Real Cascade U-Nets for Anime Image Super Resolution

English | Simplified Chinese

Real-CUGAN is an AI super resolution model for anime images, trained in a million scale anime dataset, using the same architecture as Waifu2x-CUNet. It supports 2x, 3x, 4x super resolving. For different enhancement strength, now 2x Real-CUGAN supports 5 model weights, 3x and 4x Real-CUGAN supports 3 model weights.

How to run

Download models and extract weights from archive to weights_v3/ folder

https://github.com/ZHCSOFT/Real-CUGAN/releases/download/Real-CUGAN/weights_v3.zip

Copy image files to pending/ for further processing, support .JPEG, .JPG, .PNG, .BMP file format. Some png format related issue will be fixed in future

Run python run.py following hints it prints

Output files will be saved to finish/ folder

Environment required

torch>=1.0.0
numpy
opencv-python

Sample (L: origin, R: up4x-latest-no-denoise)