Awesome

Zero-shot Single Image Restoration through Controlled Perturbation of Koschmieder's Model

This is the PyTorch implementation for our CVPR 2021 paper:

**Aupendu Kar, Sobhan Kanti Dhara, Debashis Sen, Prabir Kumar Biswas. Zero-shot Single Image Restoration through Controlled Perturbation of Koschmieder's Model. [Project Website] [PAPER]

Dependencies

• Python 3.6
• imageio==2.6.1
• numpy==1.17.4
• torch==1.8.0+cu111
• torchvision==0.9.0+cu111
• tqdm==4.40.2

Test Datasets and Results of Our Algorithm

All the processed datasets that are used in this paper and the output images of the proposed algorithm are given below:

1. Image Dehazing (Table 1 of Main Paper) Google Drive
2. Underwater Image Enhancement (Table 2 of Main Paper) Google Drive
3. Lowlight Image Enhancement (Table 3 of Main Paper) Google Drive

Test Codes

1. Run HazeZeroShot.py to perform single image dehazing
2. Run UnderWaterZeroShot.py to perform underwater image enhancement
3. Run LowLightZeroShot.py to perform lowlight image enhancement

• Give test image set path through --TestFolderPath argument
• Give the path of results through --SavePath argument

Citation

@InProceedings{Kar_2021_CVPR,
  author = {Kar, Aupendu and Dhara, Sobhan Kanti and Sen, Debashis and Biswas, Prabir Kumar},
  title = {Zero-shot Single Image Restoration through Controlled Perturbation of Koschmieder’s Model},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  month = {June},
  year = {2021},
  pages = {16205-16215}
}

Contact

Aupendu Kar: mailtoaupendu[at]gmail[dot]com