More sliding window detection: Discriminative part-based models Many slides based on P....

23
More sliding window detection: Discriminative part-based models Many slides based on P . Felzenszwalb

Transcript of More sliding window detection: Discriminative part-based models Many slides based on P....

Page 1: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

More sliding window detection:Discriminative part-based models

Many slides based on P. Felzenszwalb

Page 2: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Challenge: Generic object detection

Page 3: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Pedestrian detection• Features: Histograms of oriented gradients (HOG)

• Partition image into 8x8 pixel blocks and compute histogram of gradient orientations in each block

• Learn a pedestrian template using a linear support vector machine• At test time, convolve feature map with template

N. Dalal and B. Triggs, Histograms of Oriented Gradients for Human Detection, CVPR 2005

TemplateHOG feature map Detector response map

Page 4: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Discriminative part-based models

P. Felzenszwalb, R. Girshick, D. McAllester, D. Ramanan, Object Detection with Discriminatively Trained Part Based Models, PAMI 32(9),

2010

Root filter

Part filters

Deformation weights

Page 5: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Object hypothesis• Multiscale model: the resolution of part

filters is twice the resolution of the root

Page 6: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Scoring an object hypothesis• The score of a hypothesis is the sum of filter scores

minus the sum of deformation costs

),,,()(),...,( 22

0 10 ii

n

i

n

iiiiiin dydxdydxDpHFppscore

Filters

Subwindow features

Deformation weights

Displacements

Page 7: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Scoring an object hypothesis• The score of a hypothesis is the sum of filter scores

minus the sum of deformation costs

• Recall: pictorial structures

i ji

jiijiin lldlmllE,

1 ),()(),...,(

Matching cost

Deformation cost

),,,()(),...,( 22

0 10 ii

n

i

n

iiiiiin dydxdydxDpHFppscore

Filters

Subwindow features

Deformation weights

Displacements

Page 8: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Scoring an object hypothesis• The score of a hypothesis is the sum of filter scores

minus the sum of deformation costs

)()( zHwzscore

Concatenation of filter and deformation

weights

Concatenation of subwindow features and displacements

),,,()(),...,( 22

0 10 ii

n

i

n

iiiiiin dydxdydxDpHFppscore

Filters

Subwindow features

Deformation weights

Displacements

Page 9: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Detection• Define the score of each root filter location as the

score given the best part placements:

),...,(max)( 0,...,

01

npp

ppscorepscoren

Page 10: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Detection• Define the score of each root filter location as the

score given the best part placements:

• Efficient computation: generalized distance transforms• For each “default” part location, find the best-

scoring displacement

),...,(max)( 0,...,

01

npp

ppscorepscoren

),,,(),(max),( 22

,dydxdydxDdyydxxHFyxR ii

dydxi

Head filterHead filter responsesDistance transform

Page 11: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Detection

Page 12: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Matching result

Page 13: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Training• Training data consists of images with labeled

bounding boxes• Need to learn the filters and deformation parameters

Page 14: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Training• Our classifier has the form

• w are model parameters, z are latent hypotheses

• Latent SVM training:• Initialize w and iterate:

• Fix w and find the best z for each training example (detection)• Fix z and solve for w (standard SVM training)

• Issue: too many negative examples• Do “data mining” to find “hard” negatives

),(max)( zxHwxf z

Page 15: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Car model

Component 1

Component 2

Page 16: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Car detections

Page 17: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Person model

Page 18: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Person detections

Page 19: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Cat model

Page 20: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Cat detections

Page 21: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Bottle model

Page 22: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

More detections

Page 23: More sliding window detection: Discriminative part-based models Many slides based on P. FelzenszwalbP. Felzenszwalb.

Quantitative results (PASCAL 2008)

• 7 systems competed in the 2008 challenge• Out of 20 classes, first place in 7 classes and

second place in 8 classes

Bicycles Person Bird

Proposed approach Proposed approach

Proposed approach