SlideShare a Scribd company logo
1 of 19
Download to read offline
Dominant	Codewords	Selec2on	with	Topic	Model	
for	Ac2on	Recogni2on
Hirokatsu	KATAOKA,	Masaki	Hayashi†,	Yoshimitsu	AOKI†,	Kenji	IWATA,	
Yutaka	SATOH,	Slobodan	Ilic‡	
Na2onal	Ins2tute	of	Advanced	Industrial	Science	and	Technology	(AIST)	
†	Keio	University	
‡	Technische	Universität	München	
		
hPp://www.hirokatsukataoka.net/
Human	Sensing	
•  Several	tasks	are	included	in	vision-based	human	sensing	
–  Detec2on,	tracking,	face	recogni2on	
–  Posture	es2ma2on,	ac2on	analysis	(event	recogni2on)	
–  Ac2on	recogni2on	is	able	to	extend	human	sensing	applica2ons	
Mental	state
Body	Situa2on
APen2on
Ac2on	Analysis
shakinghands
Look	at	people
Detec2on
Gaze	Es2ma2on
Ac2on	Recogni2on
Posture	Es2ma2on
Face	Recogni2on
Trajectory	extrac2on
Tracking
Understanding	Human	Ac2ons	
•  Ac2on	is	defined	as	something	that	people	do	or	cause	to	happen	
–  e.g.	walking,	running, si^ng	
This	image	contains	a	man	walking	
•  Ac2on	recogni2on	
Classifica2on	
•  Ac2on	detec2on	
Classifica2on	&	localiza2on	
Walking	
Walking	
Ac2on	Recogni2on
Ac2on	Detec2on
Current	Ac2on	Recogni2on	
•  Trajectory-based	representa2on	
Improved	Dense	Trajectories	(IDT)	[Wang+,	ICCV13]
Trajectory-pooled	Deep-convolu2onal	Descriptors		
(TDD)	[Wang+,	CVPR15]
•  “Dense”	representa2on	is	important	
•  Trajectory-based	hand-craied	
descrip2on	(HOG/HOF/MBH/Traj.)	
•  Intersec2on	of	hand-craied	and	deep-
learned	features	
•  Intui2vely,	IDT	feature	is	replaced	by	
ConvMaps	
•  CNN	is	based	on	Two-stream	ConvNet	
[Simonyan+,	NIPS14]
Typical	Ac2on	Recogni2on	Pipeline	
Trajectories	/	keypoints	extrac2on Feature	descrip2on
Codeword	pooling Classifica2on
e.g.	BoF,	VLAD,	Fisher	Vectors hPp://www.analy2calway.com/images/SMV/svmFeatureSpace2.gif	
e.g.	SVM
Typical	Ac2on	Recogni2on	Pipeline	
•  Two	strategies	for	more	sophis2cated	feature	vector	
Trajectory	extrac2on Trajectory	descrip2on
1.	Trajectory	refinement!	
(main	contribu2on)	
2.	BePer	feature!
Our	proposal	
•  1.	Dominant	codewords	selec2on	(DCS)	
–  Significant	ac2on	representa2on	by	using	topic	model	(typical	LDA)	
–  Topic-based	grouping	from	dense	trajectories	
–  Noise	elimina2on	at	each	topic	
•  2.	Co-occurrence	feature	representa2on	for	bePer	feat.	
–  As	an	addi2onal	feature	into	typical	HOG/	HOF	/	MBH	/	Traj.	
–  The	feature	is	based	on	[Kataoka+,	ACCV14]	which	is	improved	co-occurrence	
feature	focusing	on	extrac2ng	subtle	mo2on	
×	×	×	×	
×	×	
×	
×	×	
×	
×	
×	
×	
×	
×	
There	are	some	noise	even	in	the	sta2c	scene!
Flowchart	of	dominant	codewords	selec2on	(DCS)	
•  1.	Feature	extrac2on	
–  Dense	Trajectories	
•  2.	Topic	modeling	to	divide	primi2ve	mo2on	
–  Each	topic	is	approxima2ng	each	primi2ve	mo2on	
•  3.	Noise	canceling	at	each	topic	
•  4.	Dominant	dense	trajectories	(DDT)
Feature	extrac2on	
•  Improved	dense	trajectories	[Wang+,	ICCV13]	
–  HOG/	HOF/	MBHx,	/MBHy	
–  CoHOG,	Extended	CoHOG	[Kataoka+,	ACCV14]	
–  Bag-of-features	(BoF)	to	input	of	topic	model	
Extended	CoHOG
Topic	Model	
•  Latent	Dirichlet	alloca2on	(LDA)	[Blei+,	JMLR03]	
–  We	applied	simplified	model	[Griffiths+,	04]	
–  Typical	LDA	
–  Parameters	
•  Probability	distribu2on	of	topic:	Θ	
