Fig. 5: Effect of continual learning.
From: Deep learning at the edge enables real-time streaming ptychographic imaging
a Evolution of structural similarity against the ground truth for two selected test-sample areas during continual learning. Those areas are shown respectively in (b) and (c). The training set consists of pairs of iteratively retrieved phases and corresponding diffraction patterns on an area with randomly etched features. After about 80,000 sets, the training data starts to include the edge of the patterned areas. The scale bar is 500 nm. Source data are provided as a Source data file.
