Weakly Supervised Object Localization with Progressive Domain Adaptation (CVPR 2016)

This is the research code for the paper:

Dong Li, Jia-Bin Huang, Yali Li, Shengjin Wang, and Ming-Hsuan Yang. "Weakly Supervised Object Localization with Progressive Domain Adaptation" In Proceedings of IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016

Project page

Citation

If you find the code and pre-trained models useful in your research, please consider citing:

@inproceedings{Huang-CVPR-2016,
  author  = {Dong, Li and Huang, Jia-Bin and Li, Yali and Wang, Shengjin and Yang, Ming-Hsuan},
  title   = {Weakly Supervised Object Localization with Progressive Domain Adaptation},
  booktitle = {Proceedings of the IEEE  Conference on Computer Vision and Pattern Recognition)},
  year    = {2015},
  volume  = {},
  number  = {},
  pages   = {}  
  }

System Requirements

• MATLAB (tested with R2014a on 64-bit Linux)
• Caffe

Installation

1. Download and unzip the project code.

2. Install Caffe. Please follow the Caffe installation instructions to install dependencies and then compile Caffe:

   # We call the root directory of the project code `WSL_ROOT`.
   cd $WSL_ROOT/caffe-wsl
   make all -j8
   make pycaffe
   make matcaffe
   

3. Download the PASCAL VOC 2007 dataset. Extract all the tars into one directory named VOCdevkit. It should have this basic structure:

   $VOCdevkit/                           # development kit
   $VOCdevkit/VOCcode/                   # VOC utility code
   $VOCdevkit/VOC2007                    # image sets, annotations, etc.
   # ... and several other directories ...
   # Then create symlinks for the dataset:
   cd $WSL_ROOT/data
   ln -s $VOCdevkit VOCdevkit2007
   

4. Download the pre-trained ImageNet model and put it into $WSL_ROOT/data/imagenet_models.

5. Download the pre-computed EdgeBox proposals and put them into $WSL_ROOT/data/edgebox_data.

   • voc_2007_trainval
   • voc_2007_test

6. Install the project.

   cd $WSL_ROOT
   # Start MATLAB
   matlab
   >> startup
   

Usage

You will need about 150GB of disk space free for the feature cache (which is stored in $WSL_ROOT/cache by default. The final adapted model will be stored in $WSL_ROOT/output/default/voc_2007_trainval.

1. Classification adaptation.

   >> prepare_for_cls_adapt
   cd $WSL_ROOT
   sh cls_adapt.sh
   

2. Class-specific proposal mining.

   >> maskout
   

3. MIL for confident proposal mining.

   >> mil
   

4. Detection adaptation.

   >> prepare_for_det_adapt
   cd $WSL_ROOT
   sh det_adapt.sh
   

5. Evaluation.

   cd $WSL_ROOT
   sh test.sh