Table 10 Computation of feature space entropy (H) on the basis of quantitative evaluations of the respective GANs between unstained, H&E, and virtually stained images.
From: VISGAB: Virtual staining-driven GAN benchmarking for optimizing skin tissue histology
Quantitative evaluations | CycleGAN | CUTGAN | DCLGAN | H&E |
|---|---|---|---|---|
SSIM | 0.36–0.93 | 0.22–0.59 | 0.20–0.49 | 1 |
PSNR (dB) | 10.40–29.06 | 10.85–13.12 | 11.74–12.61 | ∞ |
FID | 27. 7–345.4 | 233.6–352.8 | 228.3–341.1 | 0 |
KID | 0.010–0.19 | 0.036–0.25 | 0.017–0.22 | 0 |
LPIPS | 0.08–0.78 | 0.49–0.71 | 0.48–0.76 | 0 |
HSFI | 0.23–0.81 | 0.19–0.58 | 0.21–0.59 | 1 |
Feature space entropy (H) | 2.1 | 3.5 | 3.8 | 4.2 |
