Xintao Wang et al., Tencent ARC Lab and Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences · Image restoration

Real-ESRGAN

Real-ESRGAN is an image and video restoration project from Xintao Wang, Liangbin Xie, Chao Dong and Ying Shan at Tencent ARC Lab and the Shenzhen Institutes of Advanced Technology. It extends ESRGAN into a practical blind super-resolution system trained with pure synthetic data. The repository includes models for general scenes (realesr-general-x4v3, with a -dn option to control denoising strength), anime images (RealESRGAN_x4plus_anime_6B) and anime video (AnimeVideo-v3). It also supports face enhancement through GFPGAN and finetuning on your own data. Inference runs through a Python script, online demos on Replicate, Colab and Hugging Face Spaces, or portable ncnn-vulkan executables for Intel, AMD and Nvidia GPUs. It is released under the BSD-3-Clause licence and has 37,405 stars. The latest release is v0.3.0 (20 September 2022), and the last push was on 6 August 2024.