Table 3 Faster R-CNN performance on the DeepPCB dataset.
Method | Input | mAP | \(AP_{s}\) | \(AP_{m}\) | mAR | \(AR_{s}\) | \(AR_{m}\) |
|---|---|---|---|---|---|---|---|
Faster R-CNN w/ Resnet50 (baseline1a) | \(640 \times 640\) | 74.4 | 74.4 | 74.8 | 79.9 | 79.4 | 80 |
Faster R-CNN w/ Resnet50 (baseline1b) | \(800 \times 800\) | 76.3 | 75.8 | 76.9 | 82.0 | 82.1 | 80.9 |
Faster R-CNN w/ Resnet101 (baseline2a) | \(640 \times 640\) | 73.8 | 72.8 | 74.8 | 79.9 | 79.2 | 80.0 |
Faster R-CNN w/ Resnet101 (baseline2b) | \(800 \times 800\) | 76.2 | 76.0 | 77.0 | 82.4 | 83.0 | 82.2 |
Faster R-CNN w/Resnet50+ACASEM | \(640 \times 640\) | 76.7 | 76.9 | 77.2 | 82.7 | 83.5 | 82.2 |
Faster R-CNN w/Resnet50+ACASEM | \(800 \times 800\) | 78.1 | 78.9 | 78.0 | 83.7 | 84.4 | 83.5 |
Faster R-CNN w/Resnet101+ACASEM | \(640 \times 640\) | 75.8 | 75.4 | 76.8 | 81.9 | 81.6 | 82 |
Faster R-CNN w/Resnet101+ACASEM | \(800 \times 800\) | 77.1 | 78.1 | 77.4 | 83.5 | 84.1 | 83.2 |
