Table 2 Evaluation of different backbones on the proposed DBD400 testing set
From: STEF: a Swin Transformer-Based Enhanced Feature Pyramid Fusion Model for Dongba character detection
Backbones | Precision | Recall | F-measure | FPS |
|---|---|---|---|---|
PyramidVisionTransformer [38] | 79.52 | 80.34 | 79.93 | 1.62 |
ResNet18 [36] | 82.33 | 76.39 | 79.25 | 2.17 |
ResNet50 [36] | 86.77 | 82.14 | 82.14 | 1.94 |
MobileNetV2 [37] | 85.70 | 81.84 | 83.72 | 2.14 |
SwinTransformer-10 [41] | 86.31 | 81.80 | 84.00 | 4.86 |
SwinTransformer-12 [41] | 87.47 | 83.02 | 85.18 | 4.46 |
SwinTransformer-8 [41] | 87.28 | 86.08 | 86.67 | 4.93 |
Res2Net [39] | 86.17 | 89.52 | 87.81 | 1.77 |
ResNeXt [40] | 88.61 | 86.87 | 87.73 | 1.82 |
SwinTransformer-6 [41] | 88.88 | 88.65 | 88.76 | 5.03 |
