Research map / Image generation / Image restoration
Semantic-Embedded Unsupervised Spectral Reconstruction From Single RGB Images in the Wild
Zhiyu Zhu, Hui Liu, Junhui Hou, Huanqiang Zeng, Qingfu 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 |
|---|---|---|---|
| add back-projection constraint S_bp (ablation) | adversarial-only baseline · ICVL (synthetic RGB, 20 test images) | PSNR: 14.47 -> 20.48 | Improved |
| add positive constraint S_pos (ablation) | S_bp model · ICVL (synthetic RGB, 20 test images) | PSNR: 20.48 -> 21.72 | Improved |
| add smoothness constraint S_1 (ablation) | S_bp + S_pos model · ICVL (synthetic RGB, 20 test images) | PSNR: 21.72 -> 26.11 | Improved |
| add L1 gradient clipping (ablation) | S_bp + S_pos + S_1 model · ICVL (synthetic RGB, 20 test images) | PSNR: 26.11 -> 34.51 | Improved |
| add semantic-embedded regularization S_sem (ablation) | model with L1 gradient clipping · ICVL (synthetic RGB, 20 test images) | PSNR: 34.51 -> 37.83 | Improved |
| use L1 gradient clipping instead of 1-GP (ablation) | GAN regularization · ICVL (synthetic RGB, 20 test images) | PSNR: 28.15 -> 37.83 | Improved |
| use L1 gradient clipping instead of 0-GP (ablation) | GAN regularization · ICVL (synthetic RGB, 20 test images) | PSNR: 33.33 -> 37.83 | Improved |
| train with PASCAL-Context instead of Cityscapes RGB source (ICVL HS) (ablation) | Ours · ICVL (synthetic RGB, 20 test images) | PSNR: 37.68 -> 37.83 | Improved |
| train with ICVL instead of NTIRE'20 HS source (PASCAL-Context RGB) (ablation) | Ours · ICVL (synthetic RGB, 20 test images) | PSNR: 35.87 -> 37.83 | Improved |
| use unsupervised Ours instead of unsupervised DBF | DBF · ICVL (synthetic RGB, 20 test images) | PSNR: 30.75 -> 37.83 | Improved |
| use Ours instead of supervised FM-Net on ICVL | FM-Net · ICVL (synthetic RGB, 20 test images) | PSNR: 40.26 -> 37.83 | Hurt |
| use Ours instead of supervised FM-Net on NTIRE'20 | FM-Net · NTIRE'20 (synthetic RGB) | PSNR: 34.25 -> 37.53 | Improved |
Closest papers by meaning
- PhySpec: Physically Consistent Spectral Reconstruction via Orthogonal Subspace Decomposition and Self-Supervised Meta-Auxiliary Learning (ICML 2025 · similarity 0.90)
- Deeply Learned Filter Response Functions for Hyperspectral Reconstruction (CVPR 2018 · similarity 0.90)
- Joint Camera Spectral Sensitivity Selection and Hyperspectral Image Recovery (ECCV 2018 · similarity 0.90)
- EigenSR: Eigenimage-Bridged Pre-Trained RGB Learners for Single Hyperspectral Image Super-Resolution (AAAI 2025 · similarity 0.87)
- Hyperspectral Image Reconstruction via Combinatorial Embedding of Cross-Channel Spatio-Spectral Clues (AAAI 2024 · similarity 0.87)
- FRN: Fractal-Based Recursive Spectral Reconstruction Network (NeurIPS 2025 · 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.