•  Topic:	T	
•  Hyper	parameter:	α,	β	
•  DT-BoF	from	video:	v
Topic	Model	for	DCS	
•  Noise	rejec2on	at	each	topic	
–  Each	topic	indicates	each	“primi2ve	mo2on”	
–  In-topic	adap2ve	thresholding	
×	×	×	×	
×	×	
×	
×	×	
×	
×	
“cut	off	ends”	 Topic1	 Topic2	 Topic3	 Topic4	
MPII	cooking	
Each	topic	indicates	each	“primi2ve	mo2on”
×	 ×	
×	×	 ×	×	
×	 ×	
×	×	×	×	
×	×	
×	
×	×	
×	
×	
“cut	off	ends”	
×	
×	×	 ×	 ×	×	×	
×	
×	
×	
×	×	×	 Unsophis2cated	rejec2on	is	not	desirable
×	 Eliminated	vector
Dominant	dense	trajectoies	(DDT)	
•  All	refined	topics	are	integrated	
–  DDT	
–  DDT	is	a	mixture	of	dominant	codewords	using	AND	opera2on		
	 e.g.		
•  T1	=	{vw1,	vw3,	vw5}	=>	T1’	=	{vw1,	vw5}	
•  T2	=	{vw1,	vw2,	vw7}	=>	T2’	=	{vw1,	vw2,	vw7}	
•  T3	=	{vw3,	vw6,	vw7}	=>	T3’	=	{vw7}	
•  DDT	=	{vw1,	vw2,	vw5,	vw7}	
•  DDT	is	sophis2cated	representa2on	with	
dominant	codewords
Experiments	
•  Four	datasets	for	ac2on	recogni2on	
【INRIA surgery】 View: 4 Activity: 4	
view1	 view2	
view3	 view4	
【IXMAS】 View: 5 Activity: 11	
【MPII cooking activities】 View: 1 Activity: 65	
view1	 view2	
view3	 view4	
view5	
【NTSEL traffic】 View: 1 Activity: 4
Parameters	
•  DT,	classifica2on	&	LDA	are	based	on	the	previous	works	
–  # of	codeword: 4,000	
–  # of	trajectory	pooling:	15	frame	
–  The	parameters	are	following	original	DT	
–  α	and	β	of	LDA	are	set	at	1.0	and	0.01	
–  1,000	Gibbs	sampler	itera2ons	
–  The	topic	modeling	params	are	based	on	[Griffiths	and	Steyvers,	04]	
•  Dominant	codewords	selec2on	(DCS)	
–  Topic-based	codewords	are	set	as	1%	of	the	frequency	
–  #	of	topic:	#	of	view		x	#	of	ac2on
Comparison	of	DT	
•  Dominant	DT	vs	DT	
–  By	using	HOG,	HOF,	MBHx,	MBHy	and	combined	features	
–  Dominant	codewords	selec2on	is	effec2ve	on	the	DT
Co-occurrence	feature	in	DDT	
•  On	fine-grained	dataset	
–  MPII	cooking	[Rohrbach+,	CVPR12]	
–  Extended	co-occurrence	feature	is	improved	by	DCS	
–  DDT	is	improved	with	extended	co-occurrence	feature	
–  [Ni+,	ECCV14]	and	[Ni+,	CVPR15]	are	state-of-the-art	with	mid-level	features
Topic	Visualiza2on	
×	
×	
×	
×	
×	
×	
×	
×	
×	
×	 ×	×	
×	×	×	
×	
×	×	×	×	×	
×	
×	×	
×	
×	
×	×	
×	 ×	
×	
×	
×	×	 ×	
×	
×	
×	
×	×	×	×	×	
×	
×	
×	
×	×	×	×	×	×	
×	×	
×	
×	×	
×	×	 ×	
×	×	
×	×	
×	×	
×	
×	×	×	
×	
×	
×	×	
×	
×	×	
×	
×	
×	 ×	
×	
×	
×	
×	×	
×	
×	×	
×	
×	
×	×	×	×	×	×	
×	
×	
×	
×	×	
×	
×	×	×	
×	×	×	×	
×	×	
×	
×	×	
×	
×	
Topic1	(View	2)	
“sit”	View	1	
“sit”	View	2	
Topic1	(View	1)	
Topic2	(View	2)	
Topic2	(View	1)	
Topic3	(View	2)	
Topic3	(View	1)	
Topic4	(View	2)	
Topic4	(View	1)	
“getup”	View	2	 Topic1	(View	2)	 Topic2	(View	2)	 Topic3	(View	2)	 Topic4	(View	2)	
“crossing”	 Topic1	 Topic2	 Topic3	 Topic4	
“cut	off	ends”	 Topic1	 Topic2	 Topic3	 Topic4	
INRIA	surgery	
INRIA	surgery	
IXMAS	
NTSEL	traffic	
MPII	cooking
Conclusion	
•  Dominant	codewords	selec2on	(DCS)	for	ac2on	recogni2on	
–  We	applied	topic	model	for	noise	canceling	in	trajectory-based	representa2on	
–  Extra	feature	(co-occurrence	feature)	
•  Future	works	
–  Exploring	parameters	and	more	effec2ve	visualiza2on	(on-going…)	
–  Seman2c	flow	with	topic	models
【CVPR2016_LAP】Dominant Codewords Selection with Topic Model for Action Recognition

