j30206868 / gcfsr Goto Github PK

[Paper]   [Project Page]   [Demo]
Jingwen He, Wu Shi, Kai Chen, Lean Fu, Chao Dong
Bytedance, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shanghai AI Laboratory, Shanghai, China.

• 2022.11.21: Release inference code for blind face restoration.
• 2022.03.23: This repo is created.

• Python >= 3.7 (Recommend to use Anaconda or Miniconda)
• PyTorch >= 1.7
• Option: NVIDIA GPU + CUDA
• Option: Linux

1. Clone repo

   git clone https://github.com/hejingwenhejingwen/GCFSR
   cd GCFSR

2. Install dependent packages

   pip install -r requirements.txt
   python setup.py develop

Inference for blind face restoration

python inference_gcfsr_blind.py --model_path experiments/pretrained_models/gcfsr_blind_512.pth --input inputs/sample 

Inference (TODO)

python inference_gcfsr.py --model_path experiments/ --scale 32 --input inputs/sample --output outputs/tmp

We provide the training codes for GCFSR.

Procedures

1. Training dataset preparation: FFHQ

2. Validation dataset preparation: CelebA-val

python make_val_dataset.py --input datasets/celeba_val --output datasets/celeba_val_input

1. Modify the configuration file options/train_gcfsr.yml accordingly.

2. Training

python -m torch.distributed.launch --nproc_per_node=8 --master_port=22021 basicsr/train.py -opt options/train/train_gcfsr.yml --launcher pytorch

| gcfsr-512-blind | blind face restoration. | | gcfsr-1024 | Controllable face super resolution. |

@inproceedings{he2022gcfsr,
  title={GCFSR: a Generative and Controllable Face Super Resolution Method Without Facial and GAN Priors},
  author={He, Jingwen and Shi, Wu and Chen, Kai and Fu, Lean and Dong, Chao},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={1889--1898},
  year={2022}
}

If you have any question, please email [email protected].