Download BFR128 and BFR512, and put them in the appropriate directory according to your needs. The two datasets provide
Download the pre-trained models of STUNet and put them in STUNet/check_points/
. Next, modify the path of datasets and pre-trained models in test.py
. Last, use python test.py
to generate results.
You can also train our STUNet by yourself. The training configurations is stored in STUNet/options/opt.json
. Modify the configuration file according to your needs. Then use python train.py
to train the STUNet.
DFDNet: https://github.com/csxmli2016/DFDNet
HiFaceGAN: https://github.com/chaofengc/PSFRGAN
PSFRGAN: https://github.com/Lotayou/Face-Renovation
GFPGAN: https://github.com/TencentARC/GFPGAN
GPEN: https://github.com/yangxy/GPEN
DFDNet: https://github.com/csxmli2016/DFDNet (We use the pre-train model provided by the author.)
HiFaceGAN: https://pan.baidu.com/s/1Yiof155wF1GUNOuw2wUWXw (4kya)
PSFRGAN: https://pan.baidu.com/s/1GgTOc1FMF34b1Nf0_9rlGg (4kya)
GFPGAN: https://pan.baidu.com/s/1E7yt_FLLZghMJFYO8cSj9g (4kya)
GPEN: https://pan.baidu.com/s/1mySckCwIKIUGJbyhW28Idw (4kya)
STUNet: https://pan.baidu.com/s/16E4cqM7pbnh9w236l7_79Q (4kya)
DFDNet: https://pan.baidu.com/s/1I2UysJ5vDfwGMvRgVVSqzg (exb4)
HiFaceGAN: https://pan.baidu.com/s/17tTcEYdy6AGNh9UZCQ4CiA (exb4)
PSFRGAN: https://pan.baidu.com/s/1fzcnJote018v6g3Tn1-URw (exb4)
GFPGAN: https://pan.baidu.com/s/1_P8OMwGaDyCe7H0G2a9fjA (exb4)
GPEN: https://pan.baidu.com/s/19OaTwlqvJlgOc_sIFESxpw (exb4)
STUNet: https://pan.baidu.com/s/1Mi-TlCmnFXgvY_jk17Tzeg (exb4)
Our codes are heavily based on SwinIR. We also borrow some codes from HiFaceGAN and Restormer.
If you think this work is useful for your research, please cite the following papers.
@inproceedings{zhang2022blind,
title={Blind Face Restoration: Benchmark Datasets and a Baseline Model},
author={Zhang, Puyang and Zhang, Kaihao and Luo, Wenhan and Li, Changsheng and Wang, Guoren},
booktitle={arXiv:2206.03697},
year={2022}
}
@inproceedings{zhang2022edface,
title={EDFace-Celeb-1M: Benchmarking Face Hallucination with a Million-scale Dataset},
author={Zhang, Kaihao and Li, Dongxu and Luo, Wenhan and Liu, Jingyu and Deng, Jiankang and Liu, Wei and Stefanos Zafeiriou},
booktitle={IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI)},
year={2022}
}