Research map / 3D vision / 3D generation
CrossHOI: Learning Cross-View Representations for Monocular 3D Human-Object Interaction Reconstruction
Pei Geng, Shanshan Zhang, Jian Yang
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 cross-view generated features to CONTHO-style pipeline (ablation) | CrossHOI · BEHAVE | CD human: 5.13 -> 4.27 | Improved |
| add cross-view features in initial reconstruction (ablation) | CONTHO reimplementation · BEHAVE | CD human: 5.13 -> 4.81 | Improved |
| add cross-view features in contact estimation (ablation) | CrossHOI · BEHAVE | Contact precision: 0.648 -> 0.679 | Improved |
| add cross-view features in refinement (ablation) | CrossHOI · BEHAVE | CD human: 4.36 -> 4.27 | Improved |
| bidirectional vertex fusion instead of A->B only (ablation) | CrossHOI · BEHAVE | Contact recall: 0.541 -> 0.576 | Improved |
| use real view A as query rather than generated view B (ablation) | CrossHOI · BEHAVE | CD human: 4.85 -> 4.57 | Improved |
| spatial cross-attention fusion instead of weighted sum (ablation) | CrossHOI · BEHAVE | Contact recall: 0.552 -> 0.576 | Improved |
| spatial cross-attention fusion instead of element-wise add (ablation) | CrossHOI · BEHAVE | CD human: 4.47 -> 4.27 | Improved |
| cross-view features vs CONTHO on occluded samples | CrossHOI · BEHAVE occluded subset (500 samples) | Contact precision: 0.573 -> 0.629 | Improved |
| CrossHOI vs HOI-TG | CrossHOI · BEHAVE | CD human: 4.59 -> 4.27 | Improved |
| CrossHOI vs HOI-TG | CrossHOI · InterCap | Contact recall: 0.473 -> 0.491 | Improved |
Closest papers by meaning
- Reconstructing In-the-Wild Open-Vocabulary Human-Object Interactions (CVPR 2025 · similarity 0.90)
- End-to-End HOI Reconstruction Transformer with Graph-based Encoding (CVPR 2025 · similarity 0.90)
- Real2Sim in HOI: Toward Physically Plausible HOI Reconstruction from Monocular Videos (NeurIPS 2026 · similarity 0.89)
- PICO: Reconstructing 3D People In Contact with Objects (CVPR 2025 · similarity 0.88)
- InteractVLM: 3D Interaction Reasoning from 2D Foundational Models (CVPR 2025 · similarity 0.88)
- ContactField: Implicit Field Representation for Multi-Person Interaction Geometry (NeurIPS 2024 · similarity 0.88)
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.
