Phase mask-based lensless image reconstruction using self-attention

Phase mask-based lensless imaging systems acquire multiplexed measurements that are not immediately interpretable by the human eye, requiring reconstruction algorithms to recover the final image. Current reconstruction methods suffer from long run times, extensive parameter tuning requirements, and model mismatch artifacts. However, attention-based neural networks have the ability to capture large-scale features in measurements, circumventing several issues present in previous methods. In this work, we use a self-attention-based machine learning architecture for lensless image reconstruction and compare our neural network to current reconstruction approaches.