Active Deformable Part Models is an active approach for part-based object detection, which optimizes the order of part filter evaluations and the time at which to stop and make a prediction. During inference, a learned policy is used as a look-up table to choose the part order and the stopping time based on the observed filter responses.
If you are using the code, please cite the followings:
Active Deformable Part Models Inference, Menglong Zhu, Nikolay Atanasov, George Pappas and Kostas Daniilidis, European Conference on Computer Vision (ECCV) 2014. Active Deformable Part Models, Menglong Zhu, Nikolay Atanasov, George Pappas and Kostas Daniilidis, arXiv:1404.0334 [cs.CV]
To run the demo, simply run the following in MATLAB:
The code assumes mex is setup in the system. If not, the pre-compiled Linux 64-bit and MacOSX 64-bit mex binaries are already provided under ./bin folder.
Note: You can comment out
at line 21 in
demo_pca_policy.m, if having trouble
compiling the code on your Linux/Mac system.
The demo will show the top ADPM detection of person and car in the example images.
Standalone API for detecting a class of object is provided as the following function call:
>> ds = detect_object(image_file, object_class, display)
where the input/output arguments are
image_file: image file path object_class: object class name display: boolean flag of displaying top detection ds: detection bounding boxes with scores
Our implementation is based on the publicly available original DPM and Cascade-DPM code. Main detection code is in ./src and ./mex folder. The main detection process is implemented in
Files under ./external are a subset of publicly available DPM code, refer to ./external/README for copyright.
Original VOC DPM models and learned policies are contained in ./models. Currently, person and car models are provided. More PASCAL VOC models will be available soon.
Copyright (C) 2013-2014 Menglong Zhu, Nikolay Atanasov, Samarth Brahmbhatt
This file is part of the Active Deformable Part Models code (http://cis.upenn.edu/~menglong/adpm.html) and is available under the terms of an MIT-like license provided in COPYING. Please retain this notice and COPYING if you use this file (or a portion of it) in your project.
Please direct all questions to:
Menglong Zhu firstname.lastname@example.org