Table 5 Validate the contribution of each of the modules in Faster R-CNN and CenterNet object detection frameworks. The best results are highlighted in bold.
Method | C3kHR | EAFN | ADown | P | R | mAP |
|---|---|---|---|---|---|---|
Faster R-CNN | \(\times\) | \(\times\) | \(\times\) | 78.9 | 8.7 | 17.7 |
Faster R-CNN-A | \(\checkmark\) | \(\times\) | \(\times\) | 79.5 | 11.2 | 19.0 |
Faster R-CNN-B | \(\times\) | \(\checkmark\) | \(\times\) | 79.9 | 10.8 | 19.7 |
Faster R-CNN-C | \(\times\) | \(\times\) | \(\checkmark\) | 79.1 | 8.7 | 17.8 |
Faster R-CNN-D | \(\checkmark\) | \(\checkmark\) | \(\checkmark\) | 80.3 | 13.2 | 21.4 |
CenterNet | \(\times\) | \(\times\) | \(\times\) | 88.9 | 6.2 | 20.1 |
CenterNet-A | \(\checkmark\) | \(\times\) | \(\times\) | 89.0 | 8.4 | 23.2 |
CenterNet-B | \(\times\) | \(\checkmark\) | \(\times\) | 88.9 | 8.0 | 23.9 |
CenterNet-C | \(\times\) | \(\times\) | \(\checkmark\) | 88.8 | 6.3 | 20.0 |
CenterNet-D | \(\checkmark\) | \(\checkmark\) | \(\checkmark\) | 89.4 | 10.7 | 25.3 |
