Research map / Video / Video segmentation and tracking
TurboVSR: Fantastic Video Upscalers and Where to Find Them
Zhongdao Wang, Guodongfang Zhao, Jingjing Ren, Bailan Feng, Shifeng Zhang, Wenbo Li
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 factorized first-frame conditioning instead of plain video condition (ablation) | TurboVSR · SPMCS | DOVER: 34.23 -> 79.15 | Improved |
| add factorized first-frame conditioning instead of plain video condition (ablation) | TurboVSR · SPMCS | NIQE: 6.162 -> 4.113 | Improved |
| add factorized first-frame conditioning instead of plain video condition (ablation) | TurboVSR · SPMCS | PSNR: 24.81 -> 22.15 | Hurt |
| add shortcut bootstrap loss to flow matching training (ablation) | TurboVSR · SPMCS (4 steps) | DOVER: 74.09 -> 79.15 | Improved |
| add shortcut bootstrap loss to flow matching training (ablation) | TurboVSR · SPMCS (1 step) | DOVER: 68.31 -> 77.08 | Improved |
| add shortcut bootstrap loss to flow matching training (ablation) | TurboVSR · SPMCS (4 steps) | NIQE: 4.84 -> 4.11 | Improved |
| add non-uniform timestep t sampling (with uniform d) (ablation) | TurboVSR shortcut · Shortcut sampling ablation (4 steps) | DOVER: 71.44 -> 78.14 | Improved |
| add non-uniform step size d sampling (with non-uniform t) (ablation) | TurboVSR shortcut · Shortcut sampling ablation (4 steps) | E_warp: 4.42 -> 3.57 | Improved |
| add non-uniform step size d sampling (with non-uniform t) (ablation) | TurboVSR shortcut · Shortcut sampling ablation (4 steps) | LPIPS: 0.218 -> 0.184 | Improved |
| reduce sampling steps from 10 to 4 (ablation) | TurboVSR · SPMCS | DOVER: 79.29 -> 79.15 | No clear effect |
| reduce sampling steps from 10 to 4 (ablation) | TurboVSR · VSR latency (1080p) | latency ms/frame: 337 -> 140 | Improved |
| reduce sampling steps from 10 to 4 (ablation) | TurboVSR · UDM10 | MUSIQ: 59.21 -> 56.07 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- Upscale-A-Video — Upscale-A-Video: Temporal-Consistent Diffusion Model for Real-World Video Super-Resolution (CVPR 2024) · better on 2 of 4 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
- FlashVSR: Towards Real-time Diffusion-Based Streaming Video Super Resolution (CVPR 2026 · similarity 0.91)
- DiffVSR: Revealing an Effective Recipe for Taming Robust Video Super-Resolution Against Complex Degradations (ICCV 2025 · similarity 0.91)
- DOVE: Efficient One-Step Diffusion Model for Real-World Video Super-Resolution (NeurIPS 2025 · similarity 0.90)
- PatchVSR: Breaking Video Diffusion Resolution Limits with Patch-wise Video Super-Resolution (CVPR 2025 · similarity 0.90)
- SkipSR: Faster Super-Resolution with Token Skipping (NeurIPS 2026 · similarity 0.89)
- Upscale-A-Video: Temporal-Consistent Diffusion Model for Real-World Video Super-Resolution (CVPR 2024 · similarity 0.89)
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.
