Research map / 3D vision / 3D generation
Novel View Synthesis as Video Completion
Qi Wu, Khiem Vuong, Minsik Jeon, Srinivasa G. Narasimhan, Deva Ramanan
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 |
|---|---|---|---|
| encode each input view independently as a 1-frame video (ablation) | FrameCrafter · DL3DV (192x336) | PSNR (dB): 11.50 -> 15.69 | Improved |
| use pixel unshuffle for Plucker ray maps (ablation) | FrameCrafter · DL3DV (192x336) | LPIPS: 0.359 -> 0.246 | Improved |
| supervise only the target frame (ablation) | FrameCrafter · DL3DV (192x336) | PSNR (dB): 15.08 -> 15.69 | Improved |
| remove temporal RoPE (ablation) | FrameCrafter · DL3DV (192x336) | PSNR (dB): 15.30 -> 15.69 | Improved |
| replace RoPE with PRoPE (ablation) | FrameCrafter · DL3DV (192x336) | PSNR (dB): 15.69 -> 14.91 | Hurt |
| query-centered camera frame and no temporal RoPE (ablation) | FrameCrafter · DL3DV permutations | PSNR std: 0.4730 -> 0.0081 | Improved |
| train with shuffled (permutation-invariant) instead of ordered inputs (ablation) | FrameCrafter · DL3DV ordered evaluation | PSNR (dB): 15.04 -> 15.69 | Improved |
| use video backbone Wan2.1-14B instead of CogVideoX-5B | FrameCrafter · DL3DV | PSNR (dB): 13.65 -> 17.18 | Improved |
| video diffusion with 1K scenes vs image diffusion EscherNet with 10K scenes | FrameCrafter · DL3DV | PSNR (dB): 12.07 -> 17.18 | Improved |
| FrameCrafter (Wan) vs SEVA on SSIM | FrameCrafter · DL3DV | SSIM: 0.470 -> 0.445 | Hurt |
| FrameCrafter (Wan) vs SEVA on DreamSim | FrameCrafter · Mip-NeRF 360 | DreamSim: 0.137 -> 0.111 | Improved |
| increase training scenes (20 to 2000) | FrameCrafter · DL3DV | PSNR (dB) | Improved |
Closest papers by meaning
- Novel View Synthesis from A Few Glimpses via Test-Time Natural Video Completion (NeurIPS 2025 · similarity 0.91)
- NVS-Solver: Video Diffusion Model as Zero-Shot Novel View Synthesizer (ICLR 2025 · similarity 0.90)
- UMAMI: Unifying Masked Autoregressive Models and Deterministic Rendering for View Synthesis (NeurIPS 2025 · similarity 0.90)
- Novel View Synthesis with Pixel-Space Diffusion Models (CVPR 2025 · similarity 0.90)
- ViVid-1-to-3: Novel View Synthesis with Video Diffusion Models (CVPR 2024 · similarity 0.90)
- MVSplat360: Feed-Forward 360 Scene Synthesis from Sparse Views (NeurIPS 2024 · 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.
