Skip to content

Latest commit

 

History

History
50 lines (41 loc) · 2.5 KB

File metadata and controls

50 lines (41 loc) · 2.5 KB

AutoPTZ Notices

AutoPTZ is licensed under the GNU Affero General Public License version 3.0. See LICENSE.md.

Third-party code and model assets can have their own licenses. AutoPTZ tries to make those boundaries visible in the app under Services -> Optional Setup.

Models and optional features:

  • YOLO detector and pose models are used for person boxes, automatic PTZ following, pose overlay, and torso-stable framing. Ultralytics states that its YOLO trained models are AGPL-3.0 by default. AutoPTZ can download/export these into its own model cache, and can remove the AutoPTZ-managed cache files from the Services panel.
  • InsightFace is used only for optional face recognition. The InsightFace code is MIT-licensed, but the upstream model packs are labeled for non-commercial research unless a commercial license is obtained from InsightFace. AutoPTZ does not delete the upstream InsightFace cache because it may be shared by other tools.
  • BoxMOT/ReID is used only for optional stable re-acquire after occlusion. Its package/model dependencies are managed by the upstream package ecosystem, not by AutoPTZ's model cache.
  • NDI support is optional and depends on cyndilib.

Missing models or optional dependencies should never prevent AutoPTZ from launching. The app disables unavailable feature controls and keeps live preview available where possible.

AutoPTZ Mark benchmark video clips:

The bundled demo clips under autoptz/assets/ are used only by the "AutoPTZ Mark" benchmark to drive synthetic cameras. Each was trimmed/transcoded (re-scaled, re-timed, audio removed) from a permissively-licensed source:

  • mark_people_1080p.mp4 — derived from the OpenCV sample vtest.avi (pedestrians). OpenCV samples are BSD-3-Clause.
  • mark_crowd_30.mp4 — derived from "Shibuya Scramble Crossing" by Gst, via Wikimedia Commons, licensed CC BY-SA 3.0. The AutoPTZ-derived clip is likewise made available under CC BY-SA 3.0 (ShareAlike).
  • mark_people_24.mp4 and mark_people_60.mp4 — derived from "Tears of Steel" (CC) Blender Foundation | mango.blender.org, licensed CC BY 3.0. The mark_people_60.mp4 variant was frame-interpolated to 60 fps.
  • mark_faces_30.mp4 — derived from a U.S. White House Daily Press Briefing (April 11, 2016). A work of the U.S. federal government, in the Public Domain (17 U.S.C. § 101). Used as the face-recognition benchmark scene.

These clips are demo assets only and are independent of the AGPL-3.0 license that covers AutoPTZ's own source code.