High performance “non-local” generic face reconstruction model using the lightweight Speckle-Transformer (SpT) UNet

Significant progress has been made in computational imaging (CI), in which deep convolutional neural networks (CNNs) have demonstrated that sparse speckle patterns can be reconstructed. However, due to the limited “local” kernel size of the convolutional operator, for the spatially dense patterns, such as the generic face images, the performance of CNNs is limited. Here, we propose a “non-local” model, termed the Speckle-Transformer (SpT) UNet, for speckle feature extraction of generic face images. It is worth noting that the lightweight SpT UNet reveals a high efficiency and strong comparative performance with Pearson Correlation Coefficient (PCC), and structural similarity measure (SSIM) exceeding 0.989, and 0.950, respectively.