Getting started

Try it with the simulation (no GPU)

You need Docker with Compose.

git clone https://github.com/eyesonplay/eyesonplay.git
cd eyesonplay
docker compose up --build

Open http://localhost:3000 and sign in with the local development account admin@example.com / eyesonplay-admin.

1. Click New match.
2. Paste any HLS URL, for example https://test-streams.mux.dev/x36xhzz/x36xhzz.m3u8.
3. Press Create & start processing.

The worker runs in INFERENCE_MODE=mock: a simulated football match (or tennis rally, if you choose tennis) is projected through a synthetic broadcast camera and fed through the real tracking and event pipeline. Every part of the dashboard works, but the detections are not related to the video’s content.

URL
http://localhost:3000	Dashboard
http://localhost:8000/docs	API docs (OpenAPI)
ws://localhost:8000/ws/matches/{id}	Live feed (signed in)

Analyse real video

Real analysis needs model weights and, for live speed, a GPU:

# NVIDIA GPU (needs the NVIDIA Container Toolkit)
docker compose -f docker-compose.yml -f docker-compose.gpu.yml up --build
# CPU (slow, any machine)
docker compose -f docker-compose.yml -f docker-compose.real.yml up --build

Then set the match’s sport and detection model:

• Football: Ball + Players + Pitch with the pitch keypoint model for the mini pitch.
• Tennis: Ball + Players + Court, 25 fps.

See Football, Tennis and Third-party software and models for the weights.

On a Mac, Docker cannot use the Apple GPU. Run the worker natively against the Docker stack instead; the README has the commands.

Next

• How it works
• Deploy to production
• Integration feed