https://github.com/funnyplanter/CuNNy

Convolutional upscaling Neural Network, yeah!
https://github.com/funnyplanter/CuNNy

anime magpie mpv upscaling

Last synced: 3 months ago
JSON representation

Convolutional upscaling Neural Network, yeah!

• Host: GitHub
• URL: https://github.com/funnyplanter/CuNNy
• Owner: funnyplanter
• License: lgpl-3.0
• Created: 2024-06-17T15:00:18.000Z (5 months ago)
• Default Branch: master
• Last Pushed: 2024-08-01T22:27:31.000Z (4 months ago)
• Last Synced: 2024-08-02T06:11:41.201Z (4 months ago)
• Topics: anime, magpie, mpv, upscaling
• Language: GLSL
• Homepage:
• Size: 4.33 MB
• Stars: 13
• Watchers: 1
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-mpv - CuNNy - Cute and funny CNN-based upscaler optimized for anime. (Other)

README

        
# CuNNy - Convolutional upscaling Neural Network, yeah!

Nice, small, and fast realtime CNN-based upscaler. Trained on visual novel

screenshots/CG.

Currently very new and immature 😭.

Supports exporting to an mpv ~~meme~~shader!

And now a Magpie effect!

# Usage

mpv shaders are found inside the `mpv/` directory.

Metric-focused variants are found inside the `results/` directory.

Magpie effects are found inside the `magpie/` directory.

The order of quality is 8x32 > 4x32 > 4x24 > 4x16 > 4x12 > 3x12 > 2x12 > fast >

faster > veryfast. Conversely the order of speed would be the reverse.

The `CuNNy-fast` shader is the recommended shader if you're on a slow machine.

Variants:

- (No-suffix): Trained to be neutral and do no denoising or sharpening.

- `SOFT`: Trained to anti-alias & produce a soft output. Is probably the most

  'artifact-free' variant if you don't count blur.

- `DS`: Trained to denoise & sharpen.

- `NVL`: Trained to upscale visual novel games & high-quality illustrations.

There are versions of the mpv shaders use 8-bit `dp4a` instructions. They can be

many times faster than the standard upscaling shader depending on if your

hardware supports accelerated `dp4a` instructions. Requires `vo=gpu-next` with

`gpu-api=vulkan`. They can be found inside the `dp4a/` subdirectories.

# Training

Tested training with PyTorch nightly. If any errors arise try using nightly.

Prepare data by running `sh scripts/build.sh`, then `sh scripts/split.sh

 `, then `py scripts/proc.py <128-grids> `.

To train `py train.py   ` where `N` is the number of internal

convolutions and `D` is the number of feature layers.

Convert the resulting model to an mpv shader by running

`py mpv.py `.

Convert the resulting model to a Magpie effect by running

`py magpie.py `.

Trains very fast on my machine.

# Quality

See `results/`.

# License

LGPL v3