Skip to content

Third-party software and models

EyesOnPlay is licensed under the GNU AGPL-3.0 (see LICENSE), with commercial licences available. This page lists what it builds on, under which licence, and which model weights are not distributed with the project.

Versions are those in use at release 0.1.0; licences are taken from each package’s own metadata. Check them again before shipping a commercial build.

Model weights (not included in this repository)

Section titled “Model weights (not included in this repository)”

No model weights are committed. The worker loads them from MODELS_DIR (default ~/.pitchside/models for the native worker, /data/models in Docker).

File Used for Source Licence Status
yolov8n.pt, yolov8s.pt Players and ball (COCO) Ultralytics, downloaded automatically AGPL-3.0 Usable under AGPL; a commercial closed-source product needs an Ultralytics Enterprise licence or another detector
football-player-detection.pt, football-ball-detection.pt, football-pitch-detection.pt Football players/ball, pitch keypoints roboflow/sports Code MIT; YOLO-based weights trained on Roboflow Universe datasets Check the training datasets’ licences and the Ultralytics licence before commercial use
tracknet_ball.pt Tennis ball (TrackNet) yastrebksv/TennisProject No licence published Personal evaluation only. Not redistributable; must be replaced with your own trained model before any release that ships weights or any commercial use
tennis_bounce.cbm Tennis bounce classifier (CatBoost) yastrebksv/TennisProject No licence published Same as above. Without it the engine falls back to its built-in bounce rule

The network definition in inference-worker/worker/detect/tracknet.py is our own implementation of the published TrackNet architecture (it only mirrors the layer names so compatible weights load).

Package Version Licence Used by
fastapi 0.142 MIT api
uvicorn 0.54 BSD-3-Clause api
sqlalchemy 2.1 MIT api
asyncpg 0.31 Apache-2.0 api
alembic 1.20 MIT api
python-multipart 0.0.32 Apache-2.0 api
pydantic, pydantic-settings 2.13, 2.15 MIT api, worker
redis (redis-py) 8.1 MIT api, worker
structlog 26.1 MIT OR Apache-2.0 api, worker
numpy 2.5 BSD-3-Clause (and bundled permissive licences) worker
opencv-python-headless 5.0 Apache-2.0 worker (real mode)
ultralytics 8.4 AGPL-3.0 worker (real mode)
torch 2.14 Apache-2.0 (per package metadata) worker (real mode, via ultralytics)
torchvision 0.29 BSD worker (real mode)
catboost 1.2 Apache-2.0 worker (optional, tennis bounces)
rapidocr 3.9 Apache-2.0 worker (real mode, scoreboard OCR for goals; bundles PaddleOCR ONNX models, Apache-2.0)
onnxruntime 1.2x MIT worker (real mode, runs the OCR models)
scipy 1.18 BSD worker (real mode, via ultralytics)
nvidia-ml-py 13.6 BSD worker (GPU metrics)

The event engine (event-engine/) has no runtime dependencies.

Package Version Licence
next 16.3 MIT
react, react-dom 19.2 MIT
@base-ui/react 1.8 MIT
shadcn 4.21 MIT
tailwindcss 4.3 MIT
tw-animate-css 1.4 MIT
class-variance-authority 0.7 Apache-2.0
@tanstack/react-query 5.104 MIT
@tanstack/react-virtual 3.14 MIT
react-hook-form, @hookform/resolvers 7.89, 5.9 MIT
zod 4.6 MIT
hls.js 1.7 Apache-2.0
lucide-react 1.52 ISC
next-themes 0.4 MIT
sonner 2.0 MIT

Services and tools (run alongside, not linked)

Section titled “Services and tools (run alongside, not linked)”
Component Licence Note
PostgreSQL (postgres:16-alpine) PostgreSQL License
Redis (redis:7-alpine) Redis 7.4+: RSALv2 / SSPLv1 (earlier 7.x: BSD-3-Clause) Fine to run unmodified as your own backing service; it may not be offered to others as a managed Redis service. Valkey (BSD-3-Clause) is a drop-in alternative
FFmpeg LGPL-2.1+ / GPL-2.0+ depending on the build Called as a separate program; the Debian packages in the images are GPL builds
NVIDIA CUDA images NVIDIA Deep Learning Container licence GPU image only