ForeHOI
Tao-11-chen
Feed-forward object reconstruction and pose estimation from partially observed hand–object interaction videos.
What you can explore
Explore object geometry recovery, interactive demos and the released HO-3D and HOT3D evaluation paths.
Before you use it
Prebuilt FoundationPose components target Python 3.11; other environments need rebuilding. External pretrained models and datasets have separate access requirements.
Original contributors
Yuantao Chen and ForeHOI collaborators
License / access: Code: MIT · check external data, models and hardware terms separately
Keep exploring
HOLD
Joint reconstruction of a hand and a held object from video.
HO-3D
Images and three-dimensional annotations of a human hand manipulating objects under occlusion.
HOT3D
Multi-view egocentric recordings of human hands interacting with objects, with tracking annotations.
Editorial reference · checked 2026-09-22