Research map / 3D vision / 3D generation
Stable Virtual Camera: Generative View Synthesis with Diffusion Models
Jensen Zhou, Hang Gao, Vikram Voleti, Aaryaman Vasishta, Chun-Han Yao, Mark Boss, Philip Torr, Christian Rupprecht, 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 interp procedural two-pass sampling instead of one-pass sampling for trajectory NVS (ablation) | SEVA · DL3DV trajectory NVS (P=3) | motion smoothness: 95.39 -> 95.76 | Improved |
| use gt+nearest instead of nearest procedural sampling (append input views to each pass) for set NVS (ablation) | SEVA · DL3DV trajectory NVS (P=3) | PSNR: 13.74 -> 15.58 | Improved |
| add optional temporal pathway (3D convs) at inference for ordered frames (ablation) | SEVA · DL3DV trajectory NVS (P=3) | TSED: 116.7 -> 109.0 | Improved |
| pad forward pass to T frames by repeating first input image instead of changing context window zero-shot (ablation) | SEVA · sparse-view NVS (P+Q <= T) | sample quality | Improved |
| extend context window T zero-shot with one-pass sampling (ablation) | SEVA · T&T set NVS, semi-dense views (P>=9) | PSNR / 3D consistency | Improved |
| extend context window T zero-shot in the sparse-view regime (ablation) | SEVA · T&T set NVS, sparse views (P<9) | Image Quality | Hurt |
| use nearest-K anchor procedural sampling for many input views instead of one-pass (ablation) | SEVA · T&T set NVS, P>T | PSNR | Hurt |
| retrieve spatially nearest anchors from a memory bank instead of temporally nearest ones (ablation) | SEVA · long-trajectory NVS (telephone booth loops) | long-range 3D consistency | Improved |
| sweep camera normalization unit length per scene for single-view input | SEVA · RealEstate10K single view (P=1) | PSNR | Improved |
| higher CFG scale under high generation uncertainty, lower scale under low uncertainty (ablation) | SEVA · qualitative samples | sample quality | Mixed |
| diffusion model without 3D distillation versus regression models on small-viewpoint interpolation (P=1 RE10K) | SEVA · RealEstate10K (D split, P=1) | PSNR: 17.99 vs 20.90 (DepthSplat) | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 4 of 4 on most shared results.
- DepthSplat — DepthSplat: Connecting Gaussian Splatting and Depth (CVPR 2025) · better on 4 of 5 shared results
- ViewCrafter — ViewCrafter: Taming Video Diffusion Models for High-fidelity Novel View Synthesis (TPAMI 2025) · better on 2 of 3 shared results
- CAT3D — CAT3D: Create Anything in 3D with Multi-View Diffusion Models (NeurIPS 2024) · better on 3 of 3 shared results
- MVSplat — MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images (ECCV 2024) · better on 1 of 1 shared result
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
- NVS-Solver: Video Diffusion Model as Zero-Shot Novel View Synthesizer (ICLR 2025 · similarity 0.90)
- ViVid-1-to-3: Novel View Synthesis with Video Diffusion Models (CVPR 2024 · similarity 0.90)
- Dynamic View Synthesis as an Inverse Problem (NeurIPS 2025 · similarity 0.89)
- Novel View Synthesis with Pixel-Space Diffusion Models (CVPR 2025 · similarity 0.89)
- NAMVIS: Next-Scale Autoregressive Multi-View Image Synthesis (NeurIPS 2026 · similarity 0.89)
- MVGenMaster: Scaling Multi-View Generation from Any Image via 3D Priors Enhanced Diffusion Model (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.