More Related Content

Viewers also liked

【ECCV 2016 BNMW】Human Action Recognition without Human
【ECCV 2016 BNMW】Human Action Recognition without Human【ECCV 2016 BNMW】Human Action Recognition without Human
【ECCV 2016 BNMW】Human Action Recognition without Human
Hirokatsu Kataoka
 
【ITSC2015】Fine-grained Walking Activity Recognition via Driving Recorder Dataset
【ITSC2015】Fine-grained Walking Activity Recognition via Driving Recorder Dataset【ITSC2015】Fine-grained Walking Activity Recognition via Driving Recorder Dataset
【ITSC2015】Fine-grained Walking Activity Recognition via Driving Recorder Dataset
Hirokatsu Kataoka
 

Viewers also liked (10)

CVPR 2016 速報
CVPR 2016 速報CVPR 2016 速報
CVPR 2016 速報
 
【チュートリアル】コンピュータビジョンによる動画認識
【チュートリアル】コンピュータビジョンによる動画認識【チュートリアル】コンピュータビジョンによる動画認識
【チュートリアル】コンピュータビジョンによる動画認識
 
【ECCV 2016 BNMW】Human Action Recognition without Human
【ECCV 2016 BNMW】Human Action Recognition without Human【ECCV 2016 BNMW】Human Action Recognition without Human
【ECCV 2016 BNMW】Human Action Recognition without Human
 
