01 / THE REFERENCEREAL FOX · 25 FPS
Play or scrub this clip. Every view below follows the same source time.
REAL FOOTAGE. VISIBLE EVIDENCE.
Watch real movement beside its estimated skeleton. Then see what a timing model can, and cannot, reliably learn from nine seconds in the wild.
Play or scrub this clip. Every view below follows the same source time.
Confident 2D observations. Low-confidence points are omitted, not invented. Watch the diagnostic video ↗
Horizontal paw position relative to the torso center, scaled by torso length. Raw masked observations are shown. The shaded final 30% was held out when fitting each timing model.
The pose network was pretrained by its authors. This project fits a small timing model, not a new animal foundation model. These are single-camera estimates, not 3D motion capture or learned game physics.
Inspect the full measurement report ↗