2024-10-13 20:56:50 +08:00
2024-04-13 22:56:17 +08:00
2024-10-13 20:56:50 +08:00
2024-10-13 20:56:50 +08:00
2024-04-13 22:56:17 +08:00
2024-04-12 04:02:42 -04:00
2024-04-13 22:56:17 +08:00
2024-04-12 04:09:54 -04:00
2024-04-13 13:14:41 -04:00
2024-04-12 04:02:42 -04:00
2024-04-14 12:54:14 -04:00
2024-04-12 04:02:42 -04:00
2024-04-12 04:02:42 -04:00
2024-04-12 04:02:42 -04:00
2024-04-12 04:02:42 -04:00

web-realesrgan

Run Real-ESRGAN in the browser with tensorflow.js

Usage

Drag in or select an image, then choose a model and WebGPU/WebGL to upscale the image.

Sure, PNG with alpha channel is supported.

Online Demo

https://cappuccino.moe

demo

Accelaration

  • WebGL: Enabled on most devices by default.
  • WebGPU: Enabled on Chrome with chrome://flags/#enable-unsafe-webgpu flag or other browsers with WebGPU support. Much faster than WebGL.

Development

npm install
npm run dev

Models

I've converted 4 models to tensorflow.js format, you can find the original pytorch models in xinntao/Real-ESRGAN

  • anime (fast) RealESRGAN-animevideov3
  • anime (plus) RealESRGAN_x4plus_anime_6B
  • general (fast) RealESRGAN-general-x4v3
  • general (plus) RealESRGAN_x4plus

The first run will download the models and cache them in the browser indexedDB. No need to download them again.

Credits

Pytorch model -> ONNX -> Tensorflow saved model -> Tensorflow.js

Description
No description provided
Readme 1.9 MiB
Languages
Vue 62.6%
JavaScript 15.6%
CSS 10.4%
TypeScript 8.2%
C 2%
Other 1.2%