Skip to content
BetaThis project is currently in its beta version.Updates might take a while to roll out.There's still plenty of work left to do.Thanks for your patience while we keep building.
BetaThis project is currently in its beta version.Updates might take a while to roll out.There's still plenty of work left to do.Thanks for your patience while we keep building.
mHumanTrackAImHumanTrackAI

Changelog

Version history for mHumanTrackAI. To add a release, copy a block below and update the version, date and notes.

v1.3.4 — current

The current release. Highlights of where the project stands:

  • Whole-body tracking — RTMW 133-keypoint model (33 body + feet + face + 21 landmarks per hand) with balanced and performance tiers.
  • Two modes — Accessibility (cursor / clicks / scroll / keyboard) and Gaming (virtual Xbox 360 controller via ViGEm).
  • Multi-camera fusion — per-joint best-view selection for front + back occlusion coverage.
  • Occlusion prediction — One-Euro + Kalman + IK extrapolation, rendered ghosted/dashed.
  • Recording & playback — record landmark streams; export to CSV / BVH.
  • Automatic GPU/CPU detection — a single installer with a CUDA 12 stack and optional TensorRT FP16 fast path (~90 fps on an RTX 2060) on NVIDIA GPUs, falling back to RTMW on CPU or MediaPipe everywhere else.
  • Local-only by design — two processes, loopback (127.0.0.1) only, no cloud and no telemetry.

Maintainers

This page is plain MDX. Add new entries at src/app/docs/changelog/page.mdx — each release is just an ## vX.Y.Z — date heading followed by a bullet list.