Research map / Label-efficient and robust learning / Self-supervised and contrastive
SGDE: Self-supervised Geometry Degradation Estimation Framework for Coded Aperture Compressive Spectral Imaging
Yuqiao He, Xiaoyan Liu, Jianxu Mao, Yaonan Wang, Hui Zhang, Lizhu Liu, Yurong Chen, Wenbin He
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 self-supervised affine mask estimation (SK-LCTC) to LCTC | LCTC · KAIST simulation, combined perturbation | PSNR: 21.46 -> 36.15 | Improved |
| add multi-kernel estimation with 8 kernels to LCTC | LCTC · KAIST simulation, combined perturbation | PSNR: 21.46 -> 36.04 | Improved |
| use multi-kernel instead of single-kernel under large rotation perturbation (ablation) | SGDE · KAIST simulation, 0.4 deg rotation | PSNR: SK-LCTC drops 10 dB at 0.4 deg rotation; MK-LCTC stays above 35 dB | Improved |
| add affine estimation to unperturbed reconstruction | LCTC · KAIST simulation, no perturbation | PSNR: 38.51 -> 38.64 (SK), 38.39 (MK) | No clear effect |
| place kernel initializations far from the true perturbation (forward order) vs near it (reverse order) (ablation) | MK-LCTC · KAIST Scene 01 | PSNR: 25.50 -> 36.30 with 1 kernel | Improved |
| increase number of kernels to 8 when initialized forward (ablation) | MK-LCTC · KAIST Scene 01 | PSNR: 28.62 -> 36.13 (7 -> 8 kernels) | Improved |
| increase number of kernels when initialized in reverse order (ablation) | MK-LCTC · KAIST Scene 01 | PSNR: 36.30 -> 36.21 | No clear effect |
| transfer estimated affine parameters to existing reconstruction methods | LCTC, DDIP, RDLUF, MIDET, DERNN · TSA real scenes | reconstruction quality (qualitative) | Improved |
| use a single kernel with wide output range for estimation (ablation) | SK-LCTC · KAIST Scene 01, perturbation beyond 1 pixel | PSNR | Hurt |
Closest papers by meaning
- Degradation-Aware Unfolding Half-Shuffle Transformer for Spectral Compressive Imaging (NeurIPS 2022 · similarity 0.89)
- End-to-End Low Cost Compressive Spectral Imaging with Spatial-Spectral Self-Attention (ECCV 2020 · similarity 0.89)
- SAH-SCI: Self-Supervised Adapter for Efficient Hyperspectral Snapshot Compressive Imaging (ECCV 2024 · similarity 0.89)
- Residual Degradation Learning Unfolding Framework With Mixing Priors Across Spectral and Spatial for Compressive Spectral Imaging (CVPR 2023 · similarity 0.88)
- Spectral Compressive Imaging via Chromaticity-Intensity Decomposition (NeurIPS 2025 · similarity 0.87)
- Dual-Window Multiscale Transformer for Hyperspectral Snapshot Compressive Imaging (AAAI 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.
