R2H-Diff: Guided Spectral Diffusion Model for RGB-to-Hyperspectral Reconstruction

TL;DR

R2H-Diff is a diffusion-based framework for RGB-to-hyperspectral image reconstruction that achieves high fidelity with low model complexity by using a guided iterative refinement process and specialized modules.

Contribution

The paper introduces R2H-Diff, a novel spectral diffusion model with a guided refinement process and spectral-adaptive modules, enabling efficient high-quality hyperspectral reconstruction.

Findings

R2H-Diff achieves 35.37 dB PSNR on NTIRE2022 with only 0.58M parameters.

The method balances reconstruction quality and computational efficiency effectively.

It outperforms existing methods in spectral fidelity with minimal model complexity.

Abstract

RGB-to-hyperspectral image reconstruction is a highly ill-posed inverse problem, since multiple plausible spectral distributions may correspond to the same RGB observation. Existing regression-based methods usually learn a deterministic mapping, which limits their ability to model reconstruction uncertainty and often leads to over-smoothed spectral responses. Although diffusion models provide strong distribution modeling capability, their direct application to hyperspectral reconstruction remains challenging due to the high spectral dimensionality, strong inter-band correlations, and strict requirement for spectral fidelity. To this end, we propose R2H-Diff, an efficient diffusion-based framework tailored for RGB-to-HSI reconstruction. Specifically, R2H-Diff formulates spectral recovery as a conditional iterative refinement process, enabling progressive reconstruction under RGB guidance. We proposed a Guided Spectral Refinement Module for RGB-conditioned feature fusion and a Hyperspectral-Adaptive Transposed Attention module for efficient spatial--spectral dependency modeling. Furthermore, a normalization-free denoising backbone is adopted to preserve spectral amplitude consistency, while a task-adapted linear noise schedule enables high-quality reconstruction with only five denoising steps. Extensive experiments on NTIRE2022, CAVE, and Harvard demonstrate that R2H-Diff achieves a favorable balance between reconstruction quality and computational efficiency. Notably, on NTIRE2022, R2H-Diff obtains 35.37 dB PSNR with a sub-million-parameter model of 0.58M parameters and 12.25G FLOPs, achieving the lowest model complexity among the evaluated methods while maintaining strong reconstruction fidelity.