Research map / 3D vision / Human pose estimation
Reconstructing 3D Human-Object Interaction via a Unified Triplane Space
Yuhang Chen, Chenxing Wang
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 semi-continuous (SDF-based) occupancy instead of binary (ablation) | HOIT-VAE · BEHAVE | F-score combined: 0.9804 -> 0.9913 at 512 res | Improved |
| increase voxel resolution from 128 to 512 (ablation) | HOIT-VAE · BEHAVE | F-score combined: 0.9740 -> 0.9913 | Improved |
| add cosine similarity loss to MSE loss (ablation) | HOIT-T · BEHAVE | F-score human/object: human 0.5922 -> 0.4838; object 0.3885 -> 0.4434 | Mixed |
| add KL loss on top of MSE + SIM (ablation) | HOIT-T · BEHAVE | F-score combined: 0.4934 -> 0.5723 | Improved |
| add camera-aware embedding (ablation) | HOIT-T · BEHAVE | F-score combined: 0.5144 -> 0.5723 | Improved |
| add SMPL-H pose-fitting stage | HOIT-T · BEHAVE | F-score combined: 0.5723 -> 0.5497 | Hurt |
| add SMPL-H pose-fitting stage | HOIT-T · BEHAVE | contact recall: 0.684 (fit) with human mesh now parametric | Improved |
| use unified triplane representation instead of template-based vertex methods | HOIT-T · BEHAVE | F-score human: 0.5913 (HOI-TG) vs 0.5560 | Hurt |
| use unified triplane representation instead of template-based vertex methods | HOIT-T · InterCap | F-score combined: 0.3953 (HOI-TG) -> 0.6432 | Improved |
| train HOIT-VAE only on synthetic ProciGen and test on real data | HOIT-VAE · HODome | F-score combined: 0.9594 on unseen categories | Improved |
Closest papers by meaning
- End-to-End HOI Reconstruction Transformer with Graph-based Encoding (CVPR 2025 · similarity 0.89)
- Reconstructing In-the-Wild Open-Vocabulary Human-Object Interactions (CVPR 2025 · similarity 0.88)
- Template Free Reconstruction of Human-object Interaction with Procedural Interaction Generation (CVPR 2024 · similarity 0.87)
- CG-HOI: Contact-Guided 3D Human-Object Interaction Generation (CVPR 2024 · similarity 0.87)
- Real2Sim in HOI: Toward Physically Plausible HOI Reconstruction from Monocular Videos (NeurIPS 2026 · similarity 0.87)
- ContactField: Implicit Field Representation for Multi-Person Interaction Geometry (NeurIPS 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.
