Table 7 5-fold cross-validation results of best-performing models in each method.
Task | Metric | Method | Fold 1 | Fold 2 | Fold 3 | Fold 4 | Fold 5 | Mean |
|---|---|---|---|---|---|---|---|---|
Whole building façade | F1 score | Swin Transformer | 90.22 | 90.09 | 90.22 | 90.22 | 90 | 90.15 |
ViT | 88.93 | 88.79 | 88.93 | 88.93 | 88.69 | 88.85 | ||
PVT | 88.35 | 88.2 | 88.35 | 88.35 | 88.09 | 88.27 | ||
MobileViT | 87.43 | 87.28 | 87.43 | 87.43 | 87.17 | 87.35 | ||
Axial Transformer | 86.96 | 86.8 | 86.96 | 86.96 | 86.69 | 86.87 | ||
Accuracy | Swin Transformer | 91.78 | 91.62 | 91.78 | 91.78 | 91.51 | 91.69 | |
ViT | 90.58 | 90.39 | 90.58 | 90.58 | 90.26 | 90.48 | ||
PVT | 89.99 | 89.78 | 89.99 | 89.99 | 89.64 | 89.88 | ||
MobileViT | 89.12 | 88.91 | 89.12 | 89.12 | 88.76 | 89.01 | ||
Axial Transformer | 88.69 | 88.48 | 88.69 | 88.69 | 88.34 | 88.58 | ||
First story | F1 score | Swin Transformer | 89.74 | 89.59 | 89.74 | 89.74 | 89.49 | 89.66 |
ViT | 88.29 | 88.14 | 88.29 | 88.29 | 88.03 | 88.21 | ||
PVT | 87.69 | 87.53 | 87.69 | 87.69 | 87.41 | 87.6 | ||
MobileViT | 86.74 | 86.57 | 86.74 | 86.74 | 86.46 | 86.65 | ||
Axial Transformer | 86.12 | 85.95 | 86.12 | 86.12 | 85.83 | 86.03 | ||
Accuracy | Swin Transformer | 92.29 | 92.14 | 92.29 | 92.29 | 92.04 | 92.21 | |
ViT | 90.92 | 90.77 | 90.92 | 90.92 | 90.66 | 90.84 | ||
PVT | 90.56 | 90.4 | 90.56 | 90.56 | 90.28 | 90.47 | ||
MobileViT | 89.64 | 89.47 | 89.64 | 89.64 | 89.36 | 89.55 | ||
Axial Transformer | 89.26 | 89.09 | 89.26 | 89.26 | 88.97 | 89.17 |
