Discriminatively Trained Multiscale Deformable Part Models

Version 2
This is an old release
The latest release is available here

Below is an implementation of our object detection system based on discriminatively trained deformable part models.

This is exactly the system we used in the 2008 PASCAL VOC Challenge.
It extends the work in [1] as briefly described in [2]. A forthcoming paper will describe the new system in full detail. The current release contains the basic object detection code and models trained on several PASCAL datasets. The model learning code and the context rescoring code will be available soon.

[1] P. Felzenszwalb, D. McAllester, D. Ramanan
A Discriminatively Trained, Multiscale, Deformable Part Model
Proceedings of the IEEE CVPR 2008 pdf

[2] P. Felzenszwalb, R. Girshick, D. McAllester, D. Ramanan
Discriminatively Trained Mixtures of Deformable Part Models
PASCAL VOC Challenge 2008 pdf
presentation slides

The system is implemented in matlab, with a few helper functions implemented in C++ for efficiency reasons.
The README file has instructions on how to compile and use the system.

This software is released under the MIT License.
To download, click here: voc-release2.tgz (updated on 10/04/08)

This project is supported by the National Science Foundation under Grant No. 0534820 and 0746569.


Examples:


Recognition results (AP)

The models included with the source code were trained on the train+val dataset from each year and evaluated on the corresponding test dataset.
This is exactly the protocol of the "comp3" competition.

2006 data bicycle bus car cat cow dog horse mbike person sheep
with context 0.653 0.544 0.636 0.216 0.428 0.167 0.442 0.621 0.401 0.446
without context 0.643 0.525 0.641 0.180 0.408 0.145 0.389 0.584 0.382 0.412

2007 data aero bicycle bird boat bottle bus car cat chair cow table dog horse mbike person plant sheep sofa train tv
with context 0.306 0.569 0.022 0.155 0.279 0.399 0.479 0.177 0.178 0.180 0.251 0.127 0.495 0.395 0.360 0.167 0.183 0.219 0.367 0.373
without context 0.278 0.561 0.014 0.146 0.257 0.383 0.470 0.151 0.164 0.167 0.231 0.110 0.444 0.373 0.355 0.140 0.169 0.193 0.319 0.373