Table 2 The comparison of the presented diffractive optical networks, in terms of (1) all-optical overlapping object classification accuracies on T2 and (2) the quality of the image reconstruction achieved through separately-trained, shallow, electronic networks (decoder).

From: Classification and reconstruction of spatially overlapping phase images using diffractive optical networks

  

Optical classification

Image reconstruction

Diffractive network

Number of detectors (\(M\) = 10)

Accuracy on T2 (%)

SSIM

PSNR (dB)

D2NN-D1

2 \(M\)

82.70

0.52 ± 0.12

15.09 ± 2.32

D2NN-D1d

4 \(M\)

85.82

0.57 ± 0.10

16.02 ± 2.21

D2NN-D2

\(M\) + 2

82.61

0.49 ± 0.10

14.55 ± 2.17

D2NN-D2d

2 \(M\) + 2

85.22

0.57 ± 0.12

15.60 ± 2.37

  1. The mean SSIM and PSNR values and their standard deviations were computed over the entire 10 K blind test inputs (T2).
Back to article page