Research map / 3D vision / Face and shape
Generative 3D Part Assembly via Dynamic Graph Learning
jialei huang, Guanqi Zhan, Qingnan Fan, Kaichun Mo, Lin Shao, Baoquan Chen, Leonidas Guibas, Hao Dong
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 |
|---|---|---|---|
| replace per-part MLP with iterative graph learning (ablation) | backbone · PartNet Table | Part accuracy: 26.05 -> 42.09 | Improved |
| add dynamic relation reasoning to backbone (ablation) | backbone · PartNet Table | Part accuracy: 42.09 -> 46.85 | Improved |
| add dynamic part aggregation to backbone (ablation) | backbone · PartNet Table | Part accuracy: 42.09 -> 48.01 | Improved |
| add both relation reasoning and part aggregation (ablation) | backbone · PartNet Table | Part accuracy: 42.09 -> 49.51 | Improved |
| learn relation weights instead of static GT adjacency (ablation) | full model · PartNet Table | Part accuracy: 45.43 -> 49.51 | Improved |
| reason relation from poses instead of transformed geometry (ablation) | full model · PartNet Table | Part accuracy: 45.11 -> 49.51 | Improved |
| swap order of dense and sparse node sets (ablation) | full model · PartNet Table | Part accuracy: 49.51 -> 49.19 | No clear effect |
| use max pooling instead of average pooling in aggregation (ablation) | full model · PartNet Table | Part accuracy: 48.92 -> 49.51 | Improved |
| use 5 GNN iterations instead of 3 or 7 (ablation) | full model · PartNet Table | Part accuracy: 45.51/49.51/44.6 for 3/5/7 | Mixed |
| add translation loss (ablation) | full model · PartNet Table | Part accuracy: 24.48 -> 49.51 | Improved |
| add rotation loss (ablation) | full model · PartNet Table | Part accuracy: 46.94 -> 49.51 | Improved |
| add shape loss (ablation) | full model · PartNet Table | Part accuracy: 48.23 -> 49.51 | Improved |
Closest papers by meaning
- Imagine: Image-Guided 3D Part Assembly with Structure Knowledge Graph (AAAI 2025 · similarity 0.92)
- Category-Level Multi-Part Multi-Joint 3D Shape Assembly (CVPR 2024 · similarity 0.91)
- Generative 3D Part Assembly via Part-Whole-Hierarchy Message Passing (CVPR 2024 · similarity 0.91)
- Neural Assembler: Learning to Generate Fine-Grained Robotic Assembly Instructions from Multi-View Images (AAAI 2025 · similarity 0.91)
- Two by Two: Learning Multi-Task Pairwise Objects Assembly for Generalizable Robot Manipulation (CVPR 2025 · similarity 0.89)
- IKEA Manuals at Work: 4D Grounding of Assembly Instructions on Internet Videos (NeurIPS 2024 · 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.
