Research map / 3D vision / 3D generation
Physically Grounded 3D Generative Reconstruction under Hand Occlusion using Proprioception and Multi-Contact Touch
Gabriele Mario Caddeo, Pasquale Marra, Lorenzo Natale
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 proprioception and touch (vs vision-only) (ablation) | Stage A flow model · YCB sim | CD: 0.142 -> 0.033 | Improved |
| add proprioception (hand geometry) to vision-only (ablation) | Stage A flow model · YCB sim | Voxel IoU: 0.357 -> 0.430 | Improved |
| add touch on top of proprioception (ablation) | Stage A flow model · YCB sim | Voxel IoU: 0.430 -> 0.586 | Improved |
| add physics losses during training (ablation) | Stage A finetuning · YCB sim | CD: 0.052 -> 0.033 | Improved |
| add decoder-based physics guidance at inference (ablation) | Stage A sampling · YCB sim | Voxel IoU: 0.530 -> 0.586 | Improved |
| add distance-based contact channel to binary contact (ablation) | touch representation · YCB sim | Voxel IoU: 0.505 -> 0.586 | Improved |
| add binary contact to distance contact (ablation) | touch representation · YCB sim | Voxel IoU: 0.532 -> 0.586 | Improved |
| tactile noise 3 mm at test time (ablation) | Ours · YCB sim | CD: 0.033 -> 0.033 | No clear effect |
| tactile noise 5 mm at test time (ablation) | Ours · YCB sim | Voxel IoU: 0.586 -> 0.563 | Hurt |
| kinematic noise level 2 (1 cm) at test time (ablation) | Ours · YCB sim | Voxel IoU: 0.586 -> 0.501 | Hurt |
| mask noise level 2 (5 px) at test time (ablation) | Ours · YCB sim | Voxel IoU: 0.586 -> 0.537 | Hurt |
| replace Sam3D vision-only with physically grounded Ours | pose estimation · YCB sim | ADD-S@0.1: 0.658 -> 0.73 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Sam3D — SAM 3D: 3Dfy Anything in Images (CVPR 2026) · better on 10 of 11 shared results
- Amodal3R — Amodal3R: Amodal 3D Reconstruction from Occluded 2D Images (ICCV 2025) · better on 6 of 6 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- MEgoHand: Multimodal Egocentric Hand-Object Interaction Motion Generation (NeurIPS 2025 · similarity 0.89)
- Hand-held Object Reconstruction from RGB Video with Dynamic Interaction (CVPR 2025 · similarity 0.88)
- ComPose: When to Trust Hands for Object Pose Tracking (NeurIPS 2026 · similarity 0.88)
- DiffGrasp: Whole-Body Grasping Synthesis Guided by Object Motion Using a Diffusion Model (AAAI 2025 · similarity 0.87)
- HandBooster: Boosting 3D Hand-Mesh Reconstruction by Conditional Synthesis and Sampling of Hand-Object Interactions (CVPR 2024 · similarity 0.87)
- In-Hand 3D Object Reconstruction from a Monocular RGB Video (AAAI 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.
