2021.00165	iccv48922.2021.00165	PROPN
fcis-28893	261	9	.	.	PUNCT
fcis-28893	262	1	[	[	X
fcis-28893	262	2	20	20	NUM
fcis-28893	262	3	]	]	PUNCT
fcis-28893	262	4	x.	x.	NOUN
fcis-28893	262	5	wu	wu	PROPN
fcis-28893	262	6	,	,	PUNCT
fcis-28893	262	7	w.	w.	PROPN
fcis-28893	262	8	ma	ma	PROPN
fcis-28893	262	9	,	,	PUNCT
fcis-28893	262	10	d.	d.	PROPN
fcis-28893	262	11	guo	guo	PROPN
fcis-28893	262	12	,	,	PUNCT
fcis-28893	262	13	t.	t.	PROPN
fcis-28893	262	14	zhou	zhou	PROPN
fcis-28893	262	15	,	,	PUNCT
fcis-28893	262	16	s.	s.	PROPN
fcis-28893	262	17	zhao	zhao	PROPN
fcis-28893	262	18	,	,	PUNCT
fcis-28893	262	19	and	and	CCONJ
fcis-28893	262	20	z.	z.	PROPN
fcis-28893	262	21	cai	cai	PROPN
fcis-28893	262	22	,	,	PUNCT
fcis-28893	262	23	"	"	PUNCT
fcis-28893	262	24	textbased	textbase	VERB
fcis-28893	262	25	occluded	occluded	ADJ
fcis-28893	262	26	person	person	NOUN
fcis-28893	262	27	re	re	NOUN
fcis-28893	262	28	-	-	NOUN
fcis-28893	262	29	identification	identification	NOUN
fcis-28893	262	30	via	via	ADP
fcis-28893	262	31	multigranularity	multigranularity	NOUN
fcis-28893	262	32	contrastive	contrastive	ADJ
fcis-28893	262	33	consistency	consistency	NOUN
fcis-28893	262	34	learning	learning	NOUN
fcis-28893	262	35	,	,	PUNCT
fcis-28893	262	36	"	"	PUNCT
fcis-28893	262	37	proceedings	proceeding	NOUN
fcis-28893	262	38	of	of	ADP
fcis-28893	262	39	the	the	DET
fcis-28893	262	40	aaai	aaai	PROPN
fcis-28893	262	41	conference	conference	NOUN
fcis-28893	262	42	on	on	ADP
fcis-28893	262	43	artificial	artificial	ADJ
fcis-28893	262	44	intelligence	intelligence	NOUN
fcis-28893	262	45	,	,	PUNCT
fcis-28893	262	46	vol	vol	NOUN
fcis-28893	262	47	.	.	PROPN
fcis-28893	263	1	38	38	NUM
fcis-28893	263	2	,	,	PUNCT
fcis-28893	263	3	no	no	INTJ
fcis-28893	263	4	.	.	NOUN
fcis-28893	263	5	6	6	NUM
fcis-28893	263	6	,	,	PUNCT
fcis-28893	263	7	pp	pp	ADJ
fcis-28893	263	8	.	.	PUNCT
fcis-28893	264	1	6162	6162	NUM
fcis-28893	264	2	-	-	SYM
fcis-28893	264	3	6170	6170	NUM
fcis-28893	264	4	,	,	PUNCT
fcis-28893	264	5	03/24	03/24	NUM
fcis-28893	264	6	2024	2024	NUM
fcis-28893	264	7	,	,	PUNCT
fcis-28893	264	8	doi	doi	NOUN
fcis-28893	264	9	:	:	PUNCT
fcis-28893	264	10	10.1609	10.1609	NUM
fcis-28893	264	11	/	/	SYM
fcis-28893	264	12	aaai	aaai	NOUN
fcis-28893	264	13	.	.	PUNCT
fcis-28893	265	1	v38i6.28433	v38i6.28433	X
fcis-28893	265	2	.	.	PUNCT
fcis-28893	266	1	[	[	X
fcis-28893	266	2	21	21	NUM
fcis-28893	266	3	]	]	X
fcis-28893	266	4	y.	y.	PROPN
fcis-28893	266	5	bai	bai	PROPN
fcis-28893	266	6	et	et	PROPN
fcis-28893	266	7	al	al	PROPN
fcis-28893	266	8	.	.	PROPN
fcis-28893	266	9	,	,	PUNCT
fcis-28893	266	10	"	"	PUNCT
fcis-28893	266	11	rasa	rasa	NOUN
fcis-28893	266	12	:	:	PUNCT
fcis-28893	266	13	relation	relation	NOUN
fcis-28893	266	14	and	and	CCONJ
fcis-28893	266	15	sensitivity	sensitivity	NOUN
fcis-28893	266	16	aware	aware	ADJ
fcis-28893	266	17	representation	representation	NOUN
fcis-28893	266	18	learning	learn	VERB
fcis-28893	266	19	for	for	ADP
fcis-28893	266	20	text	text	NOUN
fcis-28893	266	21	-	-	PUNCT
fcis-28893	266	22	based	base	VERB
fcis-28893	266	23	person	person	NOUN
fcis-28893	266	24	search	search	NOUN
fcis-28893	266	25	,	,	PUNCT
fcis-28893	266	26	"	"	PUNCT
fcis-28893	266	27	presented	present	VERB
fcis-28893	266	28	at	at	ADP
fcis-28893	266	29	the	the	DET
fcis-28893	266	30	proceedings	proceeding	NOUN
fcis-28893	266	31	of	of	ADP
fcis-28893	266	32	the	the	DET
fcis-28893	266	33	thirty	thirty	NUM
fcis-28893	266	34	-	-	PUNCT
fcis-28893	266	35	second	second	NOUN
fcis-28893	266	36	international	international	ADJ
fcis-28893	266	37	joint	joint	ADJ
fcis-28893	266	38	conference	conference	NOUN
fcis-28893	266	39	on	on	ADP
fcis-28893	266	40	artificial	artificial	ADJ
fcis-28893	266	41	intelligence	intelligence	NOUN
fcis-28893	266	42	,	,	PUNCT
fcis-28893	266	43	macao	macao	PROPN
fcis-28893	266	44	,	,	PUNCT
fcis-28893	266	45	p.r.china	p.r.china	PROPN
fcis-28893	266	46	,	,	PUNCT
fcis-28893	266	47	2023	2023	NUM
fcis-28893	266	48	.	.	PUNCT
fcis-28893	267	1	[	[	X
fcis-28893	267	2	online	online	X
fcis-28893	267	3	]	]	X
fcis-28893	267	4	.	.	PUNCT
fcis-28893	268	1	available	available	ADJ
fcis-28893	268	2	:	:	PUNCT
fcis-28893	268	3	https://doi.org/	https://doi.org/	VERB
fcis-28893	268	4	10.24963/	10.24963/	NUM
fcis-28893	268	5	ijcai	ijcai	NOUN
fcis-28893	268	6	.	.	PUNCT
fcis-28893	269	1	2023/62	2023/62	NUM
fcis-28893	269	2	.	.	PUNCT