ACE-Ego-Hand

Checkpoints for ACE-Ego-Hand: Repurposing Video Diffusion Models for Occlusion-Robust Egocentric 3D Hand Motion Recovery (arXiv:2608.20308).

ACE-Ego-Hand recovers bimanual 3D hand motion (MANO pose, shape and camera-space translation, per frame) from egocentric RGB video by repurposing a video diffusion model (Wan2.2-Fun-5B-Control) as a deterministic geometry encoder. Inference code: https://github.com/ggxxii/ACE-Ego-Hand · project page: https://ggxxii.github.io/ace-ego-hand/

Checkpoints

model	checkpoint	camera intrinsics
K-given	ace_ego_hand_k.pt	required
K-free	ace_ego_hand_kfree.pt	not needed

Both are the paper checkpoints. Inference code and usage: https://github.com/ggxxii/ACE-Ego-Hand

⚠️ Pinhole only. These checkpoints do not support native fisheye input. To run on fisheye footage, undistort it to a pinhole view first and pass the pinhole intrinsics.

License

The weights are released for research purposes under CC BY-NC 4.0: they were trained on research-licensed datasets and derive from the MANO body model, so commercial use is not permitted. The code in the GitHub repository is MIT-licensed.