Bidirectional Feature Pyramid Network with Recurrent Attention Residual Modules for Shadow Detection
by Lei Zhu^, Zijun Deng^, Xiaowei Hu, Chi-Wing Fu, Xuemiao Xu, Jing Qin, and Pheng-Ann Heng (^ joint 1st authors)
This implementation is written by Zijun Deng at the South China University of Technology.
Citation
@inproceedings{zhu18b,
    author = {Zhu, Lei and Deng, Zijun and Hu, Xiaowei and Fu, Chi-Wing and Xu, Xuemiao and Qin, Jing and Heng, Pheng-Ann},
    title = {Bidirectional Feature Pyramid Network with Recurrent Attention Residual Modules for Shadow Detection},
    booktitle = {ECCV},
    year = {2018}
}
Shadow Maps
The results of shadow detection on SBU and UCF can be found at Google Drive.
Trained Model
You can download the trained model which is reported in our paper at Google Drive.
Requirement
- Python 2.7
- PyTorch 0.4.0
- torchvision
- numpy
- Cython
- pydensecrf (here to install)
Preparation
- Set the path of pretrained ResNeXt model in resnext/config.py
- Set the path of SBU dataset in config.py
The pretrained ResNeXt model is ported from the official torch version, using the convertor provided by clcarwin. You can directly download the pretrained model ported by me.
Usage
Training
- Run by
python train.py
Hyper-parameters of training were gathered at the beginning of train.py and you can conveniently change it as you need.
Training a model on a single GTX 1080Ti GPU takes about 40 minutes.
Testing
- Put the trained model in ckpt/BDRAR
- Run by
python infer.py