Research map / Video / Video generation
Fine-flow Distilling Coarse-flow Video Generation for Long-Term Driving World Model
Xiaodong Wang, Zhirong Wu, Peixi Peng
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 coarse-to-fine decoupled learning (ablation) | DiT · Short video prediction | FVD: 347.7 -> 321.8 | Improved |
| add structure preservation loss (ablation) | CDiT + coarse-to-fine · Short video prediction | FVD: 321.8 -> 300.3 | Improved |
| add fine-flow to coarse-flow distillation (ablation) | CDiT · Short video prediction | FVD: 300.3 -> 290.6 | Improved |
| warp-guided control instead of Vista-style conditioning (ablation) | CogVideoX · Trajectory-based prediction | FVD: 89.8 -> 69.6 | Improved |
| warp-guided control instead of Vista-style conditioning (ablation) | CogVideoX · Trajectory-based prediction | FID: 13.0 -> 12.7 | No clear effect |
| Fine DiT vs other interpolation models | frame interpolation · Frame interpolation | FVD: 147.6 -> 102.9 | Improved |
| use the model vs Vista on long rollouts | Vista · 113-frame prediction | FVD: 398.5 -> 289.5 | Improved |
| use the model vs Vista on short rollouts | Vista · 25-frame prediction | FVD: 174.1 -> 188.5 | Hurt |
| use the model vs Vista on image fidelity | Vista · NuScenes val | FID: 7.6 -> 12.3 | Hurt |
| parallel interpolation of long videos vs autoregressive rollouts | Vista · 113-frame inference time | seconds: 475s -> 70s | 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.
- DynamiCrafter — DynamiCrafter: Animating Open-domain Images with Video Diffusion Priors (ECCV 2024) · better on 2 of 2 shared results
- FRAMER — Framer: Interactive Frame Interpolation (ICLR 2025) · better on 1 of 1 shared result
- SEINE — SEINE: Short-to-Long Video Diffusion Model for Generative Transition and Prediction (ICLR 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
- DriveLaW: Unifying Planning and Video Generation in a Latent Driving World (CVPR 2026 · similarity 0.89)
- MaskGWM: A Generalizable Driving World Model with Video Mask Reconstruction (CVPR 2025 · similarity 0.88)
- From Slow Bidirectional to Fast Autoregressive Video Diffusion Models (CVPR 2025 · similarity 0.87)
- DriveDreamer4D: World Models Are Effective Data Machines for 4D Driving Scene Representation (CVPR 2025 · similarity 0.86)
- ARLON: Boosting Diffusion Transformers with Autoregressive Models for Long Video Generation (ICLR 2025 · similarity 0.86)
- ODEWorld: A Continuous Predictive Architecture via Physical-Time Flow (NeurIPS 2026 · 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.
