catnip
A work in progress
Catnip is a small Jupyter notebook for those obsessive lovers of a specific manga character with spare time and computing power.
Input some manga pages, add some seeds, and rejoice at the sight of 40 previews of matches. I haven’t implemented an actual method to export the files. Yet.
Experimental Branches
While the main branch acts as a baseline, there is active experimentation in several alternative branches exploring different bounding box (BBO) approaches, models, and environments:
legacy-keras: Older experimental Keras implementations.yolo-bbo: Utilizing YOLO-based models specifically tuned for Anime/Manga bounding boxes.yolo-windows: Windows-specific testing and pathing for YOLO pipelines.
Usage
This is how I’m testing this. So far.
- I suggest using Pixi for package management
- Set up a directory with sample manga pages where the desired character appears
- Run panel and facial cropping cells
- Collect samples of the desired character in a separate directory
- Run panel/facial cropping cells for the rest of the manga and wait
- Point to seeds (samples) directory, run the nearest neighbor algorithm and wait (again)
Comments
Adenzu’s Panel Extractor algorithm is excellent but takes a long time to run, even without split-cell. panelExtraction.py includes a faster albeit more basic and somewhat less accurate approach.