Proximal Algorithm Unrolling: Flexible and Efficient Reconstruction Networks for Single-Pixel Imaging

TL;DR

An efficient deep image restorer (DIR) for the unrolling of HQS and ADMM and a general proximal trajectory (PT) loss function is proposed to train HQS/ADMM-unrolling networks such that learned DIR approximates the proximal operator of an ideal explicit restoration regularizer.

Abstract

Deep-unrolling and plug-and-play (PnP) approaches have become the de-facto standard solvers for single-pixel imaging (SPI) inverse problem. PnP approaches, a class of iterative algorithms where regularization is implicitly performed by an off-the-shelf deep denoiser, are flexible for varying compression ratios (CRs) but are limited in reconstruction accuracy and speed. Conversely, unrolling approaches, a class of multi-stage neural networks where a truncated iterative optimization process is transformed into an end-to-end trainable network, typically achieve better accuracy with faster inference but require fine-tuning or even retraining when CR changes. In this paper, we address the challenge of integrating the strengths of both classes of solvers. To this end, we design an efficient deep image restorer (DIR) for the unrolling of HQS (half quadratic splitting) and ADMM (alternating direction method of multipliers). More importantly, a general proximal trajectory (PT) loss function is proposed to train HQS/ADMM-unrolling networks such that learned DIR approximates the proximal operator of an ideal explicit restoration regularizer. Extensive experiments demonstrate that, the resulting proximal unrolling networks can not only flexibly handle varying CRs with a single model like PnP algorithms, but also outperform previous CR-specific unrolling networks in both reconstruction accuracy and speed. Source codes and models are available at https://github.com/pwangcs/ProxUnroll.

Figures (7)

• Figure 1: Average PSNR (left) of different methods at compressive ratios (CRs) $\{0.01,0.04,0.10,0.25\}$ in which only methods in the yellow area can flexibly handle varying CRs with a single model. Optimization trajectories (right) of four classes of methods for solving inverse problems. Our ProxUnroll efficiently achieves SOTA performance with high flexibility and fast convergence.
• Figure 2: Degradations under different compressive ratios (CRs).
• Figure 3: Proximal algorithm unrolling via trajectory loss.
• Figure 4: Illustration of the proposed deep image restorer (DIR) $\mathcal{R}_{\theta}$.
• Figure 5: Visual comparisons on "barbara" from Set11 dataset at CR $=0.04$.