Research map / Video / Video generation
RAYNOVA: Scale-Temporal Autoregressive World Modeling in Ray Space
Yichen Xie, Chensheng Peng, Mazen Abdelfattah, Yihan Hu, Jiezhi Yang, Eric Higgins, Ryan Brigden, Masayoshi Tomizuka, Wei Zhan
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 global spatio-temporal attention with relative ray embedding (ablation) | RayNova (130M) · nuScenes val multi-view generation | FVD: 214 -> 124 | Improved |
| use relative instead of absolute Plucker ray position embedding (ablation) | RayNova (130M) · nuScenes val multi-view generation | FVD: 138 -> 124 | Improved |
| use relative instead of absolute Plucker ray position embedding (ablation) | RayNova (130M) · nuScenes val multi-view generation | FID: 17.2 -> 17.2 | No clear effect |
| condition past frames on all scales instead of prefix scales (ablation) | RayNova (130M) · nuScenes val multi-view generation | FID: 17.2 -> 60.4 | Hurt |
| condition past frames on same scale only instead of prefix scales (ablation) | RayNova (130M) · nuScenes val multi-view generation | FVD: 124 -> 151 | Hurt |
| scale model from 130M to 2.8B parameters (ablation) | RayNova · nuScenes val multi-view generation | FID: 17.2 -> 10.5 | Improved |
| scale model from 130M to 2.8B parameters (ablation) | RayNova · nuScenes val multi-view generation | FVD: 124 -> 91 | Improved |
| add recurrent training stage (ablation) | RayNova · nuScenes val multi-view generation | FVD: 100 -> 91 | Improved |
| shift camera 4m (novel view) versus no shift, no 3D prior | RayNova · nuScenes novel view synthesis | FID: 14.11 -> 17.48 (1m vs 4m) | No clear effect |
| geometry-free ray-space model instead of 3D-prior reconstruction (OmniRe) | novel view synthesis · nuScenes novel view synthesis, shift 4m | FVD: 428.20 -> 143.48 | Improved |
| autoregressive architecture instead of diffusion baselines | multi-view world model · nuScenes val multi-view generation | throughput (images/s): 0.67 -> 1.96 (Panacea vs RayNova) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- Panacea — Panacea: Panoramic and Controllable Video Generation for Autonomous Driving (CVPR 2024) · better on 3 of 3 shared results
- DriveDreamer — DriveDreamer: Towards Real-world-driven World Models for Autonomous Driving (ECCV 2024) · better on 2 of 2 shared results
- DrivingDiffusion — DrivingDiffusion: Layout-Guided Multi-View Driving Scenarios Video Generation with Latent Diffusion Model (ECCV 2024) · better on 2 of 2 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
- SAMPO: Scale-wise Autoregression with Motion Prompt for Generative World Models (NeurIPS 2025 · similarity 0.89)
- Video World Models with Long-term Spatial Memory (NeurIPS 2025 · similarity 0.89)
- WorldStereo: Bridging Camera-Guided Video Generation and Scene Reconstruction via 3D Geometric Memories (CVPR 2026 · similarity 0.88)
- Beyond Pixel Histories: World Models with Persistent 3D State (ICML 2026 · similarity 0.88)
- PanoWorld: Geometry-Consistent Panoramic Video World Modeling (NeurIPS 2026 · similarity 0.87)
- UMAMI: Unifying Masked Autoregressive Models and Deterministic Rendering for View Synthesis (NeurIPS 2025 · similarity 0.86)
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.
