Fig. 28: Performance comparison between the fringe-to-phase linked deep-learning method and the deep-learning approach combining the physical model of the phase-shifting method.
a For the end-to-end network structure, the fringe image can be fed into DNN1 to directly output the corresponding wrapped phases. b However, such an end-to-end approach makes the training process fails to converge because it is difficult to follow the 2π phase truncation. c Our group proposed to incorporate the physical model of the traditional phase-shifting method into deep learning and applied deep learning to predict from the fringe image the numerator and denominator of the arctangent function used to calculate the phase information50. d Such a physics-informed strategy results in a stable convergence to the minimum training and validation loss. b, d Adapted with permission from ref. 50. Distributed under Creative Commons (CC BY 4.0) license https://creativecommons.org/licenses/by/4.0/legalcode
