Research map / Image generation / Fast diffusion sampling
HyperDiffusion: Generating Implicit Neural Fields with Weight-Space Diffusion
Ziya Erkoç, Fangchang Ma, Qi Shan, Matthias Nießner, Angela Dai
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 |
|---|---|---|---|
| initialize all MLPs from the first optimized MLP instead of random init (ablation) | HyperDiffusion · ShapeNet Airplane | FPD: 4.01 -> 3.49 | Improved |
| initialize all MLPs from the first optimized MLP instead of random init (ablation) | HyperDiffusion · ShapeNet Airplane | MMD: 3.34 -> 3.45 | Hurt |
| add positional encoding to the per-sample MLP (ablation) | HyperDiffusion · ShapeNet Airplane | FPD: 6.40 -> 3.49 | Improved |
| add positional encoding to the per-sample MLP (ablation) | HyperDiffusion · ShapeNet Airplane | 1-NNA (%): 76.5 -> 69.3 | Improved |
| diffuse MLP weights instead of voxel grids | unconditional 4D generation · DeformingThings4D | 1-NNA (%): 85 -> 62 | Improved |
| diffuse MLP weights instead of voxel grids | unconditional 3D generation · ShapeNet Airplane | FPD: 38.9 -> 3.5 | Improved |
| weight-space diffusion versus point-cloud diffusion baselines (PVD, DPC) | unconditional 3D generation · ShapeNet Chair | FPD: 3.5 -> 1.7 | Improved |
| weight-space diffusion versus point-cloud diffusion baselines on MMD | unconditional 3D generation · ShapeNet Chair | MMD: 6.3 -> 7.1 | Hurt |
| add a naive secondary surface reconstruction loss to the diffusion | HyperDiffusion · 3D shape generation | generation quality | No clear effect |
Later papers that compared with this method
3 later accepted papers report HyperDiffusion as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Weight Space Representation Learning via Neural Field Adaptation (CVPR 2026 · better on 4 of 4 shared results)
- Few-shot Implicit Function Generation via Equivariance (CVPR 2025 · better on 6 of 6 shared results)
- DDMI: Domain-agnostic Latent Diffusion Models for Synthesizing High-Quality Implicit Neural Representations (ICLR 2024 · 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
- Implicit-ARAP: Efficient Handle-Guided Neural Field Deformation via Local Patch Meshing (NeurIPS 2025 · similarity 0.86)
- Single Mesh Diffusion Models with Field Latents for Texture Generation (CVPR 2024 · similarity 0.85)
- Scaling Diffusion Transformers Efficiently via $\mu$P (NeurIPS 2025 · similarity 0.85)
- Efficient-3Dim: Learning a Generalizable Single-image Novel-view Synthesizer in One Day (ICLR 2024 · similarity 0.85)
- Diffusing Differentiable Representations (NeurIPS 2024 · similarity 0.85)
- Denoising Diffusion via Image-Based Rendering (ICLR 2024 · similarity 0.84)
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.
