DAEFR

KO | EN

gitlite — search

#php #java #python #javascript #ruby #shell #html #jquery #linux #nodejs #r #ios

★ 59 Open GitHub ↗

[ICLR 2024] DAEFR: Dual Associated Encoder for Face Restoration

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# News  ✨ ✨ 

- **[2024-01-16]** Our paper is accepted by **ICLR 2024**.
- **[2023-12-26]** Create project.

# DAEFR

This repo includes the source code of the paper: "[Dual Associated Encoder for Face Restoration](https://arxiv.org/abs/2308.07314)" by [Yu-Ju Tsai](https://liagm.github.io/), [Yu-Lun Liu](https://yulunalexliu.github.io/), [Lu Qi](https://luqi.info/), [Kelvin C.K. Chan](https://ckkelvinchan.github.io/), and [Ming-Hsuan Yang](https://faculty.ucmerced.edu/mhyang/).

![](https://raw.githubusercontent.com/LIAGM/DAEFR/master/assets/DAEFR_results.png)

We propose a novel dual-branch framework named **DAEFR**. Our method introduces an auxiliary LQ branch that extracts crucial information from the LQ inputs. Additionally, we incorporate association training to promote effective synergy between the two branches, enhancing code prediction and output quality. We evaluate the effectiveness of DAEFR on both synthetic and real-world datasets, demonstrating its superior performance in restoring facial details.

![](https://raw.githubusercontent.com/LIAGM/DAEFR/master/assets/DAEFR_arch.png)

## Environment

- python>=3.8
- pytorch>=1.10.0
- pytorch-lightning==1.0.8
- omegaconf==2.0.0
- basicsr==1.3.3.4

Please properly set up the environment by the following command:
```shell
conda create -n DAEFR python=3.8 -y
conda activate DAEFR
pip install torch==1.10.1+cu113 torchvision==0.11.2+cu113 --extra-index-url https://download.pytorch.org/whl/cu113
pip install -r requirements.txt
```

**Warning:** Different versions of pytorch-lightning and omegaconf may lead to errors or different results.

## Preparations of dataset and models

**Training Dataset**: 
- Training data: **HQ Codebook**, **LQ Codebook** and **DAEFR** are trained with **FFHQ** which attained from [FFHQ repository](https://github.com/NVlabs/ffhq-dataset). 
- The original size of the images in FFHQ are 1024x1024. We resize them to 512x512 with bilinear interpolation in our work. 
- We provide our resized 512x512 FFHQ on [HuggingFace](https://huggingface.co/datasets/LIAGM/FFHQ_datasets/tree/main). 
- Link this 512x512 version dataset to ```./datasets/FFHQ/image512x512.```
- If you use our FFHQ dataset, please change the filename or change the code in the line 43 to ```f'{v:05d}.png'``` in 
```/Your/conda/envs/DAEFR/lib/python3.8/site-packages/basicsr/data/ffhq_dataset.py```

**Testing Dataset**:

Please put the following datasets in the ```./dataset/``` folder.

<table>
<tr>
    <th>Datasets</th>
    <th>Short Description</th>
    <th>Download</th>
    <th>DAEFR results</th>
</tr>
<tr>
    <td>CelebA-Test (HQ)</td>
    <td>3000 (HQ) ground truth images for evaluation</td>
    <td><a href="https://huggingface.co/datasets/LIAGM/DAEFR_test_datasets/blob/main/celeba_512_validation.zip">celeba_512_validation.zip</a></td>
    <td>None</td>
</tr>
<tr>
    <td>CelebA-Test (LQ)</td>
    <td>3000 (LQ) synthetic images for testing</td>
    <td><a href="https://huggingface.co/datasets/LIAGM/DAEFR_test_datasets/blob/main/self_celeba_512_v2.zip">self_celeba_512_v2.zip</a></td>
    <td><a href="https://huggingface.co/datasets/LIAGM/DAEFR_results/blob/main/DAEFR_celeba.zip">Link</a></td>
</tr>  
<tr>
    <td>LFW-Test (LQ)</td>
    <td>1711 real-world images for testing</td>
    <td><a href="https://huggingface.co/datasets/LIAGM/DAEFR_test_datasets/blob/main/lfw_cropped_faces.zip">lfw_cropped_faces.zip</a></td>
    <td><a href="https://huggingface.co/datasets/LIAGM/DAEFR_results/blob/main/DAEFR_lfw.zip">Link</a></td>
</tr>
<tr>
    <td>WIDER-Test (LQ)</td>
    <td>970 real-world images for testing</td>
    <td><a href="https://huggingface.co/datasets/LIAGM/DAEFR_test_datasets/blob/main/Wider-Test.zip">Wider-Test.zip</a></td>
    <td><a href="https://huggingface.co/datasets/LIAGM/DAEFR_results/blob/main/DAEFR_wider.zip">Link</a></td>
</tr>
</table>

---

**Model**:

Pretrained models used for training and the trained model of our DAEFR can be attained from [HuggingFace](https://huggingface.co/LIAGM/DAEFR_pretrain_model/tree/main). 
Link these models to ```./experiments```.

You can use the following command to download the model:
```shell
python download_model_from_huggingface.py
```

