A Discriminatively Trained, Multiscale, Deformable Part Model - - PowerPoint PPT Presentation

▶
a discriminatively trained multiscale deformable part
SMART_READER_LITE
LIVE PREVIEW

A Discriminatively Trained, Multiscale, Deformable Part Model - - PowerPoint PPT Presentation

Jan 18, 2024 122 likes •848 views

A Discriminatively Trained, Multiscale, Deformable Part Model February 24, 2016 Adam Allevato CS 381V University of Texas at Austin Outline Partial matching Non-maximum suppression Train image results Live demo Outline

slide-1
SLIDE 1

A Discriminatively Trained, Multiscale, Deformable Part Model

February 24, 2016 Adam Allevato CS 381V University of Texas at Austin

slide-2
SLIDE 2

Outline

  • Partial matching
  • Non-maximum suppression
  • Train image results
  • Live demo
slide-3
SLIDE 3

Outline

  • Partial matching
  • Non-maximum suppression
  • Train image results
  • Live demo
slide-4
SLIDE 4

Partial Matching

  • Deformable Part Models allows parts of objects

to shift around

  • What happens when one of the parts is

completely missing?

  • What happens when the images are hacked to

move parts of them around?

slide-5
SLIDE 5

Source Image

slide-6
SLIDE 6

Learned HOG Features from INRIA

slide-7
SLIDE 7

INRIA Person Dataset Matches

slide-8
SLIDE 8

Source Image

slide-9
SLIDE 9

Modified Source Image

slide-10
SLIDE 10

INRIA Person Dataset Matches

slide-11
SLIDE 11

Bad Background = Bad Detection

slide-12
SLIDE 12

Blocked Parts

  • Take the list of part filter responses in a

detection

  • One by one, replace their area with black pixels
  • Test intersection over union against ground

truth

slide-13
SLIDE 13

Source Image

slide-14
SLIDE 14

Detection

slide-15
SLIDE 15

1 Filter Blocked

slide-16
SLIDE 16

3 Filters Blocked

slide-17
SLIDE 17

Degradation (VOC 2010 Detector)

slide-18
SLIDE 18

Source Image

slide-19
SLIDE 19

0 Blocked Filters

slide-20
SLIDE 20

1 Blocked Filters

slide-21
SLIDE 21

2 Blocked Filters

slide-22
SLIDE 22

3 Blocked Filters

slide-23
SLIDE 23

Degradation (VOC 2007 Detector)

slide-24
SLIDE 24

Source Image

slide-25
SLIDE 25

0 Blocked Filters

slide-26
SLIDE 26

1 Blocked Filter

slide-27
SLIDE 27

2 Blocked Filters

slide-28
SLIDE 28

3 Blocked Filters

slide-29
SLIDE 29

Degradation (VOC 2007 Detector)

slide-30
SLIDE 30

Blocked Filters

  • DPM is great against this, especially with

canonical views

  • Shows robustness to occlusion
slide-31
SLIDE 31

Random Window Shifts

  • Window is shifted by random amount
  • The pixels covered are moved to the gap left

behind

  • All pixel information is maintained
slide-32
SLIDE 32

VOC 2010 Bicycle Detector

slide-33
SLIDE 33

Ground Truth

slide-34
SLIDE 34

No Shifts

slide-35
SLIDE 35

One Shift

slide-36
SLIDE 36

Static Parts to the Rescue

slide-37
SLIDE 37

One Shift

slide-38
SLIDE 38

Two Shifts

slide-39
SLIDE 39

Three Shifts

slide-40
SLIDE 40

Four Shifts

slide-41
SLIDE 41

Does how far we shift affect performance?

