Research map / Image generation / Super-resolution
CUST : Clustered Unit-level Similarity Transformer for Lightweight Image Super-Resolution
Jeongsoo Kim
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 MEDA to CANA-only network (ablation) | CUST-Small · Urban100 x4 | PSNR: 26.31 -> 26.60 | Improved |
| add CANA to MEDA-only network (ablation) | CUST-Small · Urban100 x4 | PSNR: 26.21 -> 26.60 | Improved |
| add MEDA to CANA-only network (ablation) | CUST-Small · Manga109 x4 | PSNR: 30.80 -> 31.14 | Improved |
| combine both modules vs parameter-matched CANA-only (ablation) | CUST-Small · Urban100 x4 | PSNR: 26.51 -> 26.60 | Improved |
| combine both modules vs parameter-matched MEDA-only (ablation) | CUST-Small · Urban100 x4 | PSNR: 26.38 -> 26.60 | Improved |
| replace Swin attention with CANA (ablation) | CUST-Small · Urban100 x4 | PSNR: 26.53 -> 26.60 | Improved |
| use key-value expansion in CANA (ablation) | CUST-Small · Urban100 x4 | PSNR: 26.58 -> 26.60 | Improved |
| increase CANA group size from 5 to 10 (ablation) | CUST-Small · Urban100 x4 | PSNR: 26.59 -> 26.60 | Improved |
| add error refiner and spatial gate in MEDA (ablation) | CUST-Small · Urban100 x4 | SSIM: 0.7982 -> 0.7995 | Improved |
| add MFA at window sizes +2 compared with +4 without MFA (ablation) | CUST-Small · Urban100 x4 | PSNR: 26.62 -> 26.64 | Improved |
| add MFA with window size -2 vs +2 without MFA (ablation) | CUST-Small · Urban100 x4 | GPU memory (M): 307.7 -> 248.5 | Improved |
| add MFA at default window sizes (ablation) | CUST-Small · Urban100 x4 | SSIM: 0.7983 -> 0.7995 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- CATANet — CATANet: Efficient Content-Aware Token Aggregation for Lightweight Image Super-Resolution (CVPR 2025) · better on 5 of 6 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
- VITRIX-UniViTAR: Unified Vision Transformer with Native Resolution (NeurIPS 2025 · similarity 0.87)
- LF-ViT: Reducing Spatial Redundancy in Vision Transformer for Efficient Image Recognition (AAAI 2024 · similarity 0.85)
- CFAT: Unleashing Triangular Windows for Image Super-resolution (CVPR 2024 · similarity 0.85)
- Recursive Generalization Transformer for Image Super-Resolution (ICLR 2024 · similarity 0.85)
- Scaling Laws in Patchification: An Image Is Worth 50,176 Tokens And More (ICML 2025 · similarity 0.85)
- AdaFormer: Efficient Transformer with Adaptive Token Sparsification for Image Super-resolution (AAAI 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.
