Code and dataset: https://github.com/cabooster/SRDTrans
SRDTrans can be used to remove noise from fluorescence images in a self-supervised manner. Benefiting from the new spatial redundancy orthogonal sampling, it has no dependence on high imaging speed and is complementary to our previously proposed DeepCAD. A lightweight spatiotemporal transformer architecture was developed to capture long-range dependencies and high-resolution features at a low computational cost. SRDTrans can overcome the inherent spectral bias of CNNs and restore high-frequency information without producing over-smoothed structures and distorted fluorescence traces.