Table 17 Model robustness and generalization. Significant values are in bold.

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

Model

Validation-test gap (Before) (%)

Validation-test gap (After) (%)

Cross-dataset performance (Before) (%)

Cross-dataset performance (After) (%)

Robustness score

ResNet-50

6.8

3.2

73.5

85.3

1.162

DenseNet-121

6.1

2.9

75.2

86.7

1.153

EfficientNet-B3

5.7

2.4

76.3

88.1

1.155

Vision Transformer

5.9

2.1

74.8

88.5

1.183

MobileNetV3

7.3

3.7

71.4

82.6

1.157

Inception-v4

6.4

2.8

74.1

85.9

1.159

Swin Transformer

5.4

2.0

76.9

89.3

1.161

ConvNeXt

5.8

2.5

75.6

87.4

1.156

RegNet-Y

6.9

3.3

72.8

84.1

1.155

NFNet

5.2

1.8

77.4

89.8

1.160

Average

6.2

2.7

74.8

86.8

1.160

Back to article page