In depth learning series 46: face image super score gfp-gan

1. Introduce

GFP-GAN By Tencent ARC Produced by the laboratory , The test results are very good .
The model is used to recover high-quality faces from low-quality faces . These lower quality portraits may be degraded for a variety of reasons , Such as low resolution , The noise , Blurred or compressed .
The mainstream image restoration technology is still against the generation network GAN, But how to use it well GAN It's a learning . This model uses generative facial prior model （Generative Facial Prior, GFP）, The spatial feature transformation layer is incorporated into the face restoration process , This enables the method to achieve a good balance between authenticity and fidelity .
There are 3 individual pretrain Model , By default v1.3：

The general drawing of the model is as follows ：

1.1 Degradation module

De degenerate the module into a UNet, stay UNet The decoder outputs an image at each resolution scale , Use L1 The loss function supervises the de degradation module .

1.2 Pretrained GAN as prior

take styleGAN Take the generator of and use it directly ~

1.3 Add spatial features

With the loss function reconstructed $F_{spatial}$ It can better improve the fidelity , As in the previous $F_{GAN}$ Connect , The connection method is SFT.

1.4 Increase the loss of facial components 、 Loss of fidelity

Loss of fidelity use ArcFace Face recognition model .

2. Quick start

2.1 Address of the test

• https://huggingface.co/spaces/akhaliq/GFPGAN, Return only faces
• https://app.baseten.co/applications/Q04Lz0d/operator_views/8qZG6Bg,GPU backened, fast , And you can return to the full graph
• https://replicate.com/xinntao/gfpgan, Need to register , Return to full picture
• colab demo Address ：https://colab.research.google.com/drive/1sVsoBd9AjckIXThgtZhGrHRfFI6UUYOo

2.2 github Address

https://github.com/TencentARC/GFPGAN

git clone https://github.com/TencentARC/GFPGAN.git
cd GFPGAN
pip install basicsr #  For reasoning and training ,https://github.com/xinntao/BasicSR
pip install facexlib #  For face detection ,https://github.com/xinntao/facexlib
pip install -r requirements.txt
python setup.py develop
pip install realesrgan #  Non face region supersession , Use Real-ESRGAN

Then download the model , You can download it offline ：

wget https://github.com/TencentARC/GFPGAN/releases/download/v1.3.0/GFPGANv1.3.pth -P experiments/pretrained_models no-check-certificate

To carry out reasoning ：
python inference_gfpgan.py -i inputs/whole_imgs -o results -v 1.3 -s 2
The parameters are described as follows ：

Usage: python inference_gfpgan.py -i inputs/whole_imgs -o results -v 1.3 -s 2 [options]...

  -h                   show this help
  -i input             Input image or folder. Default: inputs/whole_imgs
  -o output            Output folder. Default: results
  -v version           GFPGAN model version. Option: 1 | 1.2 | 1.3. Default: 1.3
  -s upscale           The final upsampling scale of the image. Default: 2
  -bg_upsampler        background upsampler. Default: realesrgan
  -bg_tile             Tile size for background sampler, 0 for no tile during testing. Default: 400
  -suffix              Suffix of the restored faces
  -only_center_face    Only restore the center face
  -aligned             Input are aligned faces
  -ext                 Image extension. Options: auto | jpg | png, auto means using the same extension as inputs. Default: auto

If you use cpu Words , There are several areas that need to be modified ：

1. inference_gfpgan.py in , allow cpu, And will half Change it to False：
   
   Besides ,python3.7/site-packages/facexlib/utils Plus compatibility cpu Code for ：