Macrodata's MacroHands hits 13.1mm stereo error and 80% detection under occlusion, beating all open baselines
gui_penedo · x · 2026-11-11
Macrodata Labs released MacroHands, a family of 3D hand tracking and camera pose models for egocentric video, in mono and stereo variants.
Key results:
- MacroHands-Mono beats every open-source baseline tested, including stereo ones: 32.2mm error on SHOW3D vs 55.9 for HaWoR and 46.4 for the best stereo baseline
- Stereo cuts error 2.5× to 13.1mm, mostly via better depth estimates
- Strong occlusion robustness: detects 80% of hands on heavily occluded clips vs at most 59% for baselines
- Distinguishes wearer from others: only 0.2% of detections land on someone else's hands (vs 24% for HaWoR); works with gloves and real-world scenes
- Camera pose is also SOTA — 4× less drift than DROID-SLAM when IMU is available
Reliable egocentric hand perception like this is key infrastructure for robots learning manipulation from human video.
Related event: MacroData Releases MacroHands for Egocentric 3D Hand Tracking(2 posts)→
More from Embodied
- SpaceX moves Super Heavy booster to the launch pad at Starbase — DimaZeniuk · 2026-11-11
- SkeLAM Uses Skeleton-Grounded Latent Actions to Decouple Motion from Appearance in Robot World Models — jiqizhixin · 2026-11-11
- Hub to open-source most diverse egocentric robot dataset on Hugging Face Nov 12 — alex_verem · 2026-11-11
- GTA V on Specs glasses: light trails and rain leap off the big screen into your room — stspanho · 2026-11-11
- MBZUAI releases open world model PAN 2: LLM planning plus 29B diffusion hits 75.1% step accuracy — rohanpaul_ai · 2026-11-11
- ProtoMotions Deployment Stack: One Exported Graph, 1.3 ms C++ Loop for Real Robots — carlosdponx · 2026-11-11