Research map / 3D vision / Face and shape
gSDF: Geometry-Driven Signed Distance Functions for 3D Hand-Object Reconstruction
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| use all hand joint transformations (K^h_3) instead of wrist only (ablation) | hand kinematic feature · DexYCB | CD_h: 0.344 -> 0.317 | Helped |
| add hand pose priors (wrist or all joints) to hand kinematic feature (ablation) | hand kinematic feature · DexYCB | CD_h: 0.364 -> 0.317 (no pose vs all joints) | Helped |
| add hand pose to object kinematic feature (on top of object center) (ablation) | object kinematic feature · DexYCB | CD_o: 1.93 -> 1.71 | Helped |
| object center alignment (K^o_2) instead of no pose priors (ablation) | object kinematic feature · DexYCB | CD_o: 2.06 -> 1.93 | Helped |
| replace global visual feature with geometry-aligned local visual feature (ablation) | visual feature · DexYCB | CD_h: 0.317 -> 0.310 | Helped |
| add spatio-temporal transformer over neighboring frames to local visual features (ablation) | visual feature · DexYCB | CD_o: 1.71 -> 1.55 (V1 -> V4) | Helped |
| use single shared image backbone for pose and SDF prediction (ablation) | image backbone sharing · DexYCB | CD_h: 0.317 -> 0.411 | Hurt |
| asymmetric backbone (separate hand-pose backbone) vs symmetric backbone (ablation) | image backbone sharing · DexYCB | CD_h: 0.324 -> 0.317 | Helped |
| gSDF vs Karunratanakul et al. on hand Chamfer distance (ObMan) | gSDF full model · ObMan | CD_h: 0.261 -> 0.112 | Helped |
| gSDF full model vs prior SOTA on DexYCB video | gSDF full model · DexYCB | CD_h: 0.344 (AlignSDF-2BB adapted Chen) -> 0.302 | Helped |
| gSDF vs Chen et al. on ObMan object metrics | gSDF full model · ObMan | E_o: 3.29 (Chen) vs 3.43 (gSDF) | Hurt |
| gSDF vs Chen et al. on DexYCB object pose error | gSDF full model · DexYCB | E_o: 1.78 (Chen) vs 1.96 (gSDF) | Hurt |
Closest papers by meaning
- HOISDF: Constraining 3D Hand-Object Pose Estimation with Global Signed Distance Fields (CVPR 2024 · similarity 0.91)
- HOLD: Category-agnostic 3D Reconstruction of Interacting Hands and Objects from Video (CVPR 2024 · similarity 0.87)
- Hand-held Object Reconstruction from RGB Video with Dynamic Interaction (CVPR 2025 · similarity 0.87)
- SRHand: Super-Resolving Hand Images and 3D Shapes via View/Pose-aware Neural Image Representations and Explicit Meshes (NeurIPS 2025 · similarity 0.87)
- HandOS: 3D Hand Reconstruction in One Stage (CVPR 2025 · similarity 0.86)
- Sensing Surface Patches in Volume Rendering for Inferring Signed Distance Functions (AAAI 2025 · similarity 0.86)
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.