Fig. 1: DNN for real-time phase retrieval of imperfect single-pulse diffraction patterns. | npj Computational Materials

Fig. 1: DNN for real-time phase retrieval of imperfect single-pulse diffraction patterns.

From: Deep-learning real-time phase retrieval of imperfect diffraction patterns from X-ray free-electron lasers

Fig. 1

a Schematic diagram of data generation and the network training. b Schematic diagram of DPR. The network consists of a WPC-based encoder and a two-stage decoder, including a base decoder and a diffraction-compensated decoder (+D). c, d Evolution of validation loss during training iterations (c) and evaluation metrics (d) for WPC-, PC-, and FFC-based encoders with and without +D. The boxes and whiskers represent the average and standard errors of each metric, respectively. Differences with WPC + D are verified by the Mann-Whitney U test (not indicated, p ≤ 10−8; ***10−8 < p ≤ 0.001; **0.001 < p ≤ 0.01; *0.01 < p ≤ 0.05; ns, p > 0.05).

Back to article page