Research map / 3D vision / 3D generation
TetraDiffusion: Tetrahedral Diffusion Models for 3D Shape Generation
Nikolai Kalischek, Torben Peters, Jan Dirk Wegner, Konrad Schindler
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 |
|---|---|---|---|
| prune unoccupied tetrahedra from the grid (ablation) | TetraDiffusion R=128 · Training memory | GPU memory (GB): 20.8 -> 12.0 | Improved |
| prune unoccupied tetrahedra from the grid (ablation) | TetraDiffusion R=192 · Training memory | GPU memory (GB): 78.2 -> 20.9 | Improved |
| convolve natively on tetrahedral grid instead of embedding in a voxel grid (MeshDiffusion) | TetraDiffusion · Training memory at R=128 | GPU memory (GB): 76.6 -> 20.8 | Improved |
| native tetrahedral convolution instead of MeshDiffusion voxel embedding | TetraDiffusion · Inference | s/obj: 714.3 -> 3.4 | Improved |
| randomize neighbor ordering in tetrahedral convolution (ablation) | TetraDiffusion · ShapeNet motorbike | convergence / sample quality | Hurt |
| replace TetraConv with permutation-invariant graph convolutions, even with matched parameter count (ablation) | TetraDiffusion · ShapeNet motorbike | convergence / sample quality | Hurt |
| increase native resolution to R=192 versus R=128 baselines (ablation) | TetraDiffusion · Shape quality | shading-FID | Improved |
| reduce sampling to 32 diffusion steps (ablation) | TetraDiffusion · Supplement step-count study | mesh quality | No clear effect |
| TetraDiffusion versus NWD on chairs | TetraDiffusion · ShapeNet Chair | shading-FID: 16.7 -> 39.9 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- GET3D — GET3D: A Generative Model of High Quality 3D Textured Shapes Learned from Images (NeurIPS 2022) · better on 3 of 4 shared results
- MeshDiffusion — MeshDiffusion: Score-based Generative 3D Mesh Modeling (ICLR 2023) · better on 1 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
- Tetrahedron Splatting for 3D Generation (NeurIPS 2024 · similarity 0.89)
- Single Mesh Diffusion Models with Field Latents for Texture Generation (CVPR 2024 · similarity 0.87)
- Cycle3D: High-quality and Consistent Image-to-3D Generation via Generation-Reconstruction Cycle (AAAI 2025 · similarity 0.87)
- Sin3DM: Learning a Diffusion Model from a Single 3D Textured Shape (ICLR 2024 · similarity 0.87)
- Diffusion$^2$: Dynamic 3D Content Generation via Score Composition of Video and Multi-view Diffusion Models (ICLR 2025 · similarity 0.87)
- DMV3D: Denoising Multi-view Diffusion Using 3D Large Reconstruction Model (ICLR 2024 · similarity 0.87)
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.
