Research map / 3D vision / 3D generation
SV3D: Novel Multi-view Synthesis and 3D Generation from a Single Image using Latent Video Diffusion
Vikram Voleti, Chun-Han Yao, Mark Boss, Adam Letts, David Pankratz, Dmitrii Tochilkin, Christian Laforte, Robin Rombach, Varun Jampani
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 pose-conditioned SV3D^c or SV3D^p instead of pose-unconditioned SV3D^u for static orbit synthesis (ablation) | SV3D · GSO static orbits | PSNR: 21.14 (u) vs 20.56 (c) vs 21.26 (p) | Mixed |
| progressive finetuning from static to dynamic orbits (SV3D^p vs SV3D^c) (ablation) | SV3D · GSO dynamic orbits | PSNR: 19.99 -> 20.38 | Improved |
| add masked SDS loss to photometric reconstruction for 3D generation (ablation) | SV3D^p · GSO 3D generation | PSNR: 16.864 -> 17.405 | Improved |
| add masked SDS loss (ablation) | SV3D^p · GSO 3D generation | Chamfer distance: 0.024 -> 0.024 | No clear effect |
| use dynamic orbit instead of static orbit for 3D reconstruction (ablation) | SV3D^p · GSO 3D generation | PSNR: 16.821 -> 17.405 | Improved |
| use triangular CFG scaling (1 to 2.5 to 1) instead of linear 1 to 4 (ablation) | SV3D · Qualitative orbit videos | back-view detail and over-sharpening | Improved |
| add disentangled spherical-Gaussian illumination model (ablation) | SV3D 3D optimization · Qualitative (school bus) | baked-in illumination | Improved |
| fine-tune a video diffusion model (SVD) instead of image diffusion for NVS | SV3D · GSO static orbits | LPIPS: 0.13 (Stable Zero123) -> 0.09 (SV3D^u) | Improved |
| naive SDS loss without masking (ablation) | SV3D 3D optimization · Qualitative | texture faithfulness and stability | Hurt |
Closest papers by meaning
- SV4D: Dynamic 3D Content Generation with Multi-Frame and Multi-View Consistency (ICLR 2025 · similarity 0.93)
- ViVid-1-to-3: Novel View Synthesis with Video Diffusion Models (CVPR 2024 · similarity 0.90)
- Diffusion$^2$: Dynamic 3D Content Generation via Score Composition of Video and Multi-view Diffusion Models (ICLR 2025 · similarity 0.89)
- Ouroboros3D: Image-to-3D Generation via 3D-aware Recursive Diffusion (CVPR 2025 · similarity 0.89)
- Era3D: High-Resolution Multiview Diffusion using Efficient Row-wise Attention (NeurIPS 2024 · similarity 0.89)
- A 3D Scene is Worth 1K Tokens: 3D-Grounded Representation for Scene Generation at Scale (NeurIPS 2026 · 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.
