Picture quality enhancer Real ESRGAN project 20220422 Wood of falling dust

The goal of Real ESRGAN is to develop Practical image inpainting algorithm 。
We use pure synthetic data to train on the basis of ESRGAN, so that it can be applied to the actual scene of image repair (as the name implies: Real ESRGAN).

Added a small model for animation video. For more information, see anime video models Medium
✅ Added ncnn implementation: Real-ESRGAN-ncnn-vulkan .
✅ Added RealESRGAN_x4plus_anime_6B.pth , optimized the binary image, and reduced the size of the model. Details and waifu2x Please check the comparison of anime_model.md
✅ Users can fine tune their own data: details
✅ Support use GFPGAN Enhance face
✅ Via Gradio Added to Huggingface Spaces (An online platform for machine learning applications): Graphio Online 。 thank @AK391
✅ Support any scaling: --outscale (actually used LANCZOS4 To further adjust the size of the output image). Added RealESRGAN_x2plus.pth Model
✅ Infer script Support: 1) Block processing tile ; 2) Band alpha channel Image of; 3) gray Image; 4) 16-bit Image
✅ The training code has been released. For details, see: Training.md 。

Portable version (green version) executable

You can download Supports Intel/AMD/Nvidia graphics cards Green version exe file of: Windows Edition / Linux / MacOS version 。

The green version means that you can run these execs directly (it's OK to copy them from the USB flash drive), because they already have the necessary files and models. It does not require CUDA or PyTorch running environment.

You can run it through the following command (Windows version example, please see the corresponding version README. md for more information):

 ./realesrgan-ncnn-vulkan.exe - i input image.jpg - o output image.png - n model name

We offer five models:

Realergan-x4plus (default)
reaesrnet-x4plus
Realergan-x4plus-anime (optimized for animation illustration images, with smaller volume)
RealESRGANv2-animevideo-xsx2 (for animation video, X2)
RealESRGANv2-animevideo-xsx4 (for animation video, X4)

You can -n Parameters to use other models, such as ./realesrgan-ncnn-vulkan.exe - i Quadratic picture. jpg - o Picture of Erythrocyter. png - n realesrgan-x4plus-anime

Usage of executable file

For more details, please refer to Real-ESRGAN-ncnn-vulkan .
Note: executable does not support python script inference_realesrgan.py All functions in, such as outscale Option)

 Usage: realesrgan-ncnn-vulkan.exe -i infile -o outfile [options]...

 -h                   show this help
 -v                   verbose output
 -i input-path        input image path (jpg/png/webp) or directory
 -o output-path       output image path (jpg/png/webp) or directory
 -s scale             upscale ratio (4, default=4)
 -t tile-size         tile size (>=32/0=auto,  default=0) can be 0,0,0 for multi-gpu
 -m model-path        folder path to pre-trained models(default=models)
 -n model-name        model name (default=realesrgan-x4plus,  can be realesrgan-x4plus | realesrgan-x4plus-anime | realesrnet-x4plus)
 -g gpu-id            gpu device to use (default=0) can be 0,1,2 for multi-gpu
 -j load:proc:save    thread count for load/proc/save (default=1:2:2) can be 1:2,2,2:2 for multi-gpu
 -x                   enable tta mode
 -f format            output image format (jpg/png/webp, default=ext/png)

Because these exe files will divide the image into several parts, and then process them separately, and then synthesize and export them, the output image may have a sense of fragmentation (and may not be the same as PyTorch's output)

These exe files are based on Tencent/ncnn as well as nihui Of realsr-ncnn-vulkan , thanks!

Github project address: https://github.com/xinntao/Real-ESRGAN