Research map / 3D vision / 3D generation
Generative Novel View Synthesis with 3D-Aware Diffusion Models
Eric R. Chan, Koki Nagano, Matthew A. Chan, Alexander W. Bergman, Jeong Joon Park, Axel Levy, Miika Aittala, Shalini De Mello, Tero Karras, Gordon Wetzstein
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 |
|---|---|---|---|
| add 3D feature volume geometry prior instead of geometry-free input concatenation (ablation) | 3D-aware diffusion NVS · ShapeNet cars (single view) | FID: 16.68 -> 6.47 | Improved |
| add geometry prior (feature volume) instead of geometry-free (ablation) | 3D-aware diffusion NVS · ShapeNet cars (single view) | LPIPS: 0.342 -> 0.104 | Improved |
| add MLP decoder to the feature volume instead of explicit representation (ablation) | 3D-aware diffusion NVS · ShapeNet cars (single view) | FID: 8.09 -> 6.47 | Improved |
| use iterative diffusion sampling instead of one-step inference (ablation) | 3D-aware diffusion NVS · ShapeNet cars (single view) | FID: 42.07 -> 6.47 | Improved |
| use iterative diffusion sampling instead of one-step inference (ablation) | 3D-aware diffusion NVS · ShapeNet cars (single view) | PSNR: 23.2 -> 20.7 | Hurt |
| use iterative diffusion sampling instead of one-step inference (ablation) | 3D-aware diffusion NVS · CO3D hydrant | KID: 0.106 -> 0.012 | Improved |
| add autoregressive conditioning on previously generated frames (ablation) | 3D-aware diffusion NVS · ShapeNet cars (single view) | FID: 6.47 -> 11.08 | Hurt |
| add autoregressive conditioning for sequence generation (ablation) | 3D-aware diffusion NVS · generated sequences (qualitative) | geometric consistency | Improved |
| add classifier-free guidance during autoregressive generation (ablation) | autoregressive sequence generation · ShapeNet (qualitative) | sequence quality | Hurt |
| use feature-volume aggregation instead of stochastic conditioning (ablation) | autoregressive generation · COLMAP 3D consistency (qualitative) | denoising steps per view: 256 -> 25 steps (10x speed) | Improved |
| replace mean-pooling with max-pooling or weighted pooling for multi-view aggregation (ablation) | multi-view feature aggregation · multi-view conditioning | quality | No clear effect |
| add more input views (ablation) | 3D-aware diffusion NVS · ShapeNet cars (qualitative) | pixel-wise std of novel renderings | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ViewFormer — ViewFormer: NeRF-Free Neural Rendering from Few Images Using Transformers (ECCV 2022) · better on 2 of 2 shared results
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
- ViVid-1-to-3: Novel View Synthesis with Video Diffusion Models (CVPR 2024 · similarity 0.92)
- Efficient-3Dim: Learning a Generalizable Single-image Novel-view Synthesizer in One Day (ICLR 2024 · similarity 0.92)
- UMAMI: Unifying Masked Autoregressive Models and Deterministic Rendering for View Synthesis (NeurIPS 2025 · similarity 0.92)
- A 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale (NeurIPS 2026 · similarity 0.91)
- Novel View Synthesis with Pixel-Space Diffusion Models (CVPR 2025 · similarity 0.91)
- LT3SD: Latent Trees for 3D Scene Diffusion (CVPR 2025 · similarity 0.91)
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.
