Fig. 2: Generalization of trained encoder-decoder pairs to previously unseen handwritten digit objects.
From: Learning diffractive optical communication around arbitrary opaque occlusions
a For different values of the occlusion width \({w}_{o}\), the performances of trained encoder-decoder pairs with different numbers of decoder layers (\(L\)) are depicted for comparison. b t-SNE-based visualization of the electronic encoder outputs for the nine different designs of Fig. 2a.
