2D articulated human pose estimation software v1.22
Marcin Eichner,
Manuel J. Marín-Jiménez,
Andrew Zisserman,
Vittorio Ferrari
Overview
We release here software for articulated human pose estimation in still images. Our algorithm [4] is designed to operate in uncontrolled images with difficult illumination conditions and cluttered backgrounds. People can appear at any location and scale in the image, and can wear any kind of clothing, in any color/texture. The only assumption the algorithm makes is that people are upright (i.e. their head is above their torso) and they are seen approximately from a frontal viewpoint.
The input is an image and a bounding-box around the head and shoulders of a person in the image. The output of our system is a set of line segments indicating location, size and orientation of the body parts.
This release can be used in conjunction with our new upper-body detector to give a fully automatic pipeline taking just an image as input.
Our software is provided for free to all users as long as they do not sell any part of it or include part of it in a commercial product. Contact us to negotiate an agreement for using the software for commercial purposes.
new in v1.22:
- code release updated to work with the latest matlab releases (R2012a, R2012b)
new in v1.21:
- system updated to [5]
- full body parsing support added
- repulsive model removed
- alternative approximate MAP stickman output added
Example of results
The next figure shows some examples of pose estimation results on Buffy the Vampire Slayer and ETHZ Pascal Stickmen datasets. Click on an image to enlarge it.
 |
 |
 |
 |
 |
 |
 |
 |
More examples  | |||
Perona November 2009 Challenge results  | |||
Digest of Ben Packer's results  | |||
| DISCLAIMER: results shown here may not be up to date and are not guaranteed to be exactly reproducible | |||
Performance
The performance of our software has been evaluated using two challenging datasets. Please follow the links for more information:
Training data
The parameters of our model [4] requires a training stage. We provide here 3 sets of parameters trained from different datasets. See the README file for details.
Downloads
| Filename | Description | Size |
|---|---|---|
| pose_estimation_code_release_v1.22.tgz | complete pose estimation system | 1259 kB |
| README.html | description of contents | 37 kB |
| LICENSE.txt | license | 1 kB |
Related Publications
[1] Ferrari, V., Marin-Jimenez, M. and Zisserman, A.
Progressive Search Space Reduction for Human Pose Estimation
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2008)
Bibtex source
|
Abstract
|
Document: ps.gz PDF
[2] Ferrari, V. and Marin-Jimenez, M. and Zisserman, A.
2D Human Pose Estimation in TV Shows
Proceedings of the Dagstuhl Seminar on Stastistical and Geometrical Approaches to Visual Motion Analysis, 2009.
Document: PDF
[3] Ferrari, V. and Marin-Jimenez, M. and Zisserman, A.
Pose search: retrieving people using their pose
Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2009)
Bibtex source
|
Abstract
|
Document: ps.gz PDF
[4] Eichner, M. and Ferrari, V.
Better Appearance Models for Pictorial Structures
Proceedings of British Machine Vision Conference (BMVC), 2009.
Document: PDF
[5] Eichner, M. and Marin-Jimenez, M. and Zisserman, A. and Ferrari, V.
Articulated Human Pose Estimation and Search in (Almost) Unconstrained Still Images
ETH Zurich, D-ITET, BIWI, Technical Report No.272, September 2010.
Document: PDF
Acknowledgements
We would like to thank Deva Ramanan, Varun Gulshan, Pushmeet Kohli and Vladimir Kolmogorov, who contributed to this code release.
This work is funded by the EU Project CLASS, the Swiss National Science Foundation SNSF and the Spanish Ministry of Education and Science (under FPU grant) and ERC Project VisRec
