Research map / Point clouds / Shape and surface completion
Autoregressive 3D Shape Generation via Canonical Mapping
An-Chieh Cheng, Xueting Li, Sifei Liu, Min Sun, Ming-Hsuan 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 |
|---|---|---|---|
| use canonical grouping instead of uniform grouping (G=128) (ablation) | Ours · ShapeNet Chair auto-encoding | Chamfer distance: 7.676 -> 6.177 | Improved |
| use canonical grouping instead of uniform grouping (G=16) (ablation) | Ours · ShapeNet Chair auto-encoding | Chamfer distance: 8.646 -> 7.542 | Improved |
| use canonical grouping instead of uniform grouping (G=256) (ablation) | Ours · ShapeNet Chair auto-encoding | Chamfer distance: 7.278 -> 7.933 | Hurt |
| increase groups from 128 to 256 with canonical grouping (ablation) | Ours · ShapeNet Chair auto-encoding | Chamfer distance: 6.177 -> 7.933 | Hurt |
| add dimension reduction 256->4 to codebook lookup (global codebook) (ablation) | Ours · ShapeNet Chair auto-encoding | Chamfer distance: 7.139 -> 6.298 | Improved |
| add grouped codebook on top of 256->4 reduction (ablation) | Ours · ShapeNet Chair auto-encoding | Chamfer distance: 6.298 -> 6.177 | Improved |
| add dimension reduction and grouped codebook (ablation) | Ours · ShapeNet Chair auto-encoding | Codebook usage (%): 11.72 -> 79.28 | Improved |
| use Fibonacci spiral order instead of random order (ablation) | Ours · ShapeNet Chair unconditional generation | MMD-CD: 7.44 -> 7.37 | Improved |
| use inverse spiral instead of spiral order (ablation) | Ours · ShapeNet Chair unconditional generation | MMD-CD: 7.37 -> 7.17 | Improved |
| raise top-p from 0.85 to 0.92 (ablation) | Ours · ShapeNet Chair unconditional generation | MMD-CD: 7.70 -> 7.37 | Improved |
| raise top-p from 0.92 to 0.99 (ablation) | Ours · ShapeNet Chair unconditional generation | COV-CD (%): 45.77 -> 44.86 | Mixed |
| raise sampling temperature from t=1 to t=3 (ablation) | Ours · ShapeNet Chair shape completion | TMD: 6.553 -> 13.341 | Improved |
Closest papers by meaning
- Learning to Tessellate: Point Cloud Generation via Recursive Spectral Partitioning (ECCV 2026 · similarity 0.90)
- PointNSP: Autoregressive 3D Point Cloud Generation with Next-Scale Level-of-Detail Prediction (CVPR 2026 · similarity 0.90)
- Learning Gradient Fields for Shape Generation (ECCV 2020 · similarity 0.90)
- EditVAE: Unsupervised Parts-Aware Controllable 3D Point Cloud Shape Generation (AAAI 2022 · similarity 0.90)
- Quartet of Diffusions: Structure-Aware Point Cloud Generation through Part and Symmetry Guidance (ICLR 2026 · similarity 0.89)
- BuildAnyPoint: 3D Building Structured Abstraction from Diverse Point Clouds (CVPR 2026 · similarity 0.89)
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.
