PoseletsPoselets capture part of the pose from a given viewpoint                                          [Bourdev & Malik...
PoseletsExamples may differ visually but have common semantics                                        [Bourdev & Malik, IC...
PoseletsBut how are we going to create training examples of poselets?
How do we train a poselet for a  given pose configuration?
Finding correspondences at training timeGiven part of a human   How do we find a similarpose                    pose confi...
Finding correspondences at training time                  Left Shoulder                  Left HipWe use keypoints to annot...
Finding correspondences at training time          Residual Error
Training poselet classifiersResidual   0.15   0.20   0.10    0.85   0.15    0.35Error:1.   Given a seed patch2.   Find the...
Training poselet classifiers1.   Given a seed patch2.   Find the closest patch for every other person3.   Sort them by res...
Which poselets should we train?• Choose thousands of random windows,  generate poselet candidates, train linear SVMs• Sele...
Selecting a small set ofcomplementary poselets
Poselets websitehttp://eecs.berkeley.edu/~lbourdev/poselets   The set of published poselet papers   H3D data set + Matla...
Upcoming SlideShare
Loading in …5
×

Poselets: Body part detectors trained using 3D human pose annotations

1,546 views

Published on

Published in: Technology, Business
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
1,546
On SlideShare
0
From Embeds
0
Number of Embeds
2
Actions
Shares
0
Downloads
10
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide
  • So the idea is to have a large number of these parts, which we call poselets. So even difficult examples can be found because a few of these poselets can be triggered.But there is one problem: In the case face detector, someone had to manually crop out thousands of faces to be able to train a classifier. In our case we have unlimited number of poselets, so we cannot have such a manual step.
  • So the idea is to have a large number of these parts, which we call poselets. So even difficult examples can be found because a few of these poselets can be triggered.But there is one problem: In the case face detector, someone had to manually crop out thousands of faces to be able to train a classifier. In our case we have unlimited number of poselets, so we cannot have such a manual step.
  • So the idea is to have a large number of these parts, which we call poselets. So even difficult examples can be found because a few of these poselets can be triggered.But there is one problem: In the case face detector, someone had to manually crop out thousands of faces to be able to train a classifier. In our case we have unlimited number of poselets, so we cannot have such a manual step.
  • Poselets: Body part detectors trained using 3D human pose annotations

    1. 1. PoseletsPoselets capture part of the pose from a given viewpoint [Bourdev & Malik, ICCV09]
    2. 2. PoseletsExamples may differ visually but have common semantics [Bourdev & Malik, ICCV09]
    3. 3. PoseletsBut how are we going to create training examples of poselets?
    4. 4. How do we train a poselet for a given pose configuration?
    5. 5. Finding correspondences at training timeGiven part of a human How do we find a similarpose pose configuration in the training set?
    6. 6. Finding correspondences at training time Left Shoulder Left HipWe use keypoints to annotate the joints, eyes, nose, etc. of people
    7. 7. Finding correspondences at training time Residual Error
    8. 8. Training poselet classifiersResidual 0.15 0.20 0.10 0.85 0.15 0.35Error:1. Given a seed patch2. Find the closest patch for every other person3. Sort them by residual error4. Threshold them
    9. 9. Training poselet classifiers1. Given a seed patch2. Find the closest patch for every other person3. Sort them by residual error4. Threshold them5. Use them as positive training examples to train a linear SVM with HOG features
    10. 10. Which poselets should we train?• Choose thousands of random windows, generate poselet candidates, train linear SVMs• Select a small set of poselets that are: • Individually effective • Complementary
    11. 11. Selecting a small set ofcomplementary poselets
    12. 12. Poselets websitehttp://eecs.berkeley.edu/~lbourdev/poselets The set of published poselet papers H3D data set + Matlab tools Java3D annotation tool + video tutorial Matlab code to detect people using poselets Our latest trained poselets

    ×