*NEW* Our #ECCV2026 @eccv.bsky.social paper Towards in-the-wild Egocentric 3D Hand-Object Pose Estimation Now on ArXiv w Dataset, Code&model sid2697.github.io/epic-contact/ arxiv.org/abs/2606.30598 Two contributions: 1. EPIC-Contact Dataset 2. HOPformer Method &Checkpoint 🧵 1/6
- EPIC-Contact provides 2.3K clips and 62.3K frames of in-the-wild egocentric footage with dense, bijective 3D hand-object contact correspondences and posed meshes.
- HOPformer is an end-to-end transformer that jointly predicts bi-manual hand and object pose in a single forward pass using a cross-attention decoder.
- The model reaches 82.4% success rate on ARCTIC, 6.2 points above prior state of the art, and nearly doubles success rate on EPIC-Contact while cutting contact deviation by 75%.