Fig. 7: Cascaded P-D2NN with a joint optimization strategy.
From: Pyramid diffractive optical networks for unidirectional image magnification and demagnification
a The design concept of the cascaded pyramid diffractive network: two P-D2NN units (P1 and P2) are cascaded, where the output plane of P1 serves as the input plane for P2. The system is optimized using three distinct loss functions—one for each P-D2NN and one for the end-to-end system performance (see “Methods” section). b Cascaded image magnification and FOV sizes. The color coding on the boundary represents the size of the FOVs and is consistent for all the cascaded designs. c Joint testing of the cascaded P-D2NN architecture with a magnification factor of \(M=3\times 3=9\). In the forward direction, the cascaded network projects a magnified version of the input image, while in the backward direction, only speckle-like noise is observed, blocking the image formation—as desired. All the images are individually normalized
