Research map / Video / Video segmentation and tracking
Learning Spatio-Temporal Downsampling for Effective Video Upscaling
Xiaoyu Xiang, Yapeng Tian, Vijay Rengarajan, Lucas D. Young, Bo Zhu, Rakesh Ranjan
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 |
|---|---|---|---|
| learned STAA downsampler instead of nearest in time (ablation) | upscaling network · Vimeo-90k (2x time, 1x space) | PSNR: 38.22 -> 46.75 | Improved |
| learned STAA_quant instead of Gaussian pre-filter (ablation) | upscaling network · Vimeo-90k (2x time, 2x space) | PSNR: 37.44 -> 40.42 | Improved |
| Gaussian pre-filter instead of nearest-bicubic (ablation) | upscaling network · Vimeo-90k (2x time, 2x space) | PSNR: 28.88 -> 37.44 | Improved |
| softmax constraint on STAA filter weights (ablation) | STAA downsampler · Vimeo-90k (2x time, 2x space) | PSNR: 39.44 -> 40.40 | Improved |
| differentiable quantization layer (ablation) | STAA downsampler · Vimeo-90k (2x time, 2x space) | PSNR: 40.40 -> 40.42 | No clear effect |
| content-adaptive filter weights (ablation) | STAA downsampler · Vimeo-90k (2x time, 2x space) | PSNR: 40.40 -> 38.13 | Hurt |
| 3D RDB instead of naive 3D conv in upsampler (ablation) | upsampler · Upsampler module ablation | PSNR: 28.67 -> 31.35 | Improved |
| deformable temporal modeling (DTM) instead of naive 3D conv (ablation) | upsampler · Upsampler module ablation | PSNR: 28.67 -> 31.55 | Improved |
| add DTM to 3D RDB upsampler (ablation) | upsampler · Upsampler module ablation | PSNR: 31.35 -> 32.00 | Improved |
| STAA downsampled input instead of nearest-bicubic (ablation) | ZSM · Vimeo-90k (4x space, 2x time) | PSNR: 33.48 -> 35.62 | Improved |
| our upsampler instead of FLAVR upsampler (ablation) | STAA framework · Vimeo-90k (2x time) | PSNR: 44.72 -> 46.59 | Improved |
| add adaptive filter vs unconstrained STAA filter (ablation) | STAA downsampler · Vimeo-90k (2x time, 1x space) | PSNR: 46.50 -> 45.32 | Hurt |
Closest papers by meaning
- Continuous Space-Time Video Resampling with Invertible Motion Steganography (CVPR 2025 · similarity 0.88)
- Time Without Time: Pseudo-Temporal Representation for Space-Time Super-Resolution (CVPR 2026 · similarity 0.88)
- Video Rescaling Networks With Joint Optimization Strategies for Downscaling and Upscaling (CVPR 2021 · similarity 0.87)
- Real-Time Video Super-Resolution With Spatio-Temporal Networks and Motion Compensation (CVPR 2017 · similarity 0.87)
- Enhancing Video Super-Resolution via Implicit Resampling-based Alignment (CVPR 2024 · similarity 0.85)
- A Video is Worth 256 Bases: Spatial-Temporal Expectation-Maximization Inversion for Zero-Shot Video Editing (CVPR 2024 · similarity 0.85)
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.
