Table 5 Cascade R-CNN performance on the DeepPCB dataset.
Method | Input | mAP | \(AP_{s}\) | \(AP_{m}\) | mAR | \(AR_{s}\) | \(AR_{m}\) |
|---|---|---|---|---|---|---|---|
Cascade R-CNN w/ Resnet50 (baseline5a) | \(640 \times 640\) | 75.3 | 75.0 | 75.6 | 80.2 | 80.2 | 80.3 |
Cascade R-CNN w/ Resnet50 (baseline5b) | \(800 \times 800\) | 77.1 | 76.8 | 77.2 | 81.7 | 81.5 | 81.6 |
Cascade R-CNN w/ Resnet101 (baseline6a) | \(640 \times 640\) | 75.2 | 74.7 | 75.6 | 80.3 | 80.4 | 80.2 |
Cascade R-CNN w/ Resnet101 (baseline6b) | \(800 \times 800\) | 76.4 | 76.0 | 76.7 | 81.0 | 80.2 | 81.3 |
Cascade R-CNN w/Resnet50+ACASEM | \(640 \times 640\) | 78.4 | 77.3 | 78.8 | 82.9 | 82.7 | 83.1 |
Cascade R-CNN w/Resnet50+ACASEM | \(800 \times 800\) | 79.5 | 78.9 | 79.8 | 83.7 | 83.1 | 83.9 |
Cascade R-CNN w/Resnet101+ACASEM | \(640 \times 640\) | 77.8 | 76.4 | 78.3 | 82.5 | 81.7 | 82.7 |
Cascade R-CNN w/Resnet101+ACASEM | \(800 \times 800\) | 78.9 | 77.7 | 79.4 | 83.2 | 82.6 | 83.4 |
