Fig. 1

Evaluation of PSSR2 against state-of-the-art microscopy image restoration methods on paired high-low-resolution test sets. a Super-resolution ResUNet model predictions of both PSSR2 and the prior PSSR implementation, in comparison to low-resolution inputs and high-resolution ground truth from a test set of 42 paired high-low-resolution images. For each set of corresponding images, a close up shows the same region of interest where PSSR2 predicts cellular structures (e.g., vesicles) more accurately than PSSR. Scale bar is 0.2\(\upmu\)m. b Image restoration performance comparison of ResUNet super-resolution models using the same test set of 42 images trained with either the prior PSSR implementation or PSSR2, benchmarked against a bilinear upscaling control. c Image denoising performance of PSSR2-trained ResUNet model against an EMDiffuse-pretrained model on a separate EMDiffuse test dataset of 32 paired high-low-quality images, with a bilinear upscaling control. All p-values measure the significance of metric performance of PSSR2 against a given image restoration method (b, PSSR; c, EMDiffuse) by paired t-test