Table 18 Ablation study individual component contributions.

From: Novel dual gland GAN architecture improves human protein localization classification using salivary and pituitary gland inspired loss functions

Configuration

FID (↓)

IS (↑)

MS-SSIM diversity (↓)

F1-score (↑)

Training epochs

Full dual-gland GAN

8.47

9.83

0.187

0.853

78

SG-loss only

12.34

8.12

0.245

0.782

95

PG-loss only

14.67

7.85

0.289

0.761

102

Without adaptive fusion

15.23

7.43

0.312

0.739

134

Without multi-scale features

16.89

6.98

0.341

0.715

145

Without biological constraints

18.45

6.52

0.378

0.692

156

Traditional single generator

22.35

5.21

0.493

0.531

250 + 

Back to article page