Table 3 Effect of different multi-scale feature fusion methods on the detection result of LEVIR data set
From: Performance analysis of different DCNN models in remote sensing image object detection
Method | mAP@.5 | mAP@[.5:.95] | Time | Memory |
|---|---|---|---|---|
FPN | 0.863 | 0.628 | 11.7 ms | 419 M |
FPN+PAN | 0.882 | 0.639 | 12.3 ms | 420.8 M |
FPN+PAN+ASFF | 0.89 | 0.638 | 14.2 ms | 497.5 M |
FPN+ASFF | 0.907 | 0.662 | 19 ms | 501 M |