Skip to content

THIRD_PARTY_LICENSES.md

Latest commit

 

History

History
40 lines (28 loc) · 3.19 KB

THIRD_PARTY_LICENSES.md

File metadata and controls

40 lines (28 loc) · 3.19 KB
Raw

Third-Party Licenses

This project bundles and depends on the following third-party software. Each is the property of its respective owners and licensed under the terms below.

Direct Python dependencies

Package Version License Notes
ultralytics ≥ 8.0.0 AGPL-3.0 Viral copyleft — forces this project to AGPL-3.0. Commercial use requires an Ultralytics Enterprise Licence.
opencv-python ≥ 4.8.0 Apache 2.0 Permissive.
Pillow ≥ 10.0.0 MIT-CMU (HPND) Permissive.
numpy ≥ 1.24.0 BSD-3-Clause Permissive.
yt-dlp ≥ 2024.1.0 Unlicense (public domain) No constraints.
requests ≥ 2.28.0 Apache 2.0 Permissive.
imageio-ffmpeg ≥ 0.4.9 BSD-2-Clause Wrapper only. Bundles an FFmpeg binary — see below.

Bundled binaries

Software Source License Notes
FFmpeg shipped inside imageio-ffmpeg LGPL-2.1+ (some builds GPL-2.1+) The build that imageio-ffmpeg ships is LGPL-compatible. We use it as a separate binary invoked by yt-dlp — no static linking — so LGPL "dynamic use" applies.
Pre-trained YOLO weights (yolo*.pt) downloaded from Ultralytics hub at runtime AGPL-3.0 (weights inherit the framework licence) If you redistribute the app with bundled weights, those weights are also covered by AGPL.

Model accuracy / dataset attribution

The pre-trained YOLO weights are trained on COCO (cocodataset.org) — Creative Commons Attribution 4.0. Models inherit dataset attribution requirements when used to publish dataset-derived predictions.

User-supplied content

  • YouTube downloads — yt-dlp downloads videos but does not grant you rights to them. YouTube's Terms of Service prohibit redistributing copyrighted content. Use only with videos you own or that are licensed (CC, public-domain, your own uploads). The maintainers of this project accept no responsibility for ToS violations by end-users.
  • Annotated images — the user retains all rights to images and labels they create with this tool. Exported datasets are the user's property.

Why AGPL-3.0 for this project

Ultralytics YOLO is AGPL-3.0. Section 13 of the AGPL ("Remote Network Interaction") means that any project linking against AGPL code must itself be AGPL-licensed when distributed, including over a network. Since this app links directly against ultralytics, AGPL-3.0 is the only legally compatible licence for redistribution unless you purchase an Ultralytics Enterprise Licence.

If you want a permissive licence for commercial closed-source use, you would need to:

  1. Buy the Ultralytics Enterprise Licence, or
  2. Replace ultralytics with a permissively-licensed detector (e.g. plain PyTorch with custom weights, MMDetection under Apache 2.0, etc.).