Fig. 4: DeepFocus convergence on an unseen sample and recalibration for a different setup.
From: DeepFocus: fast focus and astigmatism correction for electron microscopy
a Model convergence (same model as in Fig. 3d; n = 10 input patch pairs) on a sample of tin on carbon (not contained in the training data) using setup A at 100 ns dwell time. b Image from a at iterations 0 and 10. Scale bar is 1 µm. c The same model as applied to tin on carbon on setup B (see Methods for imaging parameters). d Convergence of the fine-tuned DeepFocus model (last three fully connected layers re-trained) after 50k training iterations using 100 automatically acquired samples at 10 different locations with setup B (Methods). Data in all plots are presented as mean values ± SD. Source data are provided as a Source Data file.
