Table 5 Comparison of The Impact of Different Data Homogenization Algorithms on The Final Recognition Results
From: Multi-modal ancient scripts recognition via deep learning with data homogenization and augmentation
Recognition Models | The recognition accuracy of data homogenization with different models | |||||
|---|---|---|---|---|---|---|
U-net | Diffusion | VAE | ||||
Oracle bone Inscriptions | Bronze Inscriptions | Oracle bone Inscriptions | Bronze Inscriptions | Oracle bone Inscriptions | Bronze Inscriptions | |
AlexNet | 0.861 | 0.763 | 0.699 | 0.552 | 0.774 | 0.756 |
VGG19 | 0.904 | 0.727 | 0.736 | 0.568 | 0.802 | 0.689 |
ResNet50 | 0.900 | 0.803 | 0.755 | 0.607 | 0.805 | 0.738 |
ConvNext | 0.880 | 0.778 | 0.740 | 0.564 | 0.790 | 0.575 |
EfficientNet | 0.863 | 0.744 | 0.732 | 0.561 | 0.791 | 0.742 |
ShuffleNet | 0.902 | 0.754 | 0.743 | 0.537 | 0.805 | 0.735 |
ViT | 0.774 | 0.441 | 0.629 | 0.312 | 0.623 | 0.371 |
Swin Transformer | 0.858 | 0.718 | 0.745 | 0.501 | 0.744 | 0.710 |
