Research map / Architectures and efficiency / Attention and transformers
S2-Transformer for Mask-Aware Hyperspectral Image Reconstruction
Jiamian Wang, Kunpeng Li, Yulun Zhang, Xin Yuan, Zhiqiang Tao
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 |
|---|---|---|---|
| use parallel instead of sequential spatial-spectral attention (ablation) | S2-Transformer · KAIST simulation (10 scenes, avg) | PSNR (dB): 36.06 -> 36.48 | Improved |
| use parallel spatial-spectral attention instead of spatial-only blocks (ablation) | S2-Transformer · KAIST simulation (10 scenes, avg) | PSNR (dB): 36.01 -> 36.48 | Improved |
| use parallel spatial-spectral attention instead of spectral-only blocks (ablation) | S2-Transformer · KAIST simulation (10 scenes, avg) | PSNR (dB): 35.35 -> 36.48 | Improved |
| sequential spatial-spectral instead of spatial-only attention (ablation) | attention block · KAIST simulation (10 scenes, avg) | PSNR (dB): 36.01 -> 36.06 | No clear effect |
| add mask-encoding pre-training then mask-aware loss versus L1 only (ablation) | S2-Transformer · KAIST simulation (10 scenes, avg) | PSNR (dB): 36.31 -> 36.48 | Improved |
| add mask-encoding term alone (ablation) | S2-Transformer · KAIST simulation (10 scenes, avg) | PSNR (dB): 36.31 -> 36.32 | No clear effect |
| apply mask-aware loss from scratch without ME pre-training (ablation) | S2-Transformer · KAIST simulation (10 scenes, avg) | PSNR (dB): 36.31 -> 34.88 | Hurt |
| add mask-aware learning to SRN | SRN · KAIST simulation (10 scenes, avg) | PSNR (dB): 35.07 -> 35.50 | Improved |
| add mask-aware learning to MST | MST · KAIST simulation (10 scenes, avg) | PSNR (dB): 35.18 -> 35.39 | Improved |
| later ME layer choice k_ME for SRN (ablation) | SRN with mask-aware · KAIST simulation (10 scenes, avg) | PSNR (dB): 35.20 -> 35.50 | Improved |
| add L_ME and L_MA training to S2-Transformer on real data (ablation) | S2-Transformer · Real hyperspectral data | NIQE: 6.0950 -> 5.8833 | Improved |
| S2-Transformer versus MST under Poisson noise lambda=20 | S2-Transformer · Poisson noise lambda=20 | PSNR (dB): 33.17 -> 35.34 | Improved |
Closest papers by meaning
- End-to-End Low Cost Compressive Spectral Imaging with Spatial-Spectral Self-Attention (ECCV 2020 · similarity 0.92)
- Dual-Window Multiscale Transformer for Hyperspectral Snapshot Compressive Imaging (AAAI 2024 · similarity 0.91)
- SPECAT: SPatial-spEctral Cumulative-Attention Transformer for High-Resolution Hyperspectral Image Reconstruction (CVPR 2024 · similarity 0.88)
- Improving Spectral Snapshot Reconstruction with Spectral-Spatial Rectification (CVPR 2024 · similarity 0.87)
- Spectral Compressive Imaging via Chromaticity-Intensity Decomposition (NeurIPS 2025 · similarity 0.87)
- In2SET: Intra-Inter Similarity Exploiting Transformer for Dual-Camera Compressive Hyperspectral Imaging (CVPR 2024 · similarity 0.87)
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.
