Research map / Image generation / Super-resolution
ESSAformer: Efficient Transformer for Hyperspectral Image Super-resolution
Mingjin Zhang, Chi Zhang, Qiming Zhang, Jie Guo, Xinbo Gao, Jing Zhang
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 |
|---|---|---|---|
| replace multi-head self-attention with ESSA (ablation) | ESSAformer · Chikusei 4x | PSNR: 40.4310 -> 40.7648 | Improved |
| replace windowed Swin attention with ESSA (ablation) | ESSAformer · Chikusei 4x | PSNR: 40.1122 -> 40.7648 | Improved |
| increase Taylor order from 0 to 1 in the kernel mapping (ablation) | ESSA · Chikusei 4x | PSNR: 40.5474 -> 40.7648 | Improved |
| increase Taylor order beyond 1 (ablation) | ESSA · Chikusei 4x | PSNR: 40.7648 -> 40.8012 (order 3) | No clear effect |
| increase stages from 3 to 5 (ablation) | ESSAformer · Chikusei 4x | PSNR: 40.42 -> 40.76 | Improved |
| increase stages from 5 to 7 or 9 (ablation) | ESSAformer · Chikusei 4x | PSNR: 40.7648 -> 40.7314 (7), 40.7191 (9) | No clear effect |
| use SCC spectral correlation attention instead of MHSA (ablation) | ESSAformer · Cave 4x (128x128 crops) | PSNR: 44.5341 -> 44.9210 | Improved |
| kernelize SCC attention into ESSA for linear complexity (ablation) | SCC attention · Chikusei 4x (128x128 crops) | PSNR: 41.3273 -> 41.4177 | Improved |
| use ESSA instead of other linear attention (Performer, NPRF, linear) (ablation) | ESSAformer · Chikusei 4x | PSNR: 40.5833 -> 40.7648 (vs Performer) | Improved |
| use ESSA instead of MHSA at larger input resolution (ablation) | ESSAformer · FLOPs at 40x40 input | FLOPs: 1326.80 G -> 304.97 G | Improved |
| use ESSAformer on large datasets | SwinIR · NTIRE2020 | PSNR: 34.9186 -> 34.8674 | No clear effect |
| increase scale factor from 2x to 8x | ESSAformer · Pavia | PSNR: 38.7896 -> 27.1114 | Hurt |
Later papers that compared with this method
2 later accepted papers report ESSAFormer as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- SSDCN: Spatial-Spectral Dual-Clustering-based Network for Hyperspectral Image Super-resolution (ICML 2026 · better on 1 of 1 shared result)
- EigenSR: Eigenimage-Bridged Pre-Trained RGB Learners for Single Hyperspectral Image Super-Resolution (AAAI 2025 · better on 6 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
- Learning Exhaustive Correlation for Spectral Super-Resolution: Where Spatial-Spectral Attention Meets Linear Dependence (ECCV 2024 · similarity 0.91)
- EigenSR: Eigenimage-Bridged Pre-Trained RGB Learners for Single Hyperspectral Image Super-Resolution (AAAI 2025 · similarity 0.90)
- Unsupervised Sparse Dirichlet-Net for Hyperspectral Image Super-Resolution (CVPR 2018 · similarity 0.90)
- VolFormer: Explore More Comprehensive Cube Interaction for Hyperspectral Image Restoration and Beyond (CVPR 2025 · similarity 0.88)
- SPECAT: SPatial-spEctral Cumulative-Attention Transformer for High-Resolution Hyperspectral Image Reconstruction (CVPR 2024 · similarity 0.88)
- Learning Multi-Modal Cross-Scale Deformable Transformer Network for Unregistered Hyperspectral Image Super-resolution (AAAI 2024 · similarity 0.88)
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.
