Snapshot spectral compressive imaging reconstruction using convolution and contextual Transformer

Abstract

Spectral compressive imaging (SCI) is able to encode a high-dimensional hyperspectral image into a two-dimensional snapshot measurement, and then use algorithms to reconstruct the spatio-spectral data-cube. At present, the main bottleneck of SCI is the reconstruction algorithm, and state-of-the-art (SOTA) reconstruction methods generally face problems of long reconstruction times and/or poor detail recovery. In this paper, we propose a hybrid network module, namely, a convolution and contextual Transformer (CCoT) block, that can simultaneously acquire the inductive bias ability of convolution and the powerful modeling ability of Transformer, which is conducive to improving the quality of reconstruction to restore fine details. We integrate the proposed CCoT block into a physics-driven deep unfolding framework based on the generalized alternating projection (GAP) algorithm, and further propose the GAP-CCoT network. Finally, we apply the GAP-CCoT algorithm to SCI reconstruction. Through experiments on a large amount of synthetic data and real data, our proposed model achieves higher reconstruction quality ( > 2 dB in peak signal-to-noise ratio on simulated benchmark datasets) and a shorter running time than existing SOTA algorithms by a large margin. The code and models are publicly available at https://github.com/ucaswangls/GAP-CCoT.