Research map / 3D vision / 3D generation
PixGS: Pixel-Space Diffusion for Direct 3D Gaussian Splat Generation
Cao Duy, Phong Nguyen
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 appearance and rotation regularization to flow matching loss (ablation) | PixGS · G-Objaverse-XL ablation subset | PSNR: 16.04 -> 30.11 | Improved |
| add rotation norm regularization to appearance loss (ablation) | PixGS · G-Objaverse-XL ablation subset | PSNR: NaN -> 30.11 | Improved |
| add geometry loss (ablation) | PixGS · G-Objaverse-XL ablation subset | PSNR: 30.11 -> 30.24 | Improved |
| add multi-scale LoG loss (ablation) | PixGS · G-Objaverse-XL ablation subset | LPIPS: 0.025 -> 0.022 | Improved |
| add ground-truth rendering supervision in phase 2 after flow matching phase 1 (ablation) | PixGS · G-Objaverse-XL ablation subset | PSNR: 16.04 -> 30.36 | Improved |
| add GT-only phase 3 after phase 2 (ablation) | PixGS · G-Objaverse-XL ablation subset | PSNR: 30.36 -> 31.15 | Improved |
| use View Concatenation instead of Image Prompt Adapter (ablation) | PixGS · G-Objaverse-XL ablation subset | PSNR: 30.31 -> 30.36 | No clear effect |
| use View Concatenation instead of Image Prompt Adapter (ablation) | PixGS · G-Objaverse-XL ablation subset | SSIM: 0.977 -> 0.969 | Hurt |
| use GSRecon pseudo-labels instead of LGM pseudo-labels (ablation) | PixGS · GSO | PSNR: 21.15 -> 21.21 | No clear effect |
| scale training data from 265K to 1M assets | PixGS · GSO (image-conditioned) | PSNR: 20.67 -> 21.13 | Improved |
| PixGS vs DiffSplat SD-3.5 on image-conditioned generation | PixGS · GSO | PSNR: 19.59 -> 21.21 | Improved |
| PixGS vs best baseline on text single object | PixGS · T3Bench single object | CLIP Sim (%): 30.99 -> 31.98 | 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.
- DreamGaussian — DreamGaussian: Generative Gaussian Splatting for Efficient 3D Content Creation (ICLR 2024) · better on 3 of 3 shared results
- GaussianCube — GaussianCube: A Structured and Explicit Radiance Representation for 3D Generative Modeling (NeurIPS 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
- DiffSplat: Repurposing Image Diffusion Models for Scalable Gaussian Splat Generation (ICLR 2025 · similarity 0.92)
- UVGS: Reimagining Unstructured 3D Gaussian Splatting using UV Mapping (CVPR 2025 · similarity 0.91)
- 3DGS-Enhancer: Enhancing Unbounded 3D Gaussian Splatting with View-consistent 2D Diffusion Priors (NeurIPS 2024 · similarity 0.91)
- DiffGS: Functional Gaussian Splatting Diffusion (NeurIPS 2024 · similarity 0.90)
- GaussianSpa: An "Optimizing-Sparsifying" Simplification Framework for Compact and High-Quality 3D Gaussian Splatting (CVPR 2025 · similarity 0.90)
- ContextGS : Compact 3D Gaussian Splatting with Anchor Level Context Model (NeurIPS 2024 · similarity 0.90)
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.
