Research map / 3D vision / Human pose estimation
Towards in-the-wild Egocentric 3D Hand-Object Pose Estimation
Siddhant Bansal, Zhifan Zhu, Shashank Tripathi, Jiahe Zhao, Michael Black, Dima Damen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add CDev-based interaction loss (ablation) | HOPformer · ARCTIC (egocentric, in-lab) | CDev (mm): 40.7 -> 31.9 | Improved |
| add object losses (ablation) | HOPformer · ARCTIC (egocentric, in-lab) | SR@0.05 (%): 66.8 -> 82.4 | Improved |
| add 3D keypoint loss (ablation) | HOPformer · ARCTIC (egocentric, in-lab) | MPJPE (mm): 18.7 -> 16.1 | Improved |
| add object pose loss (ablation) | HOPformer · ARCTIC (egocentric, in-lab) | SR@0.05 (%): 72.5 -> 82.4 | Improved |
| use hand-conditioned cross-attention decoder instead of concat + MLP (ablation) | HOPformer · ARCTIC (egocentric, in-lab) | SR@0.05 (%): 29.4 -> 82.4 | Improved |
| use WiLoR hand features instead of generic DINOv2 tokens (ablation) | HOPformer · ARCTIC (egocentric, in-lab) | SR@0.05 (%): 70.0 -> 82.4 | Improved |
| add self-attention in the decoder (ablation) | HOPformer · ARCTIC (egocentric, in-lab) | SR@0.05 (%): 55.8 -> 82.4 | Improved |
| add learnable aggregation module (ablation) | HOPformer · ARCTIC (egocentric, in-lab) | SR@0.05 (%): 74.0 -> 82.4 | Improved |
| increase decoder depth from 1 to 12 layers (ablation) | HOPformer · ARCTIC (egocentric, in-lab) | SR@0.05 (%): 21.9 -> 82.4 | Improved |
| increase decoder depth from 6 to 12 layers (ablation) | HOPformer · ARCTIC (egocentric, in-lab) | SR@0.05 (%): 58.6 -> 82.4 | Improved |
| initialise with ARCTIC weights instead of training from scratch (ablation) | HOPformer · EPIC-Contact (egocentric, in-the-wild) | MRRPEro (mm): 99.5 -> 65.8 | Improved |
| initialise with ARCTIC weights instead of training from scratch (ablation) | HOPformer · EPIC-Contact (egocentric, in-the-wild) | MDev (mm): 19.5 -> 11.4 | Improved |
Closest papers by meaning
- Benchmarks and Challenges in Pose Estimation for Egocentric Hand Interactions with Objects (ECCV 2024 · similarity 0.89)
- WiLoR: End-to-end 3D Hand Localization and Reconstruction in-the-wild (CVPR 2025 · similarity 0.88)
- EasyHOI: Unleashing the Power of Large Models for Reconstructing Hand-Object Interactions in the Wild (CVPR 2025 · similarity 0.88)
- Reconstructing Hands in 3D with Transformers (CVPR 2024 · similarity 0.87)
- HOT3D: Hand and Object Tracking in 3D from Egocentric Multi-View Videos (CVPR 2025 · similarity 0.87)
- HandBooster: Boosting 3D Hand-Mesh Reconstruction by Conditional Synthesis and Sampling of Hand-Object Interactions (CVPR 2024 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
