Table 2 Results of ablation experiments on lightweight metrics.
From: Lightweight aerial image object detection algorithm based on improved YOLOv5s
Algorithm | Parameters (MB) | FLOPs (GB) | Speed (ms) | FPS | mAP@0.5 (%) |
|---|---|---|---|---|---|
YOLOv5n | 1.8 | 4.2 | 5.9 | 64.5 | 24.2 |
YOLOv5s | 7.0 | 15.9 | 10.5 | 57.8 | 32.1 |
YOLOv5m | 20.9 | 48.0 | 12.1 | 55.6 | 35.9 |
YOLOv5l | 46.2 | 107.8 | 19.6 | 40.5 | 37.7 |
YOLOv3 | 61.5 | 154.7 | 22.5 | 36.4 | 39.0 |
YOLOv5s + DFM-CPFN | 5.4 | 17.3 | 12.4 | 54.3 | 37.0 |
YOLOv5s + VB | 14.9 | 34.0 | 12.5 | 52.1 | 34.3 |
YOLOv5s + C3SFN | 5.2 | 12.9 | 10.6 | 56.8 | 32.1 |
LAI-YOLOv5s | 6.3 | 29.0 | 13.2 | 51.3 | 40.4 |