```

## Test
If you put the model and dataset folder properly, you can use the following command:
```shell
bash run_test_and_evaluation.sh
```
to obtain the results.

Or you can modify the following the following script:

```shell
sh scripts/test.sh
```

Or you can use the following command for testing:
```shell
CUDA_VISIBLE_DEVICES=$GPU python -u scripts/test.py \
--outdir $outdir \
-r $checkpoint \
-c $config \
--test_path $align_test_path \
--aligned
```

## Training
### First stage for codebooks
```shell
sh scripts/run_HQ_codebook_training.sh
sh scripts/run_LQ_codebook_training.sh
```
### Second stage for Association
```shell
sh scripts/run_association_stage_training.sh
```
### Final stage for DAEFR
```shell
sh scripts/run_DAEFR_training.sh
```

**Note**. 
- Please modify the related paths to your own.
- If you encounter problems with FFHQ dataset names during training, please change the filename or change the code in the line 43 to ```f'{v:05d}.png'``` in 
```/Your/conda/envs/DAEFR/lib/python3.8/site-packages/basicsr/data/ffhq_dataset.py```
- The second stage is for model association. You need to add your trained HQ\_Codebook and LQ\_Codebook model to `ckpt_path_HQ` and `ckpt_path_LQ` in ```config/Association_stage.yaml```.
- The final stage is for face restoration. You need to add your trained HQ\_Codebook and Association model to `ckpt_path_HQ` and `ckpt_path_LQ` in ```config/DAEFR.yaml```.
- Our model is trained with 8 A100 40GB GPUs with batchsize 4.

## <a id="metrics">Metrics</a>
```shell
sh scripts/metrics/run.sh
```

**Note**. 
- You need to add the path of CelebA-Test dataset in the script if you want get IDA, PSNR, SSIM, LPIPS. You also need to modify the name of restored folders for evaluation.
- For LMD and NIQE, we use the evaluation code from [VQFR](https://github.com/TencentARC/VQFR). Please refer to their repo for more details.

## Citation
    @inproceedings{tsai2024dual,
        title={Dual Associated Encoder for Face Restoration},
        author={Tsai, Yu-Ju and Liu, Yu-Lun and Qi, Lu and Chan, Kelvin CK and Yang, Ming-Hsuan},
        booktitle={The Twelfth International Conference on Learning Representations},
        year={2024}
    }

## Acknowledgement
We thank everyone who makes their code and models available, especially [Taming Transformer](https://github.com/CompVis/taming-transformers), [basicsr](https://github.com/XPixelGroup/BasicSR), [RestoreFormer](https://github.com/wzhouxiff/RestoreFormer), [CodeFormer](https://github.com/sczhou/CodeFormer), and [VQFR](https://github.com/TencentARC/VQFR).

## Contact
For any question, feel free to email `louis19950117@gmail.com`.

DAEFR

Explore Similar Repositories

Was this content helpful?

Recent Feedback

Download README