Skip to main content
. 2021 Apr 12;12:2179. doi: 10.1038/s41467-021-22461-0

Fig. 3. Deep adjoint neural network for LIFT reconstruction.

Fig. 3

a The network architecture of DANN. The measurement data is firstly processed by the adjoint operator AT of the system and then passed to a convolutional network with skip connections. The detailed network structure and parameters are provided in Supplementary Note 8. b DANN reconstruction results of a subset of the experimentally measured test images, arranged in a montage format with an image field of view ~36 mm. Left: ground truth images; Right: DANN reconstruction. c Comparison of image reconstruction results using iterative methods (top row) and DANN (bottom row). The helical fiber structure is crafted and overlaid in each frame for visual guidance.