【ITSC2015】Fine-grained Walking Activity Recognition via Driving Recorder Dataset
【ITSC2015】Fine-grained Walking Activity Recognition via Driving Recorder Dataset【ITSC2015】Fine-grained Walking Activity Recognition via Driving Recorder Dataset
【ITSC2015】Fine-grained Walking Activity Recognition via Driving Recorder Dataset
 
【2016.08】cvpaper.challenge2016
【2016.08】cvpaper.challenge2016【2016.08】cvpaper.challenge2016
【2016.08】cvpaper.challenge2016
 
PythonによるCVアルゴリズム実装
PythonによるCVアルゴリズム実装PythonによるCVアルゴリズム実装
PythonによるCVアルゴリズム実装
 
CNN Tutorial
CNN TutorialCNN Tutorial
CNN Tutorial
 
20150930
2015093020150930
20150930
 
Convolutional Neural Networks のトレンド @WBAFLカジュアルトーク#2
Convolutional Neural Networks のトレンド @WBAFLカジュアルトーク#2Convolutional Neural Networks のトレンド @WBAFLカジュアルトーク#2
Convolutional Neural Networks のトレンド @WBAFLカジュアルトーク#2
 
CVPR 2016 まとめ v1
CVPR 2016 まとめ v1CVPR 2016 まとめ v1
CVPR 2016 まとめ v1
 

Recently uploaded

development of diagnostic enzyme assay to detect leuser virus
development of diagnostic enzyme assay to detect leuser virusdevelopment of diagnostic enzyme assay to detect leuser virus
development of diagnostic enzyme assay to detect leuser virus
NazaninKarimi6
 
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
?#DUbAI#??##{{(☎️+971_581248768%)**%*]'#abortion pills for sale in dubai@
 
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdfPests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
PirithiRaju
 
Digital Dentistry.Digital Dentistryvv.pptx
Digital Dentistry.Digital Dentistryvv.pptxDigital Dentistry.Digital Dentistryvv.pptx
Digital Dentistry.Digital Dentistryvv.pptx
MohamedFarag457087
 
The Mariana Trench remarkable geological features on Earth.pptx
The Mariana Trench remarkable geological features on Earth.pptxThe Mariana Trench remarkable geological features on Earth.pptx
The Mariana Trench remarkable geological features on Earth.pptx
seri bangash
 
Pests of mustard_Identification_Management_Dr.UPR.pdf
Pests of mustard_Identification_Management_Dr.UPR.pdfPests of mustard_Identification_Management_Dr.UPR.pdf
Pests of mustard_Identification_Management_Dr.UPR.pdf
PirithiRaju
 

Recently uploaded (20)

Justdial Call Girls In Indirapuram, Ghaziabad, 8800357707 Escorts Service
Justdial Call Girls In Indirapuram, Ghaziabad, 8800357707 Escorts ServiceJustdial Call Girls In Indirapuram, Ghaziabad, 8800357707 Escorts Service
Justdial Call Girls In Indirapuram, Ghaziabad, 8800357707 Escorts Service
 
Forensic Biology & Its biological significance.pdf
Forensic Biology & Its biological significance.pdfForensic Biology & Its biological significance.pdf
Forensic Biology & Its biological significance.pdf
 
Sector 62, Noida Call girls :8448380779 Model Escorts | 100% verified
Sector 62, Noida Call girls :8448380779 Model Escorts | 100% verifiedSector 62, Noida Call girls :8448380779 Model Escorts | 100% verified
Sector 62, Noida Call girls :8448380779 Model Escorts | 100% verified
 
Locating and isolating a gene, FISH, GISH, Chromosome walking and jumping, te...
Locating and isolating a gene, FISH, GISH, Chromosome walking and jumping, te...Locating and isolating a gene, FISH, GISH, Chromosome walking and jumping, te...
Locating and isolating a gene, FISH, GISH, Chromosome walking and jumping, te...
 
