Research map / 3D vision / Face and shape
PQ-NET: A Generative Part Seq2Seq Network for 3D Shapes
Rundi Wu, Yixin Zhuang, Kai Xu, Hao Zhang, Baoquan Chen
Check 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 |
|---|---|---|---|
| Represent shapes as sequential assemblies of part geometry and bounding boxes | PQ-NET · PartNet shape autoencoding | CD: 3.64 -> 3.38 (Chair); 6.75 -> 5.49 (Table); 12.43 -> 11.49 (Lamp) | Improved |
| Represent shapes as sequential assemblies of part geometry and bounding boxes | PQ-NET · PartNet shape autoencoding | IoU: 62.93 -> 67.29 (Chair); 56.14 -> 47.39 (Table); 41.29 -> 39.56 (Lamp) | Mixed |
| Train PQ-NET across all three categories rather than per category (ablation) | PQ-NET · PartNet Chair, Table, and Lamp shape autoencoding | CD: 2.86 -> 2.46 (Chair); 5.69 -> 4.50 (Table); 10.32 -> 4.87 (Lamp) | Improved |
| Train PQ-NET across all three categories rather than per category (ablation) | PQ-NET · PartNet shape autoencoding | LFD: 2441 -> 2501 (Chair); 2609 -> 2415 (Table); 5941 -> 4875 (Lamp) | Mixed |
| Increase reconstruction resolution from 64^3 to 256^3 (ablation) | PQ-NET · PartNet shape autoencoding | CD: 3.38 -> 2.86 (Chair); 5.49 -> 5.69 (Table); 11.49 -> 10.32 (Lamp) | Mixed |
| Use PartNet natural part order instead of the top-down order (ablation) | PQ-NET · PartNet structured reconstruction from depth | IoU: 58.68 -> 61.47 (Chair); 48.58 -> 53.67 (Table); 52.17 -> 52.94 (Lamp) | Improved |
| Use PQ-NET instead of 3D-PRNN | PartNet structured reconstruction from depth · Chair, Table, and Lamp | IoU: 37.26 -> 61.47 (Chair); 51.30 -> 53.67 (Table); 47.26 -> 52.94 (Lamp) | Improved |
| Use PQ-NET instead of IM-NET | PartNet shape generation · Chair | MMD: 7.44 -> 8.34 | Hurt |
| Use PQ-NET instead of IM-NET | PartNet shape generation · Table | JSD: 0.0047 -> 0.0057 | Hurt |
| Use PQ-NET instead of IM-NET | PartNet shape generation · Lamp | COV: 81.25 -> 87.95 | Improved |
| Use PQ-NET instead of IM-NET | PartNet shape generation · Chair | JSD: 0.0084 -> 0.0083 | Improved |
| Use PQ-NET instead of 3D-PRNN | PartNet shape generation · Chair, Table, and Lamp | COV-IoU: 51.03 -> 58.94 (Chair); 36.44 -> 59.80 (Table); 58.93 -> 76.34 (Lamp) | Improved |
Later papers that compared with this method
1 later accepted paper reports PQ-Net as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- LSD-StructureNet: Modeling Levels of Structural Detail in 3D Part Hierarchies (ICCV 2021 · better on 0 of 1 shared result)
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
- Learning Part Generation and Assembly for Structure-Aware Shape Synthesis (AAAI 2020 · similarity 0.88)
- CompoNet: Learning to Generate the Unseen by Part Synthesis and Composition (ICCV 2019 · similarity 0.88)
- Generative 3D Part Assembly via Part-Whole-Hierarchy Message Passing (CVPR 2024 · similarity 0.87)
- PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers (NeurIPS 2025 · similarity 0.86)
- PoNQ: a Neural QEM-based Mesh Representation (CVPR 2024 · similarity 0.86)
- Learning Partonomic 3D Reconstruction from Image Collections (CVPR 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.
