Table 4 The Average Precision (AP) of the baseline models.
From: HIT-UAV: A high-altitude infrared thermal dataset for Unmanned Aerial Vehicle-based object detection
Model | Dataset | Person AP (%) | Car AP (%) | Bicycle AP (%) | OtherVehicle AP (%) | mAP@0.50 (%) |
|---|---|---|---|---|---|---|
YOLOv4 | HIT-UAV | 89.88 (TP = 2370, FP = 346) | 92.64 (TP = 1241, FP = 166) | 86.48 (TP = 696, FP = 158) | 69.99 (TP = 26, FP = 8) | 84.75 (TP = 4333, FP = 678, FN = 447) |
YOLOv4-tiny | HIT-UAV | 16.86 (TP = 214, FP = 50) | 83.61 (TP = 1080, FP = 226) | 51.9 (TP = 398, FP = 182) | 49.17 (TP = 14, FP = 7) | 50.38 (TP = 1706, FP = 465, FN = 3074) |
Faster-RCNN | HIT-UAV | 75.5 | 95.6 | 86.4 | 46.8 | 76.8 |
SSD-512 | HIT-UAV | 85.6 | 96.3 | 86.0 | 74.4 | 85.6 |
YOLOv4 | COCO | \ | \ | \ | \ | 65.7 |
YOLOv4-tiny | COCO | \ | \ | \ | \ | 40.2 |
RRNet | VisDrone-2019 | \ | \ | \ | \ | 55.82 |
