Volleyball analysis

Computer vision project with ML models trained on custom datasets in order to evaluate real-world volleyball matches

Setup

If using a venv: Add [any-name].pth in venv/Lib with content ../../..

Dataset

Download unstructured video from youtube

Following command downloads the highest available video quality. Requires yt-dlp and ffmpeg

yt-dlp -f "bestvideo+bestaudio/best" --merge-output-format mp4 https://www.youtube.com/URL-TO-VIDEO

Split unstructured video into separate clips

Currently use Davinci Resolve. This is not the most efficient because all clips have to be re-coded (re-rendered) because no project currently exists that easily lets us split a video and output into separate clips without having to re-code

1. Split full video into separate clips for label serve, rally, idle
2. Export as separate clips
3. Place clips in dataset directory into their respective label folders train, val, test

gamestate-dataset/
├── train/
│   ├── rally
│   ├── serve
│   └── idle
├── val/
│   ├── rally
│   ├── serve
│   └── idle
└── test/
    ├── rally
    ├── serve
    └── idle