Research map / Video / Video segmentation and tracking
Unsupervised Video Interpolation Using Cycle Consistency
Fitsum A. Reda, Deqing Sun, Aysegul Dundar, Mohammad Shoeybi, Guilin Liu, Kevin J. Shih, Andrew Tao, Jan Kautz, Bryan Catanzaro
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 pseudo supervised loss to cycle consistency (ablation) | CC fine-tuning · Adobe->Slowflow | PSNR: 32.35 -> 33.05 | Improved |
| add pseudo supervised loss to cycle consistency (ablation) | CC fine-tuning · Adobe->Sintel | PSNR: 30.08 -> 32.53 | Improved |
| add pseudo supervised loss to cycle consistency (ablation) | CC fine-tuning · Adobe->Sintel | IE: 7.67 -> 5.21 | Improved |
| add cycle consistency fine-tuning to Adobe+YouTube baseline (ablation) | Pre-trained Super SloMo · Adobe+YouTube->Slowflow | PSNR: 33.13 -> 33.20 | Improved |
| add unsupervised CC+PS fine-tuning on target videos | Pre-trained DVF · UCF101->Slowflow | PSNR: 24.64 -> 28.38 | Improved |
| add unsupervised fine-tuning at full resolution | Quarter-resolution supervised model · Slowflow | PSNR: 30.79 -> 31.42 | Improved |
| add unsupervised fine-tuning at full resolution | Quarter-resolution supervised model · Sintel | PSNR: 29.14 -> 29.71 | Improved |
| train unsupervised with CC on Adobe-30fps instead of supervised Adobe-240fps | Super SloMo · UCF101 | PSNR: 34.63 -> 34.47 | No clear effect |
| train unsupervised with CC on BattleField-30fps | Super SloMo · UCF101 | IE: 5.48 -> 5.38 | Improved |
| use CC instead of copying inputs as predictions | Trivial Copy · UCF101 | PSNR: 31.27 -> 34.47 | Improved |
| replace cycle consistency with long-step loss (ablation) | CC · Adobe+YouTube->Slowflow | PSNR: 32.84 -> 29.03 | Hurt |
| add long-step loss to cycle consistency (ablation) | CC · Adobe+YouTube->Slowflow | PSNR: 32.84 -> 32.24 | No clear effect |
Closest papers by meaning
- Deep Video Frame Interpolation Using Cyclic Frame Generation (AAAI 2019 · similarity 0.92)
- Super SloMo: High Quality Estimation of Multiple Intermediate Frames for Video Interpolation (CVPR 2018 · similarity 0.88)
- Optimizing Video Prediction via Video Frame Interpolation (CVPR 2022 · similarity 0.87)
- Deep Bayesian Video Frame Interpolation (ECCV 2022 · similarity 0.87)
- Video Frame Interpolation without Temporal Priors (NeurIPS 2020 · similarity 0.87)
- UPR-Net: A Unified Pyramid Recurrent Network for Video Frame Interpolation (IJCV 2024 · similarity 0.87)
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.
