Table 7 Accuracy of the reference deep learning models for each PAIs technique and bona fide samples on the few-shot and unbalanced data partition.

From: Synthetic dataset of ID and Travel Documents

Model

Bona Fide

Inpainting

Crop & Replace

EfficientNet

89.5 ± 3.07

86.8 ± 2.42

91.3 ± 5.04

ResNet

91.8 ± 2.67

91.2 ± 2.72

91.6 ± 8.51

VIT

91.6 ± 5.94

88.5 ± 4.80

96.1 ± 2.11

TransFG

85.7 ± 5.60

83.6 ± 6.50

87.1 ± 7.10

CoAARC

67.4 ± 4.60

66.2 ± 6.38

70.7 ± 8.27

Back to article page