+ All Categories
Home > Documents > 0934. ONR MURI Grant No. N00014- x,y,t 31 Learning …jhchoi/paper/cvpr2016_anomaly_poster.pdf ·...

0934. ONR MURI Grant No. N00014- x,y,t 31 Learning …jhchoi/paper/cvpr2016_anomaly_poster.pdf ·...

Date post: 09-Apr-2018
Category:
Upload: trinhngoc
View: 216 times
Download: 2 times
Share this document with a friend
1
Learning Temporal Regularity in Video Sequence Data and Codes: http://www.ee.ucr.edu/~mhasan/regularity.html Mahmudul Hasan 1 Jonghyun Choi 2 Jan Neumann 2 Amit K. Roy-Chowdhury 1 Larry S. Davis 3 This research is partly supported by NSF grant IIS-1316934and ONR MURI Grant No. N00014-10-1-0934. Anomaly Detection: Comparison with State-of-the-art Methods Visualizing Temporal Regularity Motivation q Watching long hours of uncontrolled videos is extremely tedious No dataset bias compensated Regularities by the General Model More Results are in the paper Applications and Experiments A sample irregular frame Synthesized Regular frame Regularity score A sample irregular frameSynthesized Regular frame Regularity score UCSD Ped2 Dataset UT-AustinSubway-ExitDataset Predicting Near Past and Future CUHK Avenue Dataset UT-AustinSubway-ExitDataset Regularity Scores UT-AustinSubway-ExitDataset CUHK Avenue Dataset UT-Austin Subway-Enter Dataset Feature Based Fully Connected Autoencoder (FConn) End-to-End Fully Convolutional Autoencoder (FConv) UC Riverside 1 Comcast Labs, DC 2 University of Maryland, College Park 3 q We want to segment ‘meaningful’ moments in such videos without supervision Challenges q Learning a classification model of these meaningful (irregular) moments is not trivial because – q Ill defined (anything can be meaningful) q Infrequent (small training data) q Labeling them is expensive. Approach q Use two high capacity generative models - deep neural network based auto-encoders (DNN-AE): q Fully connected DNN-AE on hand-crafted feature q Fully convolutional DNN-AE on frames q Regular input -> Reconstruction cost is small. q Irregular input -> Reconstruction cost is large. Exemplar output of our model when there are irregular motions, the regularity score drops significantly. Our Objective q Learning a generative model for regularity with q Limited supervision required q Ease of learning q Multiple datasets used to train Overview of the Approach Model Architecture Input: HOG+HOF collected around the trajectory of interest point Input Data Layer and Data Augmentation q Input cuboid sizes – 5, 10, and 20 (We use 10) q Large cuboid -> better discrimination and increased running time. q Sliding window size: 10, 20, and 30 with sample rate of stride 1, 2, and 3 respectively. q Sliding windows are moved 2 frames at a time. Training a General Model Regularity Score: s(t) Optimization q LR Scheme: AdaGrad q Init. LR: 0.001 (FConv) and 0.01 (FConn) q Mini-batch size: 1024 (FConv) and 32 (FConn) q Weight initialization: Xavier e(x, y, t)= kI (x, y, t) - f W (I (x, y, t))k 2 e(t)= (x,y ) e(x, y, t) s(t)=1 - e(t) - min t e(t) max t e(t)
Transcript
Page 1: 0934. ONR MURI Grant No. N00014- x,y,t 31 Learning …jhchoi/paper/cvpr2016_anomaly_poster.pdf · Learning Temporal Regularity in Video Sequence ... Regularity score ... Austin Subway-Exit

Learning

Tempo

ralRegularity

inVideo

Seq

uence

Dataand

Cod

es:

http://w

ww.ee.ucr.edu/~mhasan/regularity.htm

l

Mah

mud

ulHasan

1Jong

hyun

Cho

i2JanNeu

man

n2Am

itK.Roy-Cho

wdh

ury1

LarryS.Davis3

Thisresearchispartly

sup

ported

byNSFgrantIIS-13

1693

4and

ONRMURIGrantNo.N00

014-10

-1-093

4.

Anom

alyDe

tection:Com

parison

with

State-of-the

-artM

etho

ds

VisualizingTempo

ralR

egularity

Motivation

qWatchinglonghoursof

uncontrolledvideosisextremely

tedious

Nodatasetbiascompensated

Regularitiesb

ytheGe

neralM

odel

MoreRe

sultsareinth

epa

per

Applicationsand

Experim

ents

Asampleirregularfram

eSynthe

sized

Regularfram

eRe

gularitys

core

Asampleirregularfram

e Syn

thesize

dRe

gularframeRe

gularitys

core

UCSDPe

d2Dataset

UT-Au

stinSu

bway-ExitD

ataset

Pred

ictin

gNearP

asta

ndFuture

CUHK

Avenu

eDa

taset

UT-Au

stinSu

bway-ExitD

ataset

RegularityScores

UT-Au

stinSu

bway-ExitD

ataset

CUHK

Avenu

eDa

taset

UT-Au

stinSu

bway-EnterDataset

FeatureBa

sedFullyCon

nected

Autoe

ncod

er(FCo

nn)

End-to-End

FullyCon

volutio

nalA

utoe

ncod

er(FCo

nv)

UCRiverside1

ComcastLabs,DC2

UniversityofMaryland,CollegePark

3

qWewanttosegment‘m

eaningful’m

omentsinsuch

videoswithoutsupervision

Challenges

qLearningaclassificationmodelofthesemeaningful

(irregular)m

omentsisnottrivialbecause–

qIlldefined(anythingcanbem

eaningful)

qInfrequent(smalltrainingdata)

qLabelingthemisexpensive.

Approa

chq

Usetwohighcapacitygenerativem

odels-deep

neuralnetw

orkbasedauto-encoders(DNN-AE):

qFullyconnectedDNN-AEonhand-craftedfeature

qFullyconvolutionalDNN-AEonframes

qRe

gularinput->Reconstructioncostissmall.

qIrregular

input->Reconstructioncostislarge.

Exe

mpl

ar o

utpu

t of o

ur m

odel

whe

n th

ere

are

irreg

ular

mot

ions

, the

regu

larit

y sc

ore

drop

s sig

nific

antly

.

OurObjectiv

eq

Learningagenerativem

odelforregularitywith

qLimitedsupervisionrequired

qEaseoflearning

qMultipledatasetsusedtotrain

Overviewofthe

App

roach

Mod

elArchitecture

Input:HOG+HOF

collectedaroundthe

trajectoryofinterest

point

Inpu

tDataLayera

ndDataAu

gmen

tatio

nq

Inputcuboidsizes–5,10,and20(Weuse10)

qLargecuboid->betterdiscrim

inationandincreasedrunningtim

e.

qSlidingwindowsize:10,20,and30withsamplerateofstride

1,2,and3respectively.

qSlidingwindowsarem

oved2framesatatim

e.

TrainingaGen

eralM

odel

RegularityScore:s(t)

Optim

izatio

nq

LRScheme:AdaGrad

qInit.LR:0.001(FConv)and

0.01(FConn)

qMini-batchsize:1024

(FConv)and32(FConn)

qWeightinitialization:Xavier

e(x,y,t)=

kI(x,y,t)�

f

W(I(x,y,t))k 2

e(t)=

⌃(x

,y)e(x,y,t)

s(t)

=1�

e(t)�

min

te(t)

max

te(t)

Recommended