Research map / Video / Video generation
Realtime Video Frame Interpolation using One-Step Diffusion Sampling
Yongrui Ma, Shijie Zhao, Mingde Yao, Junlin Li, Li zhang, Xiaohong Liu, Qi Dou, Jinwei Gu, Tianfan Xue
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 iterative flow fusion for continuous pixel trajectories (vs Direct Warping) (ablation) | RDVFI-D · ablation set (Table 3b) | FVD: 327.11 -> 209.38 | Improved |
| add iterative flow fusion (vs Direct Warping) (ablation) | RDVFI-D · ablation set (Table 3b) | FID: 42.33 -> 23.71 | Improved |
| add pixel-space perceptual loss (vs L+P-L2 loss) (ablation) | RDVFI-D · ablation set (Table 3b) | FVD: 267.42 -> 209.38 | Improved |
| add pixel-wise reconstruction losses (L loss vs L+P-L2 loss) (ablation) | RDVFI-D · ablation set (Table 3b) | FVD: 281.37 -> 267.42 | Improved |
| use motion-guided decoding with fixed low-resolution diffusion (vs one-step Wan) (ablation) | RDVFI-D · 1024x576 | runtime (ms): 267.7 -> 58.01 | Improved |
| raise output resolution from 576x320 to 1280x720 (ablation) | RDVFI-D · runtime scaling | runtime (ms): 55.3 -> 63.49 | No clear effect |
| raise output resolution from 576x320 to 1280x720 (ablation) | one-step Wan · runtime scaling | runtime (ms): 103.3 -> 314.3 | Hurt |
| use DiT backbone Wan (RDVFI-D) instead of SVD U-Net (RDVFI-U) (ablation) | RDVFI · FCVG dataset | FVD: 217.72 -> 197.86 | Improved |
| use RDVFI-D instead of Wan with same backbone | Wan 1.3B InP · FCVG dataset | runtime (sec): 2.579 -> 0.057 | Improved |
| use RDVFI-D instead of RIFE | RIFE · FCVG dataset | FVD: 366.14 -> 197.86 | Improved |
| use RDVFI-D instead of Wan | Wan · human evaluation, overall | share choosing as best (%): 14.2% -> 75.7% | Improved |
| increase frame skip to 7 so input frames barely overlap | RDVFI · failure case | interpolation quality | Hurt |
Closest papers by meaning
- EDEN: Enhanced Diffusion for High-quality Large-motion Video Frame Interpolation (CVPR 2025 · similarity 0.93)
- Generative Inbetweening: Adapting Image-to-Video Models for Keyframe Interpolation (ICLR 2025 · similarity 0.92)
- ViBiDSampler: Enhancing Video Interpolation Using Bidirectional Diffusion Sampler (ICLR 2025 · similarity 0.92)
- Hierarchical Flow Diffusion for Efficient Frame Interpolation (CVPR 2025 · similarity 0.91)
- Video Interpolation with Diffusion Models (CVPR 2024 · similarity 0.91)
- Disentangled Motion Modeling for Video Frame Interpolation (AAAI 2025 · similarity 0.90)
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.
