Table 9 Comparison of performance on the DeepPCB dataset.
Method | Resolution | mAP | mAR |
|---|---|---|---|
YOLOv350 | \(640\times 640\) | 70.71 | 78.6 |
SSD6 | \(640\times 640\) | 72.51 | 77.7 |
ID-YOLO51 | \(640\times 640\) | 71.48 | 77.62 |
Lightnet52 | \(640\times 640\) | 76.22 | 81.01 |
Cascade R-CNN w/SwinT-T | \(640\times 640\) | 77.24 | 82.74 |
Cascade R-CNN w/SwinT-S | \(640\times 640\) | 77.41 | 83.28 |
DDTR w/ReSwinT-T53 | \(640\times 640\) | 78.75 | 83.53 |
DDTR w/ResSwinT-S53 | \(640\times 640\) | 78.62 | 83.9 |
Faster R-CNN w/Resnet50 + ACASEM (Ours) | \(640\times 640\) | 76.7 | 82.7 |
Double-Head R-CNN w/Resnet50 + ACASEM (Ours) | \(640\times 640\) | 76.7 | 81.2 |
Cascade R-CNN w/Resnet50 + ACASEM (Ours) | \(640\times 640\) | 78.4 | 82.9 |
Faster R-CNN w/Resnet50 + ACASEM (Ours) | \(800\times 800\) | 78.1 | 83.7 |
Double-Head R-CNN w/Resnet50 + ACASEM (Ours) | \(800\times 800\) | 78.5 | 82.8 |
Cascade R-CNN w/Resnet50 + ACASEM (Ours) | \(800\times 800\) | 79.5 | 83.7 |