Zoology 5th semester notes( Sumit_yadav).pdf
Zoology 5th semester notes( Sumit_yadav).pdfZoology 5th semester notes( Sumit_yadav).pdf
Zoology 5th semester notes( Sumit_yadav).pdf
 
development of diagnostic enzyme assay to detect leuser virus
development of diagnostic enzyme assay to detect leuser virusdevelopment of diagnostic enzyme assay to detect leuser virus
development of diagnostic enzyme assay to detect leuser virus
 
COST ESTIMATION FOR A RESEARCH PROJECT.pptx
COST ESTIMATION FOR A RESEARCH PROJECT.pptxCOST ESTIMATION FOR A RESEARCH PROJECT.pptx
COST ESTIMATION FOR A RESEARCH PROJECT.pptx
 
chemical bonding Essentials of Physical Chemistry2.pdf
chemical bonding Essentials of Physical Chemistry2.pdfchemical bonding Essentials of Physical Chemistry2.pdf
chemical bonding Essentials of Physical Chemistry2.pdf
 
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
+971581248768>> SAFE AND ORIGINAL ABORTION PILLS FOR SALE IN DUBAI AND ABUDHA...
 
GBSN - Microbiology (Unit 1)
GBSN - Microbiology (Unit 1)GBSN - Microbiology (Unit 1)
GBSN - Microbiology (Unit 1)
 
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdfPests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
Pests of cotton_Borer_Pests_Binomics_Dr.UPR.pdf
 
Pulmonary drug delivery system M.pharm -2nd sem P'ceutics
Pulmonary drug delivery system M.pharm -2nd sem P'ceuticsPulmonary drug delivery system M.pharm -2nd sem P'ceutics
Pulmonary drug delivery system M.pharm -2nd sem P'ceutics
 
Digital Dentistry.Digital Dentistryvv.pptx
Digital Dentistry.Digital Dentistryvv.pptxDigital Dentistry.Digital Dentistryvv.pptx
Digital Dentistry.Digital Dentistryvv.pptx
 
module for grade 9 for distance learning
module for grade 9 for distance learningmodule for grade 9 for distance learning
module for grade 9 for distance learning
 
Dubai Call Girls Beauty Face Teen O525547819 Call Girls Dubai Young
Dubai Call Girls Beauty Face Teen O525547819 Call Girls Dubai YoungDubai Call Girls Beauty Face Teen O525547819 Call Girls Dubai Young
Dubai Call Girls Beauty Face Teen O525547819 Call Girls Dubai Young
 
FAIRSpectra - Enabling the FAIRification of Analytical Science
FAIRSpectra - Enabling the FAIRification of Analytical ScienceFAIRSpectra - Enabling the FAIRification of Analytical Science
FAIRSpectra - Enabling the FAIRification of Analytical Science
 
PSYCHOSOCIAL NEEDS. in nursing II sem pptx
PSYCHOSOCIAL NEEDS. in nursing II sem pptxPSYCHOSOCIAL NEEDS. in nursing II sem pptx
PSYCHOSOCIAL NEEDS. in nursing II sem pptx
 
The Mariana Trench remarkable geological features on Earth.pptx
The Mariana Trench remarkable geological features on Earth.pptxThe Mariana Trench remarkable geological features on Earth.pptx
The Mariana Trench remarkable geological features on Earth.pptx
 
Molecular markers- RFLP, RAPD, AFLP, SNP etc.
Molecular markers- RFLP, RAPD, AFLP, SNP etc.Molecular markers- RFLP, RAPD, AFLP, SNP etc.
Molecular markers- RFLP, RAPD, AFLP, SNP etc.
 
Pests of mustard_Identification_Management_Dr.UPR.pdf
Pests of mustard_Identification_Management_Dr.UPR.pdfPests of mustard_Identification_Management_Dr.UPR.pdf
Pests of mustard_Identification_Management_Dr.UPR.pdf
 

【CVPR2016_LAP】Dominant Codewords Selection with Topic Model for Action Recognition