Fig. 16: Temporal phase retrieval based on deep learning.

a Schematic of using deep learning for temporal phase retrieval using three phase-shifted sinusoidal fringe images. b, c The reconstruction results of a complex scene based on the deep-learning method illustrated in (a). d One raw image containing both low-modulation and saturated fringes. e–g The 3D reconstruction results using the traditional phase-shifting method. h–j The 3D reconstruction results using the deep-learning-based HDR method. a–c Adapted with permission from ref. ³¹², Optica Publishing. d–j Adapted with permission from ref. ³¹⁷, Copyright (2021), with permission from Elsevier