Defocus Blur Detection via Depth Distillation

This repo contains the code and results of our ECCV 2020 paper:

Defocus Blur Detection via Depth Distillation
Xiaodong Cun and Chi-Man Pun^*
University of Macau

Models | Results | Paper | Supp. | Online Demo!(Google CoLab)

Results

we provide results on two datasets under different backbone(VGG19,ResNext101), please download from Google Drive

Pretrained Models

• VGG16 backbone: vgg_best.pth
• ResNeXt101 backbone: res_best.pth

Dependences

• PyTorch
• OpenCV
• scipy
• tqdm
• scikit-learn

Demos

Online Demo!(Google CoLab) is recommanded to evaluate the performance of our method.

Also, you can run a local jupyter server to evalute on CPU or GPU.

1. Download the pretrianed models and ResNeXt101 backbone and put it to pretrained.

2. Download the DUT500 dataset and put it to dataset

3. make sure all the path in paths.py are correct, the folder may like:

depth-distillation/
    - datasets/
        * DUTDefocus/
        * CUHKDefocus/
    - pretrained/
        * res_best.pth
        * vgg_best.pth
        * resnext_101_32x4d.pth
    - models/
    
    other files...

4. run the jupyter notebook to evaluate.

Acknowledgements

The author would like to thanks Nan Chen for her helpful discussion.

Part of the code is based upon Pytorch-GAN and Shadow Detection

Citation

If you find our work useful in your research, please consider citing:

@misc{cun2020defocus,
      title={Defocus Blur Detection via Depth Distillation}, 
      author={Xiaodong Cun and Chi-Man Pun},
      year={2020},
      eprint={2007.08113},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Contact

Please contact me if there is any question (Xiaodong Cun [email protected])