Research map / Image generation / Super-resolution
N-Gram in Swin Transformers for Efficient Lightweight 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 |
|---|---|---|---|
| Add N-Gram context (ablation) | NGswin · Urban100 ×4 | PSNR: 26.25 -> 26.45 | Helped |
| Add N-Gram context (ablation) | HNCT · Urban100 ×2 | PSNR: 32.22 -> 32.35 | Helped |
| Use two N-Gram directions with sliding-WSA rather than one or four directions or CNN interaction (ablation) | NGswin · Urban100 ×2 | PSNR: 32.54; 32.52; 32.45; 32.54; 32.52 | Mixed |
| Use default stages instead of extra encoder and decoder stages (ablation) | NGswin without SCDP · Urban100 ×2 | PSNR: 32.28 -> 32.48 | Helped |
| Add the SCDP multi-scale bottleneck (ablation) | NGswin with default stages · Urban100 ×2 | PSNR: 32.48 -> 32.53 | Helped |
| Use SwinV2 scaled-cosine attention instead of SwinV1 (ablation) | NGswin · Urban100 ×4 | PSNR: 26.26 -> 26.45 | Helped |
| Use sequentially reflected window padding instead of zero padding (ablation) | NGswin · Urban100 ×4 | PSNR: 25.72 -> 26.45 | Helped |
| Warm-start SwinIR-NG rather than train from scratch (ablation) | SwinIR-NG · Urban100 ×4 | PSNR: 26.51 -> 26.61 | Helped |
| Use warm-start rather than scratch training (ablation) | SwinIR-NG · Set5 ×3 | PSNR: 34.65 -> 34.64 | Mixed |
| Add N-Gram context to SwinIR-light (ablation) | SwinIR-light · Manga109 ×4 | PSNR: 30.92 -> 31.09 | Helped |
| Add N-Gram context to HNCT (ablation) | HNCT · Set14 ×2 | PSNR: 33.65 -> 33.64 | No clear effect |
Closest papers by meaning
- Transcending the Limit of Local Window: Advanced Super-Resolution Transformer with Adaptive Token Dictionary (CVPR 2024 · similarity 0.86)
- Recursive Generalization Transformer for Image Super-Resolution (ICLR 2024 · similarity 0.85)
- Efficient Attention-Sharing Information Distillation Transformer for Lightweight Single Image Super-Resolution (AAAI 2025 · similarity 0.84)
- GSDD: Generative Space Dataset Distillation for Image Super-resolution (AAAI 2024 · similarity 0.83)
- AdaFormer: Efficient Transformer with Adaptive Token Sparsification for Image Super-resolution (AAAI 2024 · similarity 0.83)
- L-SWAG: Layer-Sample Wise Activation with Gradients Information for Zero-Shot NAS on Vision Transformers (CVPR 2025 · similarity 0.83)
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.