Phase-2 follow-up deferred from the sound–motion extension (PRs #351–353): fuse MediaPipe world landmarks from ≥2 uncalibrated camera views via per-frame Umeyama similarity transform on rigid torso landmarks + 2D-visibility-weighted averaging, reporting cross-view residual (mm) as a quality metric. Working reference implementations exist in the Westney study (concert_fuse3d.py / reh_fuse3d.py — byte-identical near-duplicates that should become one library function).
🤖 Generated with Claude Code
Phase-2 follow-up deferred from the sound–motion extension (PRs #351–353): fuse MediaPipe world landmarks from ≥2 uncalibrated camera views via per-frame Umeyama similarity transform on rigid torso landmarks + 2D-visibility-weighted averaging, reporting cross-view residual (mm) as a quality metric. Working reference implementations exist in the Westney study (concert_fuse3d.py / reh_fuse3d.py — byte-identical near-duplicates that should become one library function).
🤖 Generated with Claude Code