CVPR Findings2026
Semantic-Aware Spectral Reconstruction: A Spectral Library-Aided Unsupervised Method Based on the Diffusion Model
Keli Deng, Yuntao Qian
Abstract
Unsupervised spectral reconstruction (SR) has attracted increasing attention, aiming to reconstruct a hyperspectral image (HSI) from a single RGB image without requiring paired RGB-HSI training data. However, existing methods often fail to fully leverage semantic information in RGB images and lack reliable mechanisms for generating high-fidelity spectra. In this paper, we propose Lib-SRDM, a spectral library-aided unsupervised method based on the diffusion model that leverages RGB segmentation to extract semantics for enhancing SR. Specifically, we dynamically construct two training-free spectral models to represent the semantic-aware spectral distribution during inference, which are fused with low-level details from an HSI diffusion model pre-trained on HSIs only. Given an RGB image, segmentation identifies instances, and relevant spectra are retrieved from a spectral library. For each instance, we build two training-free spectral models, named the spectral diffusion model and the low-rank model, to capture its semantic-aware spectral distribution. These models are then plugged into the reverse diffusion process to restrict the solution space using semantics and library spectra, while the pre-trained HSI diffusion model refines detailed spatio-spectral structure to compensate for potential defects. Extensive experiments on three datasets demonstrate that our method outperforms state-of-the-art methods in terms of spectral fidelity and visual quality. The code is available at https://github.com/Dmsw/Lib-SRDM .
Ask Lune
Know this paper in minutes, not an afternoon.
Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.
Lineage
- Diffusion Models Beat GANs on Image Synthesis cites Denoising Diffusion Probabilistic Models
- Diffusion Posterior Sampling for Linear Inverse Problem Solving: A Filtering Perspective cites Diffusion Models Beat GANs on Image Synthesis
- Diffusion Posterior Sampling for Linear Inverse Problem Solving: A Filtering Perspective cites Denoising Diffusion Probabilistic Models
- Directly Denoising Diffusion Models cites Diffusion Models Beat GANs on Image Synthesis
- Directly Denoising Diffusion Models cites Denoising Diffusion Probabilistic Models
Builds on10
- Denoising Diffusion Probabilistic ModelsJonathan Ho, Ajay Jain, Pieter AbbeelNeurIPS 2020 · 35,902 citations
- Diffusion Models Beat GANs on Image SynthesisPrafulla Dhariwal, Alexander Quinn NicholNeurIPS 2021 · 13,211 citations
- Directly Denoising Diffusion ModelsDan Zhang, Jingjing Wang, Feng LuoICML 2024 · 11,724 citations
- Mask-guided Spectral-wise Transformer for Efficient Hyperspectral Image ReconstructionYuanhao Cai, Jing Lin, Xiaowan Hu, Haoqian Wang et al.CVPR 2022 · 310 citations
- Diffusion Posterior Sampling for Linear Inverse Problem Solving: A Filtering PerspectiveZehao Dou, Yang SongICLR 2024 · 162 citations
Related papers
- Semantic-embedded Unsupervised Spectral Reconstruction from Single RGB Images in the WildZhiyu Zhu, Hui Liu, Junhui Hou, Huanqiang Zeng et al.ICCV 2021 · 27 citations
- Unraveling Metameric Dilemma for Spectral Reconstruction: A High-Fidelity Approach via Semi-Supervised LearningXingxing Yang, Jie Chen, Zaifeng YangNeurIPS 2025 · 2 citations
- HIR-Diff: Unsupervised Hyperspectral Image Restoration Via Improved Diffusion ModelsLi Pang, Xiangyu Rui, Long Cui, Hongzhong Wang et al.CVPR 2024 · 32 citations
- DDS2M: Self-Supervised Denoising Diffusion Spatio-Spectral Model for Hyperspectral Image RestorationYuchun Miao, Lefei Zhang, Liangpei Zhang, Dacheng TaoICCV 2023 · 64 citations
- FRN: Fractal-Based Recursive Spectral Reconstruction NetworkGe Meng, Zhongnan Cai, Ruizhe Chen, Jingyan Tu et al.NeurIPS 2025 · 1 citation
