Efficiency analysis of generative adversarial networks for single pixel imaging

The single-pixel camera provides a prospective tool for imaging beyond conventional pixel-matrix-based devices. In recent years, neural networks have become a part of singlepixel imaging as a method to restore an image from  intensity measurements computationally. Generative adversarial networks (GANs) are particularly well suited for this task. In this paper, we investigate the performance of a generative adversarial least squares network in the task of image reconstruction from a single-pixel camera at low sampling rates. We demonstrate that stable successful image reconstruction is possible at sampling rates around 8%, and that the reconstructed images should match the  structure of the images present in the training sample.