Deep Tensor ADMM-Net for Snapshot Compressive Imaging

TLDR

Snapshot compressive imaging (SCI) captures high‑dimensional video by mapping multiple frames into a single measurement, but existing model‑based decoders require exhaustive tuning and are too slow for practical use. This work introduces a deep tensor ADMM‑Net that delivers high‑quality video reconstruction in seconds. The network unfolds a tensor ADMM algorithm into a layer‑wise deep architecture and learns a low‑rank tensor domain via stochastic gradient descent, eliminating the need for a pre‑specified sparse representation. On public video datasets, the method improves PSNR by 0.8–2.5 dB, SSIM by 0.07–0.1, and achieves 1,500–3,600× speedups over state‑of‑the‑art, while real‑world experiments show comparable visual quality in much less time.

Abstract

Snapshot compressive imaging (SCI) systems have been developed to capture high-dimensional (≥ 3) signals using low-dimensional off-the-shelf sensors, i.e., mapping multiple video frames into a single measurement frame. One key module of a SCI system is an accurate decoder that recovers the original video frames. However, existing model-based decoding algorithms require exhaustive parameter tuning with prior knowledge and cannot support practical applications due to the extremely long running time. In this paper, we propose a deep tensor ADMM-Net for video SCI systems that provides high-quality decoding in seconds. Firstly, we start with a standard tensor ADMM algorithm, unfold its inference iterations into a layer-wise structure, and design a deep neural network based on tensor operations. Secondly, instead of relying on a pre-specified sparse representation domain, the network learns the domain of low-rank tensor through stochastic gradient descent. It is worth noting that the proposed deep tensor ADMM-Net has potentially mathematical interpretations. On public video data, the simulation results show the proposed method achieves average 0.8 ~ 2.5 dB improvement in PSNR and 0.07 ~ 0.1 in SSIM, and 1500× ~ 3600× speedups over the state-of-the-art methods. On real data captured by SCI cameras, the experimental results show comparable visual results with the state-of-the-art methods but in much shorter running time.