Research map / Video / Video segmentation and tracking
A Simple Baseline for Video Restoration With Grouped Spatial-Temporal Shift
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 grouped spatial-temporal shifts instead of standard alignment-heavy fusion (ablation) | Group Shift-Net · GoPro test | PSNR: 34.81 -> 35.49 | Helped |
| Add grouped spatial shift to alternative temporal shift (ablation) | Group Shift-Net · GoPro test | PSNR: 35.14 -> 35.49 | Helped |
| Use alternative temporal shifts rather than a single bidirectional shift across three frames (ablation) | Group Shift-Net · GoPro test | PSNR: 0.33 dB | Helped |
| Replace large-kernel convolution with grouped spatial shift (ablation) | Group Shift-Net fusion layers · GoPro test | PSNR: 35.17 -> 35.49 | Helped |
| Increase fusion receptive field from 13×13 to 23×23 using spatial-temporal shifts (ablation) | Group Shift-Net · GoPro test | PSNR: 35.39 -> 35.49 | Helped |
| Increase fusion receptive field from 23×23 to 33×33 (ablation) | Group Shift-Net · GoPro test | PSNR: 35.49 -> 35.48 | No clear effect |
| Use spatial offsets {0,±5,±9} instead of only {0} (ablation) | Group Shift-Net · GoPro test | PSNR: 35.20 -> 35.49 | Helped |
| Set shift length s to 5 rather than 3, 7, or 9 (ablation) | Group Shift-Net · GoPro deblurring and DAVIS denoising | PSNR: 35.39 -> 35.49 | Helped |
| Replace shift blocks with optical flow (ablation) | Group Shift-Net · GoPro test | PSNR: 35.49 -> 34.14 | Hurt |
| Replace shift blocks with deformable convolution (ablation) | Group Shift-Net · GoPro test | PSNR: 35.49 -> 33.66 | Hurt |
| Replace shift blocks with cross-frame self-attention (ablation) | Group Shift-Net · GoPro test | PSNR: 35.49 -> 34.52 | Hurt |
| Add optical flow to BasicVSR++ (ablation) | BasicVSR++ · REDS4 super-resolution | PSNR: +0.56 dB | Helped |
Closest papers by meaning
- Gated Spatio-Temporal Attention-Guided Video Deblurring (CVPR 2021 · similarity 0.87)
- ARVo: Learning All-Range Volumetric Correspondence for Video Deblurring (CVPR 2021 · similarity 0.87)
- FMA-Net: Flow-Guided Dynamic Filtering and Iterative Feature Refinement with Multi-Attention for Joint Video Super-Resolution and Deblurring (CVPR 2024 · similarity 0.86)
- Learning Truncated Causal History Model for Video Restoration (NeurIPS 2024 · similarity 0.85)
- Improving the Learning Capability of Small-size Image Restoration Network by Deep Fourier Shifting (NeurIPS 2024 · similarity 0.85)
- Blind Bitstream-corrupted Video Recovery via Metadata-guided Diffusion Model (CVPR 2025 · similarity 0.84)
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.