A Discriminatively Trained, Multiscale, Deformable Part Model

1
A Discriminatively Trained, Multiscale,
Deformable Part Model

• ??? ??

2
??

• ????
• ????
• ????
• ?????
• ???????
• ???????
• ??????
• ??????
• ?????????
• ????
• HOG????
• ???
• ????
• ??

3
????

• Pedro Felzenszwalb
• Associate Professor at Department of Computer
  Science, University of Chicago
• Courses AI, CV, statistical model
• Professional activitiesAC (ICCV, CVPR, ECCV),
  AE(PAMI), EB(IJCV)
• David McAllester
• Professor and Chief Academic Officer, Toyota
  Technological Institute at Chicago
• AAAI, fellow
• Collaborated with Peter Bartlett, Robert E.
  Schapire
• Deva Ramanan
• Assistant Professor, Department of Computer
  Science , University of California at Irvine
• PhD from UC Berkeley, advised by David Forsyth
• Marr Prize 2009Discriminative models for
  multi-class object layout

4
??

• ????
• ????
• ????
• ?????
• HOG????
• ???????
• ???????
• ??????
• ??????
• ?????????
• ????
• ???
• ????
• ??

5
????

• Object Detection with Discriminatively Trained
  Part Based Models. To appear in PAMI.
• ???????,(1)SVM????(2)mixtured models(3)?????(4)???
  
• Cascade Object Detection with Deformable Part
  Models. In CVPR,2010.
• ???????????????????????
• Discriminative models for multi-class object. In
  ICCV, 2009.
• ????????,?????????????????????
• Distance Transforms of Sampled Functions. In TR,
  2004.
• ?????????????????,?????????????,????????????????,?
  ?????????

6
??

• ????
• ????
• ????
• ?????
• ???????
• ???????
• ??????
• ??????
• ?????????
• ????
• HOG????
• ???
• ????
• ??

7
????(1/2)

• 1. This paper describes a discriminatively
  trained, multiscale, deformable part model for
  object detection. Our system achieves a two-fold
  improvement in average precision over the best
  performance in the 2006 PASCAL person detection
  challenge. It also outperforms the best results
  in the 2007 challenge in ten out of twenty
  categories. .
• 2. The system relies heavily on deformable
  parts. While deformable part models have become
  quite popular, their value had not been
  demonstrated on difficult benchmarks such as the
  PASCAL challenge. Our system also relies heavily
  on new methods for discriminative training. We
  combine a margin-sensitive approach for data
  mining hard negative examples with a formalism we
  call latent SVM. A latent SVM, like a hidden CRF,
  leads to a non-convex training problem. However,
  a latent SVM is semi-convex and the training
  problem becomes convex once latent information is
  specified for the positive examples.
  .
• 3. We believe that our training methods will
  eventually make possible the effective use of
  more latent information such as hierarchical
  (grammar) models and models involving latent
  three dimensional pose. .

8
????(2/2)

• 1. ?????????????????????????????????????VOC20
  06????????????20??VOC2007?????20???10????????????
  ???????
  .
• 2. ???????????????????????????,????????PASCAL
  ??????????????????????????????????????????????(mar
  gin-sensitive)????????latent SVM??????????????????
  Latent SVM?????????,???????????????latent
  SVM???????,???latent???????????,????????????
  
  .
• 3. ??????????????????????????latent???????,??
  ??(??)??,??latent??????
  .

9
??

• ????
• ????
• ????
• ?????
• ???????
• ???????
• ??????
• ??????
• ?????????
• ????
• HOG????
• ???
• ????
• ??

10
???????(1/3)

• ?????

11
???????(2/3)

• ????

???????????????,???????,????????????????,?????????
???????
12
???????(3/3)

• ??????

???????,????HOG?????????,????????
13
???????(1/2)

• ????,?????????

????
14
???????(2/2)
15
??????(1/3)

• Latent SVM

16
??????(2/3)

• ??????(?)

17
??????(3/3)

• ??????(?)

18
??????(1/3)

• Latent SVM????

19
??????(2/3)

• ??coordinate decent??

20
??????(3/3)

• SVM?????
• Sequential Minimization Optimization(libSVM)
• Stochastic gradient descent

21
?????????
22
????(1/2)
23
????(2/2)
24
HOG????(1/4)

• ???????
• ??x,y?????,??????9???
• ????(spatial aggregation)
• ?????
• ???
• ????block???cell??????
• ??
• ??????????????0

25
HOG????(2/4)

• PCA??(3613,PAMI???)
• ?11??????????36??????
• ???????,?????11?????
• ?????????????????????,???9????????????????HOG????,
  ???????9?????????4???,????9????4?cell?????,??4?cel
  l??9????????

26
HOG????(3/4)

• ????(1331,PAMI???)
• ?????9???,??????????????18???,??PCA???????????(??
  ???????18???)????4x9-gt4913,????4x27-gt42731???,
  ????????13?,27????4?cell??????4?cell?27????????

27
HOG????(4/4)

• ????(PAMI???)
• ???????????????????????????????????(sin?cos),?????
  ???????????????????,?????????
• ??????????(FFT)????????????,??dd????????????,???
  ???k,??
  ????????????????????????????
• ???????????????,??????????????,?????????????????b
  in,????4?cell????

28
???

• Bounding box??
• ??????
• ????,?????????????
• ?????

29
????(1/2)

• Mixture models???????

30
????(2/2)

• ?VOC????????

2006
2007
2008
31
??

• ????
• ????
• ????
• ?????
• HOG????
• ???????
• ???????
• ??????
• ??????
• ?????????
• ????
• ???
• ????
• ??

32
??

• ????
• ?????????????
• ?????????????
• ????????????
• ????
• ??????????????????
• ?mixture model?????????
• ????????????(mixture model)

33
?????????
34
Distance Transforms of Sampled Functions

• ????????????????????

??
35
Cascade Object Detection with Deformable Part
Models

• ?CVPR2008??,????????????????
• ????cascade??,??????????,??Probability
  Approximately Admissible ????

36
??!