Averaged across 30 trials!

slide-42
SLIDE 42

10 10-Pixel Shifts

Ground truth

slide-43
SLIDE 43

Does how many times we shift affect performance?

slide-44
SLIDE 44

Does how many times we shift affect performance?

slide-45
SLIDE 45

Window Shifts

  • DPM is robust to small number of window shifts

because some part filters still fire correctly

  • More shifts give worse performance
  • The shift distance does not have appreciable

effect on the detection score loss

slide-46
SLIDE 46

Partial Matching

  • DPM is robust to object parts moving around
  • It can also infer positions of hidden or missing
  • bject parts
  • Sometimes, IoU can actually increase with
  • cclusion
slide-47
SLIDE 47

Outline

  • Partial matching
  • Non-maximum suppression
  • Train image results
  • Live demo
slide-48
SLIDE 48

Size-Matched Image

slide-49
SLIDE 49

Without NMS, N = 10

slide-50
SLIDE 50

N = 4 N = 3 N = 44

Without NMS, N = 50

slide-51
SLIDE 51

With NMS, N = 3

slide-52
SLIDE 52

Overlap = |Bi ∩ Bj| / |Bj|

Better matches Worse matches

Bi Bj

slide-53
SLIDE 53

NMS Overlap

  • 30 closely correlated matches are detected

before the second person is detected

  • 42 matches before third person is detected
  • Repeated detections for similar objects rank

similarly

  • NMS helps highlight the weaker matches
  • Asymmetric overlap metric allows good windows

to subsume smaller windows that lie inside

slide-54
SLIDE 54

Non-Maximum Supression

  • Helps avoid duplicates
  • Also helps let the weaker data show itself when

a limit is imposed on the total number of matches

slide-55
SLIDE 55

Outline

  • Partial matching
  • Non-maximum suppression
  • Train image results
  • Live demo
slide-56
SLIDE 56

Chicago Elevated Train

slide-57
SLIDE 57

VOC 2007 Train Model

slide-58
SLIDE 58

VOC 2007 Train Results, N = 1

slide-59
SLIDE 59

Without NMS, N=30

slide-60
SLIDE 60

Without NMS, N=30

Bi Bj

  • Many different modes
  • Overall high confusion
  • Some lonesome
  • utliers
slide-61
SLIDE 61

Chicago Elevated Train

  • Most detected windows contain mostly train
  • No single canonical detection window - “lots of

trains”

  • No window captures the entire train
  • No learned DPM for “train” is long enough to

capture this shape

slide-62
SLIDE 62

Outline

  • Partial matching
  • Non-maximum suppression
  • Train image results
  • Live demo
slide-63
SLIDE 63

Live Demo

  • INRIA person dataset
  • VOC 2010 dataset - “chair”
  • Can we fool it?
slide-64
SLIDE 64

Summary

  • Tested matches with parts of objects missing
  • Surveyed non-max suppression effects
  • Results on train image: technically correct, but

still did not capture entire object

  • Girshick's library is mature and can be easily

integrated into live application

slide-65
SLIDE 65

References

  • A Discriminatively Trained, Multiscale, Deformable Part Model. P. Felzenszwalb, D. McAllester,
  • D. Ramanan. IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2008
  • Original code available on GitHub: https://github.com/rbgirshick/voc-dpm
  • My code available on GitHub: https://github.com/Kukanani/voc-dpm
  • Images

http://cdn.collider.com/wp-content/image- base/Movies/P/Princess_Bride/the_princess_bride_movie_image__1_.jpg http://www.planetizen.com/files/images/ChicagoEl.jpg http://www.brinoideas.xyz/wp-content/uploads/2015/11/open-design-living-room-ideas-with- black-drume-pendant-and-blue-sofa-and-unique-glass-coffee-table-and-lovely-black-white-area- rug-and-grey-cream-pouf-also-big-window.jpg http://i.telegraph.co.uk/multimedia/archive/01947/B084FX_1947399c.jpg http://images.glaciermedia.ca/polopoly_fs/1.1346352.1410102588!/fileImage/httpImage/image.j pg_gen/derivatives/landscape_563/10175643-1-jpg.jpg

slide-66
SLIDE 66
slide-67
SLIDE 67

Live Cam Examples

slide-68
SLIDE 68

Input Image

slide-69
SLIDE 69

VOC 2010 Person Detector

slide-70
SLIDE 70

VOC 2010 Person Detector

slide-71
SLIDE 71

Chair Detector

slide-72
SLIDE 72

Chair Detector