Table 7 Comparison on VisDrones after using the normal model instead of the Backbone part. Including (Faster-RCNN19 and Cascade-RCNN62), reporting the average precision (AP) for different IOU thresholds, respectively. (Experimented on VisDrone 2019 dataset).
From: End to end polysemantic cooperative mixed task trainer for UAV target detection
Backbone | AP | AP_50 | AP_75 | AP_S | AP_M | AP_L | |
|---|---|---|---|---|---|---|---|
Faster-RCNN | ResNet-50 | 23.79 | 36.2 | 27.21 | 15.02 | 26.87 | 30.55 |
PoT-50 | 28.43 | 51.57 | 31.85 | 19.66 | 31.51 | 45.92 | |
ResNet-101 | 25.13 | 38.72 | 28.85 | 16.36 | 28.21 | 33.07 | |
PoT-101 | 29.71 | 53.53 | 33.13 | 20.94 | 32.79 | 47.88 | |
Cascade-RCNN | ResNet-50 | 26.31 | 36.49 | 29.73 | 17.54 | 29.23 | 30.84 |
PoT-50 | 30.77 | 51.41 | 34.19 | 22 | 33.88 | 45.76 | |
ResNet-101 | 32.16 | 33.73 | 35.58 | 23.39 | 35.24 | 28.08 | |
PoT-101 | 33.44 | 54.4 | 36.85 | 24.67 | 36.52 | 48.75 |
