Table 5 Ablation study results of YOLOv8 on the VisDrone2019 dataset.
From: LPAE-YOLOv8: lightweight aerial small object detection via LSE-Head and adaptive attention
Baseline(YOLOv8) | A | B | C | D | P | R | mAP(0.5)(%) | mAP(0.5:0.95) (%) | The number of model parameters (×106) | GFLOPs (×109) | FPS (RTX 4090D) | best.pt |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
√ | 42.2% | 30.6% | 30.6% | 17.7% | 3.01 | 8.1 | 170 | 6.2 MB | ||||
√ | √ | 43.9% | 34.1% | 34.5% | 20.5% | 2.01 | 11.6 | 165 | 4.3 MB | |||
√ | √ | 40.1% | 31.7% | 30.6% | 17.5% | 2.76 | 7.4 | 156 | 5.8 MB | |||
√ | √ | 44.5% | 33.6% | 33.5% | 19.5% | 3.02 | 9.7 | 144 | 6.3 MB | |||
√ | √ | 42.9% | 31.2% | 31.0% | 17.6% | 3.01 | 8.1 | 173 | 6.2 MB | |||
√ | √ | √ | 43.2% | 34.2% | 33.7% | 19.8% | 1.96 | 11 | 147 | 4.3 MB | ||
√ | √ | √ | √ | 44.3% | 35.5% | 35.6% | 21.0% | 1.97 | 12.8 | 130 | 4.3 MB | |
√ | √ | √ | √ | √ | 45.8% | 35.9% | 35.8% | 21.3% | 1.97 | 12.8 | 130 | 4.8MB |
