OpenPTV2 v0.2.1 released — faster, headless-friendly, cloud-ready

[Alex Liberzon]

Hi all,

We're excited to announce OpenPTV2 v0.2.1, now on PyPI and GitHub.

This is the first 0.2.x release and brings substantial performance, packaging, and tracking improvements since v0.1.7.

Performance

• Correspondences ~157× faster (flat NumPy arrays)
• Target recognition ~63× faster (targ_rec_fast)
• GIL-free BFS + parallel 4-camera detection (~2.3×)

Packaging & deployment

• Default pip install openptv2 is now the slim, headless batch runtime (no Qt/GUI dependencies)
• Desktop GUI: pip install "openptv2[gui]" → launch with openptv2-gui
• New Dockerfile.cloud (free-threaded CPython 3.14t) + a batch deployment guide (docs/cloud-batch.md)
• Python 3.11–3.14 supported; prebuilt wheels for Linux, Windows, and macOS (cp311–cp313)

Tracking

• Cold-start recovery + forward/backward reciprocity post-passes
• 3D segment tracking fixes (full_forward_3d)
• Fixed a forward-tracking off-by-one at the sequence end

Parameters

• YAML-first experiment discovery with explicit active-run selection
• Pydantic v2 parameter validation

Install

 

pip install openptv2==0.2.1 # headless / batch

 

pip install "openptv2[gui]==0.2.1" # + desktop GUI

 

(Note: please pin ==0.2.1 for now — an older stray 1.0.0 upload still sits on PyPI and would otherwise be picked first. We'll clean that up.)

Release notes: https://github.com/alexlib/openptv2/releases/tag/v0.2.1 PyPI: https://pypi.org/project/openptv2/

As always, bug reports, questions, and contributions are very welcome.

Happy tracking, The OpenPTV team