Research map / Architectures and efficiency / RNNs and normalization
MemSR: Training Memory-efficient Lightweight Model for Image Super-Resolution
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 |
|---|---|---|---|
| apply winning initialization derived from teacher to plain student (ablation) | Plain-M · Urban100 ×4 | PSNR: 25.31 -> 25.97 dB | Helped |
| add initialization-aware feature distillation (FDInit) on top of winning initialization (ablation) | Plain-M · Urban100 ×4 | PSNR: 25.97 -> 25.99 dB | Helped |
| add standard feature distillation without winning initialization (ablation) | Plain-M · Urban100 ×4 | PSNR: 25.30 -> 25.67 dB | Helped |
| mean-shift on student pseudo features (MST) (ablation) | Plain-M student with winning init · Set5 ×4 | PSNR: 32.030 -> 32.090 dB | Helped |
| replace SRKD/FAKD/PISR distillation with MemSR training on plain student | Plain-M and Plain-S students · Urban100 ×4 | PSNR: 25.99 vs 25.30 baseline, other distillation methods at or below baseline | Helped |
| RepVGG re-parameterized training for plain SR student | Plain-M · Urban100 ×4 | PSNR: 24.55 vs 25.30 baseline | Hurt |
| remove batch normalization from RepVGG blocks (ablation) | RepVGG student · Urban100 ×4 | PSNR: 24.55 -> 25.35 dB | Helped |
| PISR distillation on plain student | Plain-M · Urban100 ×4 | PSNR: training failed (7.169 dB) at depth 35 | Hurt |
| dynamic per-layer student width (Plain-X) with smaller error limit theta=0.05 vs equal width Plain-M | Plain-X · Urban100 ×4 | PSNR: 26.17 vs 25.99 (Plain-M) | Helped |
| MemSR plain student (Plain-M) vs FSRCNN memory | Plain-M · Urban100 ×4 | maximum memory footprint (MB): 161MB vs 111MB FSRCNN | Mixed |
Closest papers by meaning
- Image Super-Resolution via Dual-State Recurrent Networks (CVPR 2018 · similarity 0.87)
- Real-Time Single Image and Video Super-Resolution Using an Efficient Sub-Pixel Convolutional Neural Network (CVPR 2016 · similarity 0.87)
- “Zero-Shot” Super-Resolution Using Deep Internal Learning (CVPR 2018 · similarity 0.87)
- Deeply-Recursive Convolutional Network for Image Super-Resolution (CVPR 2016 · similarity 0.87)
- SAVSR: Arbitrary-Scale Video Super-Resolution via a Learned Scale-Adaptive Network (AAAI 2024 · similarity 0.86)
- Lightweight Image Super-Resolution via Flexible Meta Pruning (ICML 2024 · 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.