mirror of
https://github.com/zhw2590582/ArtPlayer.git
synced 2026-10-12 23:36:15 -08:00
Harvey Zhao
7dffd99cfd
Remove debugging border from the websr plugin's styling Enhance canvas sizing logic in artplayer-plugin-websr - Updated canvas positioning to center it within the player. - Introduced a new function to calculate canvas size based on the video aspect ratio. - Adjusted canvas dimensions during initialization and on resize events to maintain aspect ratio. - Ensured canvas styles are updated to reflect the calculated dimensions. feat: add comparison mode to artplayer-plugin-websr - Updated README.md to reflect new comparison feature and simplified content. - Modified package.json to remove unnecessary keywords. - Enhanced index.js to implement comparison functionality, allowing users to compare original and upscaled video. - Added new methods for enabling, disabling, and toggling comparison mode, as well as setting and getting the comparison position. - Updated TypeScript definitions to include new comparison options and methods.
14 lines
883 B
Markdown
14 lines
883 B
Markdown
# Network Weights
|
|
|
|
There are 3 Anime4k networks implemented:
|
|
|
|
* [cnn-2x-s](https://github.com/bloc97/Anime4K/blob/master/glsl/Upscale/Anime4K_Upscale_CNN_x2_S.glsl)
|
|
* [cnn-2x-m](https://github.com/bloc97/Anime4K/blob/master/glsl/Upscale/Anime4K_Upscale_CNN_x2_M.glsl)
|
|
* [cnn-2x-l](https://github.com/bloc97/Anime4K/blob/master/glsl/Upscale/Anime4K_Upscale_CNN_x2_VL.glsl)
|
|
|
|
The original Anime4K network weights are stored in:`cnn-2x-s-an.json`, `cnn-2x-m-an.json` and `cnn-2x-l-an.json`.
|
|
|
|
I have also retrained the same networks for Real Life and 3D content (e.g. `cnn-2x-m-rl.json` for Real Life content, or `cnn-2x-m-3d.json`). These were trained on data from [Xiph](https://media.xiph.org/video/derf/), however the upscaling results qualitatively don't seem as good as Anime4K weights on animated content.
|
|
|
|
I will continue working on improving the networks and training strategy.
|