Research map / 3D vision / Face and shape
PolyGen: An Autoregressive Generative Model of 3D Meshes
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 discrete rather than continuous vertex embeddings (ablation) | PolyGen vertex model · ShapeNet Core V2 test | bits per vertex (vertices): 2.56 -> 2.46 | Helped |
| Remove data augmentation (ablation) | PolyGen · ShapeNet Core V2 test | bits per vertex (vertices): 2.46 -> 3.39 | Hurt |
| Remove data augmentation (ablation) | PolyGen · ShapeNet Core V2 test | bits per vertex (faces): 1.79 -> 2.52 | Hurt |
| Add valid-prediction masking at evaluation (ablation) | PolyGen · ShapeNet Core V2 test | bits per vertex (vertices): 2.47 -> 2.46 | No clear effect |
| Add valid-prediction masking at evaluation (ablation) | PolyGen · ShapeNet Core V2 test | bits per vertex (faces): 1.82 -> 1.79 | Helped |
| Add cross-attention to the face model (ablation) | PolyGen · ShapeNet Core V2 test | bits per vertex (faces): 1.79 -> 1.87 | Hurt |
| Use a mixture of discretized logistics to predict whole vertices (ablation) | PolyGen vertex model · ShapeNet Core V2 test | bits per vertex (vertices): 2.46 -> 3.01 | Hurt |
| Use a MADE decoder for coordinate triplets (ablation) | PolyGen vertex model · ShapeNet Core V2 test | bits per vertex (vertices): 2.46 -> 2.65 | Hurt |
| Use a Transformer decoder over coordinate components (ablation) | PolyGen vertex model · ShapeNet Core V2 test | bits per vertex (vertices): 2.46 -> 2.50 | Hurt |
| Add Transformer-based triplet embeddings (ablation) | PolyGen vertex model · ShapeNet Core V2 test | bits per vertex (vertices): 2.46 -> 2.48 | No clear effect |
| Condition vertex generation on class labels | PolyGen · ShapeNet Core V2 test | bits per vertex (vertices): 2.46 -> 2.43 | Helped |
| Condition vertex generation on images | PolyGen · ShapeNet Core V2 test | bits per vertex (vertices): 2.46 -> 2.30 | Helped |
Closest papers by meaning
- ARMesh: Autoregressive Mesh Generation via Next-Level-of-Detail Prediction (NeurIPS 2025 · similarity 0.91)
- PartCrafter: Structured 3D Mesh Generation via Compositional Latent Diffusion Transformers (NeurIPS 2025 · similarity 0.90)
- MeshGen: Generating PBR Textured Mesh with Render-Enhanced Auto-Encoder and Generative Data Augmentation (CVPR 2025 · similarity 0.89)
- AutoPartGen: Autoregressive 3D Part Generation and Discovery (NeurIPS 2025 · similarity 0.89)
- GALA: Geometry-Aware Local Adaptive Grids for Detailed 3D Generation (ICLR 2025 · similarity 0.89)
- MeshAnything: Artist-Created Mesh Generation with Autoregressive Transformers (ICLR 2025 · 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.