:fire: RIDCP: Revitalizing Real Image Dehazing via High-Quality Codebook Priors (CVPR2023)

This is the official PyTorch codes for the paper.

RIDCP: Revitalizing Real Image Dehazing via High-Quality Codebook Priors
Ruiqi Wu, Zhengpeng Duan, Chunle Guo^*, [Zhi Chai](), Chongyi Li （ * indicates corresponding author)
The IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR), 2023

[Arxiv Paper] [中文版 (TBD)] [Website Page] [Dataset (pwd:qqqo)]

:rocket: Highlights:

• SOTA performance: The proposed RIDCP achieves superior performance in both qualitative and quantitative comparisons with SOTA methods.
• Adjustable: Our RIDCP can adjust the degree of enhancement manually.

  :page_facing_up: Todo-list

• Release code for VQGAN pre-training and CHM weights acquiring
• Website page
• Colab demo

Demo

Video examples

Dependencies and Installation

• Ubuntu >= 18.04
• CUDA >= 11.0
• Other required packages in requirements.txt

  
  # git clone this repository
  git clone https://github.com/RQ-Wu/RIDCP.git
  cd RIDCP

create new anaconda env

conda create -n ridcp python=3.8 conda activate ridcp

install python dependencies

pip install -r requirements.txt BASICSR_EXT=True python setup.py develop

## Get Started
### Prepare pretrained models & dataset
1. Downloading pretrained checkpoints
<table>
<thead>
<tr>
    <th>Model</th>
    <th> Description </th>
    <th>:link: Download Links </th>
</tr>
</thead>
<tbody>
<tr>
    <td>HQPs</td>
    <th>VQGAN pretrained on high-quality data.</th>
    <th rowspan="3">
    [<a href="https://github.com/RQ-Wu/RIDCP_dehazing/blob/master/">Google Drive (TBD)</a>] 
    [<a href="https://pan.baidu.com/s/1ps9dPmerWyXILxb6lkHihQ">Baidu Disk (pwd: huea)</a>]
    </th>
</tr>
<tr>
    <td>RIDCP</td>
    <th>Dehazing network trained on data generated by our pipeline.</th>
</tr>
<tr>
    <td>CHM</td>
    <th>Weight for adjusting controllable HQPs matching.</th>
</tr>
</tbody>
</table>

2. Preparing data for training
<table>
<thead>
<tr>
    <th>Dataset</th>
    <th> Description </th>
    <th>:link: Download Links </th>
</tr>
</thead>
<tbody>
<tr>
    <td>rgb_500</td>
    <th>500 clear RGB images as the input of our phenomenological degradation pipeline</th>
    <th rowspan="2">
    [<a href="https://github.com/RQ-Wu/RIDCP_dehazing/blob/master/">Google Drive (TBD)</a>] 
    [<a href="https://pan.baidu.com/s/1oX3AZkVlEa7S1sSO12r47Q">Baidu Disk (pwd: qqqo)</a>]
    </th>
</tr>
<tr>
    <td>depth_500</td>
    <th>Corresponding depth map generated by RA-Depth(https://github.com/hmhemu/RA-Depth).</th>
</tr>
<tr>
    <td>Flick2K, DIV2K</td>
    <th>High-quality data for VQGAN pre-training</th>
    <th>-</th>
</tr>
</tbody>
</table>

3. The final directory structure will be arranged as:

datasets |- clear_images_no_haze_no_dark_500 |- xxx.jpg |- ... |- depth_500 |- xxx.npy |- ... |- Flickr2K |- DIV2K

pretrained_models |- pretrained_HQPs.pth |- pretrained_RIDCP.pth |- weight_for_matching_dehazing_Flickr.pth

### Quick demo
Run demos to process the images in dir `./examples/` by following commands:

python inference_ridcp.py -i examples -w pretrained_models/pretrained_RIDCP.pth -o results --use weight --alpha -21.25

### Train RIDCP
Step 1: Pretrain a VQGAN on high-quality dataset

TBD

Step 2: Train our RIDCP

CUDA_VISIBLE_DEVICES=X,X,X,X python basicsr/train.py --opt options/RIDCP.yml

Step3: Adjust our RIDCP

TBD

## Citation
If you find our repo useful for your research, please cite us:

@inproceedings{wu2023ridcp, title={RIDCP: Revitalizing Real Image Dehazing via High-Quality Codebook Priors}, author={Wu, Ruiqi and Duan, Zhengpeng and Guo, Chunle and Chai, Zhi and Li, Chongyi}, booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition}, year={2023} }

## License
Licensed under a [Creative Commons Attribution-NonCommercial 4.0 International](https://creativecommons.org/licenses/by-nc/4.0/) for Non-commercial use only.
Any commercial use should get formal permission first.

## Acknowledgement
This repository is maintained by [Ruiqi Wu](https://rq-wu.github.io/).