Research Article
Improved YOLOv4 for Pedestrian Detection and Counting in UAV Images
Table 3
Comparisons with other existing methods in test set.
| Models | Backbone | AP50 | MAE | MSE | FPS |
| YOLO v4 (baseline) | CSPDarknet-53 | 17.56 | 11.36 | 19.90 | 26 | CenterNet [23] | ResNet50 | 17.67 | 13.84 | 30.38 | 22 | YOLOX [24] | Darknet-53 | 22.68 | 14.13 | 28.64 | 14 | RFBNet [25] | VGG | 17.99 | 9.58 | 19.27 | 32 | YOLO-CC (Human body detection) | CSPDarknet-34 | 39.32 | 7.29 | 12.38 | 34 | YOLO-CC (density map) | ā | 11.41 | 17.25 |
|
|