Research map / 3D vision / 3D generation
A Lesson in Splats: Teacher-Guided Diffusion for 3D Gaussian Splats Generation with 2D Supervision
Chensheng Peng, Ido Sobol, Masayoshi Tomizuka, Kurt Keutzer, Chenfeng Xu, Or Litany
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 cycle consistency regularization in stage I (ablation) | SplatDiffusion · ShapeNet-SRN Cars validation (1-view ablation) | PSNR: 22.6078 -> 23.7293 | Improved |
| add cycle consistency regularization in stage II (ablation) | SplatDiffusion · ShapeNet-SRN Cars validation (1-view ablation) | PSNR: 24.6897 -> 24.9137 | Improved |
| add multi-step fine-tuning stage II with rendering loss only after stage I (ablation) | SplatDiffusion · ShapeNet-SRN Cars validation (1-view ablation) | PSNR: 22.6078 -> 24.4936 | Improved |
| use rendering loss only rather than diffusion plus rendering in stage II (ablation) | SplatDiffusion · ShapeNet-SRN Cars validation (1-view ablation) | PSNR: 23.1323 -> 24.4936 | Improved |
| add diffusion loss to rendering loss in stage I (teacher bootstrapping) (ablation) | SplatDiffusion · ShapeNet-SRN Cars validation (1-view ablation) | PSNR: 18.8201 -> 22.6078 | Improved |
| add per-timestep weighting to the stage II rendering loss (ablation) | SplatDiffusion · ShapeNet-SRN Cars validation (1-view ablation) | PSNR: 22.8848 -> 24.4936 | Improved |
| diffusion framework vs feedforward network of same Medium size (ablation) | SplatDiffusion · ShapeNet-SRN Cars validation (1-view ablation) | PSNR: 19.9947 -> 21.7506 | Improved |
| add additional-view guidance to the diffusion model (ablation) | SplatDiffusion · ShapeNet-SRN Cars subset (2-view) | PSNR: 25.18 -> 25.36 | Improved |
| add 3DGS optimization on top of guidance (ablation) | SplatDiffusion · ShapeNet-SRN Cars subset (2-view) | PSNR: 25.36 -> 25.55 | Improved |
| replace Splatter Image teacher with SplatDiffusion on ShapeNet-SRN Cars | SplatDiffusion · ShapeNet-SRN Cars (1-view) | PSNR: 24.00 -> 24.84 | Improved |
| replace Splatter Image teacher with SplatDiffusion on ShapeNet-SRN Chairs | SplatDiffusion · ShapeNet-SRN Chairs (1-view) | PSNR: 24.43 -> 25.21 | Improved |
| replace Flash3D teacher with SplatDiffusion on RealEstate10k | SplatDiffusion · RealEstate10k 5 frames | PSNR: 28.46 -> 29.12 | 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.
- Splatter Image (Large) — Splatter Image: Ultra-Fast Single-View 3D Reconstruction (CVPR 2024) · better on 6 of 6 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
- Taming Video Diffusion Prior with Scene-Grounding Guidance for 3D Gaussian Splatting from Sparse Inputs (CVPR 2025 · similarity 0.91)
- DiffSplat: Repurposing Image Diffusion Models for Scalable Gaussian Splat Generation (ICLR 2025 · similarity 0.90)
- Denoising Diffusion via Image-Based Rendering (ICLR 2024 · similarity 0.90)
- Wonderland: Navigating 3D Scenes from a Single Image (CVPR 2025 · similarity 0.90)
- Generative Sparse-View Gaussian Splatting (CVPR 2025 · similarity 0.90)
- 3D Student Splatting and Scooping (CVPR 2025 · similarity 0.89)
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.
