This BasicSR project is released under the Apache 2.0 license.

• StyleGAN2
  • The codes are modified from the repository stylegan2-pytorch. Many thanks to the author - Kim Seonghyeon 😊 for translating from the official TensorFlow codes to PyTorch ones. Here is the license of stylegan2-pytorch.
  • The official repository is https://github.com/NVlabs/stylegan2, and here is the NVIDIA license.
• DFDNet
  • The codes are largely modified from the repository DFDNet. Their license is Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
• DiffJPEG
  • Modified from https://github.com/mlomnitz/DiffJPEG.
• pytorch-image-models
  • We use the implementation of DropPath and trunc_normal_ from pytorch-image-models. The LICENSE is included as LICENSE_pytorch-image-models.
• DISTS
  • The loss implementation of DISTS is from DISTS. The LICENSE is MIT.
• SwinIR
  • The arch implementation of SwinIR is from SwinIR. The LICENSE is included as LICENSE_SwinIR.
• ECBSR
  • The arch implementation of ECBSR is from ECBSR. The LICENSE of ECBSR is Apache License 2.0
• Omnisr
  • The arch implementation of Omnisr is from Omnisr. The LICENSE of Omnisr is [Apache License 2.0]. The LICENSE is included as LICENSE_Omnisr.
• DAT
  • The arch implementation of DAT is from DAT. The LICENSE is Apache License 2.0.
• SPAN
  • The arch implementation of SPAN is from SPAN with modifications by neosr. The LICENSE is Apache 2.0.
• PLKSR
  • The arch implementation of PLKSR is from PLKSR with modifications by neosr. The LICENSE is MIT.

1. NIQE metric: the codes are translated from the official MATLAB codes

   A. Mittal, R. Soundararajan and A. C. Bovik, "Making a Completely Blind Image Quality Analyzer", IEEE Signal Processing Letters, 2012.

2. FID metric: the codes are modified from pytorch-fid and stylegan2-pytorch.