Research map / Video / Video generation
Self-Refining Video Sampling
Sangwon Jang, Taekyung Ki, Jaehyeong Jo, Saining Xie, Jaehong Yoon, Sung Ju Hwang
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 P&P self-refinement sampling | Wan2.2-A14B T2V · Dynamic-bench | VBench motion: 98.01 -> 98.41 | Improved |
| double NFE of default sampler | Wan2.2-A14B T2V · Dynamic-bench | VBench motion: 98.01 -> 98.03 | No clear effect |
| add FlowMo guidance | Wan2.2-A14B T2V · Dynamic-bench | VBench motion: 98.01 -> 97.68 | Hurt |
| add P&P | Cosmos-Predict-2.5-2B · PAI-Bench-G | grasp success rate: 79.2 -> 89.6 | Improved |
| add P&P | Wan2.2-I2V-A14B · PAI-Bench-G | grasp success rate: 77.3 -> 85.7 | Improved |
| replace verifier best-of-4 with P&P | Cosmos-Predict-2.5-2B · PAI-Bench-G | grasp success rate: 84.4 -> 89.6 | Improved |
| add P&P | Wan2.2-A14B T2V · PhyWorldBench | physical commonsense: 29.3 -> 40.0 | Improved |
| add P&P | Wan2.2-A14B T2V · VideoPhy2 | physical commonsense: 54.5 -> 55.6 | Improved |
| add P&P | Wan2.2-A14B T2V · Spatial consistency | SSIM: 0.401 -> 0.485 | Improved |
| add P&P | Wan2.2-A14B I2V · Graph traversal | success rate: 0.1 -> 0.8 | Improved |
| add P&P | Wan2.2-A14B I2V · Maze solving | success rate | No clear effect |
| add uncertainty-aware masking to P&P with large Kf (ablation) | Wan2.2 T2V · qualitative, Kf=5 | visual quality | Improved |
Closest papers by meaning
- SIFT: Self-Imagination Fine-Tuning for Physically Plausible Motion in Video Diffusion Models (ECCV 2026 · similarity 0.89)
- LaMo: Self-Supervised Latent Motion Priors for Physical Realism in Video Generation (NeurIPS 2026 · similarity 0.88)
- When Distillation Breaks Motion Control: Restoring Generative Trajectories for Fast Video Generators (ECCV 2026 · similarity 0.87)
- Physics-Informed Video Generation via Mixture-of-Experts Latent Alignment (NeurIPS 2026 · similarity 0.86)
- Hierarchical Fine-grained Preference Optimization for Physically Plausible Video Generation (NeurIPS 2025 · similarity 0.86)
- Equilibrium Forcing: Adaptive Video Generation Without Noise Conditioning (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.
