Research map / Video / Video segmentation and tracking
Learning Blind Video Temporal Consistency
Wei-Sheng Lai, Jia-Bin Huang, Oliver Wang, Eli Shechtman, Ersin Yumer, Ming-Hsuan Yang
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 learned recurrent network instead of optimization-based Bonneel et al. | blind temporal consistency · DAVIS test (avg) | perceptual distance: 0.088 -> 0.017 | Improved |
| use learned recurrent network instead of optimization-based Bonneel et al. | blind temporal consistency · Videvo test (avg) | perceptual distance: 0.073 -> 0.012 | Improved |
| use learned recurrent network instead of Bonneel et al. | blind temporal consistency · DAVIS test (avg) | temporal warping error: 0.032 -> 0.030 | No clear effect |
| use learned recurrent network instead of Bonneel et al. | blind temporal consistency · Videvo test (avg) | temporal warping error: 0.015 -> 0.012 | Improved |
| apply the proposed network to per-frame processed video | per-frame processed video · DAVIS test (avg) | temporal warping error: 0.047 -> 0.030 | Improved |
| remove optical flow computation at test time | Bonneel et al. vs ours · 1280x720 video | speed (FPS): 0.25 -> 418 | Improved |
| raise ratio r=lambda_t/lambda_p above 10 (ablation) | loss weighting · DAVIS test (udnie style) | perceptual distance | Hurt |
| lower ratio r=lambda_t/lambda_p below 10 (ablation) | loss weighting · DAVIS test (udnie style) | temporal warping error | Hurt |
| set r = 10 with lambda_t >= 100 (ablation) | loss weighting · DAVIS test (udnie style) | temporal warping error and perceptual distance | Improved |
| held-out tasks in user study: ours vs Bonneel et al. | blind temporal consistency · User study (pairwise) | user preference: 62% | Improved |
| replace Bonneel et al. with ours on a trained task | blind temporal consistency · DAVIS test: WCT/sketch | temporal warping error: 0.028 -> 0.023 | Improved |
| replace Bonneel et al. with ours on a stylization task | blind temporal consistency · DAVIS test: WCT/antimono | temporal warping error: 0.031 -> 0.035 | Hurt |
Closest papers by meaning
- Blind Video Temporal Consistency via Deep Video Prior (NeurIPS 2020 · similarity 0.94)
- Deep Video Prior for Video Consistency and Propagation (TPAMI 2022 · similarity 0.92)
- Single-Frame Regularization for Temporally Stable CNNs (CVPR 2019 · similarity 0.89)
- Recurrent Temporal Aggregation Framework for Deep Video Inpainting (TPAMI 2019 · similarity 0.88)
- Instant Video Models: Universal Adapters for Stabilizing Image-Based Networks (NeurIPS 2025 · similarity 0.87)
- Learning Temporally Consistent Video Depth from Video Diffusion Priors (CVPR 2025 · 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.
