Jump to a moment
Jump to a moment
Jump to a moment
Jump to a moment
Jump to a moment
Jump to a moment
Jump to a moment
Jump to a moment
Measured performance
More hands captured.
Fewer identity mistakes.
In DreamVu's Ego-Exo4D evaluation, HandTrack led HaMeR and WiLoR on all six measures below. It detected 98.02% of reference hands, reducing missed hands by approximately 60% relative to either baseline.
98.02%Hand coverage
≈60%Fewer missed hands
99.80%Left/right accuracy
65 clips processed · 4,144 scored frames · 6,477 reference hand instances · Ego-Exo4D validation data
| Metric | HaMeR | WiLoR | DreamVu HandTrack V1.1 |
|---|---|---|---|
| Joint accuracy (PCK@0.10, manual joints) | 82.93% | 74.33% | 84.32% |
| Hand coverage / recall | 95.12% | 94.94% | 98.02% |
| Left/right accuracy | 99.16% | 99.22% | 99.80% |
| Reference-joint coverage | 95.63% | 95.47% | 98.20% |
| Detection precision (both hands labelled) | 99.47% | 99.36% | 100.00% |
| Detection F1 (both hands labelled) | 98.13% | 98.02% | 99.49% |
Built for real-world demonstrations
Turn human activity into more useful robot-learning data.
Keep more of the action
More complete hand coverage preserves the movements that matter during grasping, tool use and object manipulation.
Keep left and right consistent
HandTrack maintains hand identity across a recording. In this evaluation, its left/right error rate was approximately four times lower than the baselines.
Process faster than playback
In DreamVu's 1080p processing test, HandTrack achieved 36.4 frames per second—faster than 30 fps video playback.
Bring HandTrack to your robotics workflow.
Talk to DreamVu about hand tracking for first-person demonstration video.