Research map / Video / Video segmentation and tracking
Deep Video Prior for Video Consistency and Propagation
Chenyang Lei, Yazhou Xing, Hao Ouyang, Qifeng Chen
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 iteratively reweighted training (IRT) (ablation) | DVP colorization · Colorization on DAVIS | F_data (PSNR): 29.61 -> 28.20 | Mixed |
| add better initialization from pretrained early-stopped network (ablation) | DVP · CycleGAN ukiyoe | E_warp / F_data: E_warp 0.218 -> 0.217, F_data 26.25 -> 26.33 (x3 speed-up) | No clear effect |
| add coarse-to-fine training (ablation) | DVP · CycleGAN ukiyoe | E_warp / F_data: E_warp 0.218 -> 0.217, F_data 26.25 -> 26.18 (x1.6 speed-up) | No clear effect |
| combine better initialization and coarse-to-fine training (ablation) | DVP · CycleGAN ukiyoe | E_warp / F_data: E_warp 0.218 -> 0.215, F_data 26.25 -> 26.32 (x4.8 speed-up) | Improved |
| use automatic instead of manual stopping epoch (ablation) | DVP · CycleGAN ukiyoe | F_data / E_warp: F_data 25.64 -> 25.52, E_warp 0.194 -> 0.193 | No clear effect |
| replace U-net with ResUnet (ablation) | DVP · Network architecture comparison | E_warp: 0.1747 -> 0.1746 | No clear effect |
| replace U-net with FCN (ablation) | DVP · Network architecture comparison | E_warp: 0.1747 -> 0.1751 | No clear effect |
| add progressive propagation with pseudo labels (PPPL) (ablation) | DVP segmentation propagation · DAVIS2016 | J Mean / F Mean: J 68.3 -> 70.5, F 67.8 -> 70.7 | Improved |
| add crop augmentation (ablation) | DVP segmentation propagation · DAVIS2016 | J Mean / F Mean: J 60.9 -> 63.1, F 60.5 -> 61.4 | Improved |
| add flip augmentation on top of crop (ablation) | DVP segmentation propagation · DAVIS2016 | J Mean / F Mean: J 63.1 -> 65.4, F 61.4 -> 63.8 | Improved |
| add copy-paste augmentation on top of crop and flip (ablation) | DVP segmentation propagation · DAVIS2016 | J Mean / F Mean: J 65.4 -> 70.5, F 63.8 -> 70.7 | Improved |
| increase number of reference frames from 1 to 5 (ablation) | DVP color propagation · DAVIS color propagation | PSNR: 27.29 -> 29.87 (DeepRemaster 24.92 -> 25.67) | Improved |
Closest papers by meaning
- Blind Video Temporal Consistency via Deep Video Prior (NeurIPS 2020 · similarity 0.95)
- Learning Temporally Consistent Video Depth from Video Diffusion Priors (CVPR 2025 · similarity 0.88)
- Buffer Anytime: Zero-Shot Video Depth and Normal from Image Priors (CVPR 2025 · similarity 0.86)
- Instant Video Models: Universal Adapters for Stabilizing Image-Based Networks (NeurIPS 2025 · similarity 0.86)
- Motion Consistency Model: Accelerating Video Diffusion with Disentangled Motion-Appearance Distillation (NeurIPS 2024 · similarity 0.86)
- OSV: One Step is Enough for High-Quality Image to Video Generation (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.
