Research map / Video / Video segmentation and tracking
Video Frame Interpolation without Temporal Priors
Youjian Zhang, Chaoyue Wang, Dacheng Tao
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 |
|---|---|---|---|
| Use the proposed full method instead of the strongest reported baseline | Video frame interpolation · GoPro-5-5, GoPro-7-3, GoPro-9-1 and Adobe240-5-5, Adobe240-7-3, Adobe240-9-1 | interpolation PSNR: GoPro: 32.47 vs 29.13; 31.95 vs 30.92; 30.95 vs 30.82. Adobe240: 31.79 vs 29.5; 31.52 vs 31.43; 30.66 vs 30.48 | Improved |
| Use uncertain-time-interval motion interpolation rather than QVI with λ=1 | Sharp-frame interpolation · GoPro, λ=7/3 and λ=9/1 | PSNR: 32.35 -> 32.65; 29.17 -> 29.75 | Improved |
| Add optical flow refinement to UTI (ablation) | Sharp-frame interpolation · GoPro, λ=7/3 and λ=9/1 | PSNR: 32.65 -> 33.56; 29.75 -> 30.57 | Improved |
| Add second-order residual learning (ablation) | Key-state restoration · GoPro-5-5, GoPro-7-3, GoPro-9-1 | PSNR: 32.39 -> 34.00; 31.15 -> 32.63; 30.38 -> 31.72 | Improved |
| Cascade another stage-I architecture without B1 and B2 as inputs (ablation) | Key-state restoration · GoPro-5-5, GoPro-7-3, GoPro-9-1 | PSNR: 34.00 vs 32.49; 32.63 vs 31.41; 31.72 vs 30.76 | Hurt |
| Use only 2 input frames instead of the proposed input structure (ablation) | Key-state restoration · GoPro-5-5, GoPro-7-3, GoPro-9-1 | PSNR: 34.00 vs 33.07; 32.63 vs 31.83; 31.72 vs 30.74 | Hurt |
| Remove optical flow refinement (ablation) | Interpolation · GoPro-5-5, GoPro-7-3, GoPro-9-1 | PSNR: 31.82 -> 32.47; 31.34 -> 31.95; 30.3 -> 30.95 | Hurt |
| Apply the equal-interval QVI trajectory when time intervals are uneven | Sharp-frame interpolation · GoPro, λ=7/3 and λ=9/1 | PSNR: 32.65 vs 32.35; 29.75 vs 29.17 | Hurt |
| Iteratively synthesize the middle frame for TNTT* under exposure setting 5-5 | TNTT* · GoPro-5-5 interpolation | PSNR: 32.49 -> 28.39 | Hurt |
| Use a cascade of deblurring followed by interpolation | Cascade models · GoPro and Adobe240 blurry video interpolation | PSNR: about 3 dB loss | Hurt |
Closest papers by meaning
- Realtime Video Frame Interpolation using One-Step Diffusion Sampling (ICLR 2026 · similarity 0.91)
- Quadratic Video Interpolation (NeurIPS 2019 · similarity 0.90)
- Super SloMo: High Quality Estimation of Multiple Intermediate Frames for Video Interpolation (CVPR 2018 · similarity 0.90)
- Disentangled Motion Modeling for Video Frame Interpolation (AAAI 2025 · similarity 0.89)
- Framer: Interactive Frame Interpolation (ICLR 2025 · similarity 0.88)
- Generative Inbetweening: Adapting Image-to-Video Models for Keyframe Interpolation (ICLR 2025 · similarity 0.88)
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.
