Research map / Video / Video generation
SpeedVFI: One-step Diffusion for Efficient Video Frame Interpolation
Ganggui Ding, Xiaogang Xu, Hao Chen, Chunhua Shen
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 flow matching loss to DMD distillation (ablation) | SpeedVFI · VFI ablation | PSNR: 12.60 -> 20.05 | Improved |
| add pixel loss (LPIPS + MSE) on top of DMD + flow loss (ablation) | SpeedVFI · VFI ablation | FVD: 417.85 -> 370.77 | Improved |
| add temporal RoPE alignment to sequential token concatenation (ablation) | SpeedVFI · VFI ablation | FVD: 518.06 -> 417.85 | Improved |
| use sequential instead of interleaved concatenation (ablation) | SpeedVFI · VFI ablation | FVD: 641.64 -> 518.06 | Improved |
| raise TTUR interval from 1 to 5 (ablation) | SpeedVFI · VFI ablation | FID: 52.88 -> 44.19 | Improved |
| use noise-centric partial attention instead of full attention (1K steps) (ablation) | SpeedVFI · 9->33 frames ablation | DiT TFLOPs: 75.44 -> 53.80 TFLOPs; PSNR 20.11 -> 20.06 | Improved |
| use partial attention instead of full attention (ablation) | SpeedVFI · 17->65 frames latency | time (s): 10.72 -> 9.98 | Improved |
| train partial attention longer (1K to 2K steps) (ablation) | SpeedVFI partial attention · 9->33 frames ablation | LPIPS: 0.1480 -> 0.1212 | Improved |
| one-step multi-frame diffusion instead of pairwise multi-step diffusion | SpeedVFI vs Framer · DAVIS | inference time (s): 150.69 -> 9.98 | Improved |
| one-step SpeedVFI compared with GIMM-VFI on LPIPS | SpeedVFI · DAVIS | LPIPS: 0.112 -> 0.123 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Framer — Framer: Interactive Frame Interpolation (ICLR 2025) · better on 5 of 5 shared results
- BiM-VFI — BiM-VFI: Bidirectional Motion Field-Guided Frame Interpolation for Video with Non-uniform Motions (CVPR 2025) · better on 3 of 3 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
- EDEN: Enhanced Diffusion for High-quality Large-motion Video Frame Interpolation (CVPR 2025 · similarity 0.92)
- LDMVFI: Video Frame Interpolation with Latent Diffusion Models (AAAI 2024 · similarity 0.91)
- Disentangled Motion Modeling for Video Frame Interpolation (AAAI 2025 · similarity 0.91)
- Repurposing Pre-trained Video Diffusion Models for Event-based Video Interpolation (CVPR 2025 · similarity 0.90)
- ViBiDSampler: Enhancing Video Interpolation Using Bidirectional Diffusion Sampler (ICLR 2025 · similarity 0.90)
- Generative Inbetweening: Adapting Image-to-Video Models for Keyframe Interpolation (ICLR 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.
