Table 3 The experimental results of neck network improvement.
From: YOLOFM: an improved fire and smoke object detection algorithm based on YOLOv5n
Model | Precision | Recall | mAP50 | mAP50-95 | FPS | Params (MB) | GFLOPs(G) |
|---|---|---|---|---|---|---|---|
Conv+ nn.Upsample+ C3(YOLOv5n) | 91.8 | 90.9 | 95.3 | 66.8 | 80.90 | 6.72 | 4.1 |
AsymptoticFPN | 91.0 | 90.2 | 94.6 | 64.1 | 51.63 | 5.04 | 3.4 |
Conv+ Transpose + QARepNeXt | 93.4 | 90.6 | 95.8 | 69.9 | 50.60 | 8.64 | 5.6 |
SimConv+ Transpose + QARepNeXt | 92.8 | 91.5 | 95.8 | 69.7 | 54.38 | 8.64 | 5.6 |
GhostConv+ Transpose + QARepNeXt | 92.6 | 91.0 | 95.6 | 69.6 | 49.04 | 8.64 | 5.6 |
SimConv + QARepVGGB+ Transpose+ QARepNeXt | 91.6 | 92.0 | 95.8 | 68.5 | 65.05 | 8.66 | 5.6 |
GhostConv + QARepVGGB+ Transpose + QARepNeXt | 92.4 | 92.0 | 95.8 | 69.9 | 57.09 | 8.63 | 5.6 |
