Research map / 3D vision / 3D generation
Repurposing 2D Diffusion Models with Gaussian Atlas for 3D Generation
Tiange Xiang, Kai Li, Chengjiang Long, Christian Häne, Peihong Guo, Scott Delp, Ehsan Adeli, Li Fei-Fei
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 Gaussian Atlas fine-tuned 2D diffusion instead of GaussianCube 3D diffusion for text-to-3D | GaussianAtlas · Text-to-3D (100 prompts) | CLIP score: 22.31 -> 23.20 | Improved |
| use Gaussian Atlas instead of GaussianCube on VQA score | GaussianAtlas · Text-to-3D (100 prompts) | VQA score: 0.52 -> 0.61 | Improved |
| fine-tune from pretrained 2D latent diffusion UNet instead of training from scratch (ablation) | GaussianAtlas · Text-to-3D training checkpoints | generation quality / convergence | Improved |
| use deterministic sphere plus equirectangular flattening instead of per-object optimization-based flattening (ablation) | GaussianAtlas · Text-to-3D training | convergence and generation quality | Improved |
| remove the VAE and fine-tune UNet directly on atlases (vs VAE auto-encoding) (ablation) | GaussianAtlas · Gaussian atlas reconstruction | PSNR of rendering | Improved |
| add visibility-ranking pruning to bound Gaussians in fitting | Scaffold-GS fitting · 3DGS fitting quality | rendering quality vs Gaussian count | Improved |
| increase number of Gaussians in fitting (ablation) | 3DGS fitting · 3DGS fitting quality | rendering quality | No clear effect |
| user pairwise preference of Gaussian Atlas vs TriplaneGaussian | GaussianAtlas · User study | win rate: 88% of users favoring our method | Improved |
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.
- GaussianCube — GaussianCube: A Structured and Explicit Radiance Representation for 3D Generative Modeling (NeurIPS 2024) · better on 4 of 4 shared results
- DreamGaussian — DreamGaussian: Generative Gaussian Splatting for Efficient 3D Content Creation (ICLR 2024) · better on 3 of 3 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
- Atlas Gaussians Diffusion for 3D Generation (ICLR 2025 · similarity 0.93)
- GaussianDreamer: Fast Generation from Text to 3D Gaussians by Bridging 2D and 3D Diffusion Models (CVPR 2024 · similarity 0.90)
- Prometheus: 3D-Aware Latent Diffusion Models for Feed-Forward Text-to-3D Scene Generation (CVPR 2025 · similarity 0.89)
- Geometry Image Diffusion: Fast and Data-Efficient Text-to-3D with Image-Based Surface Representation (ICLR 2025 · similarity 0.89)
- Diffusion$^2$: Dynamic 3D Content Generation via Score Composition of Video and Multi-view Diffusion Models (ICLR 2025 · similarity 0.88)
- LT3SD: Latent Trees for 3D Scene Diffusion (CVPR 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.
