Table 5.
Comparison of different models on the CCTSDB2021.
| Model | Precision | Recall | F1 | mAP@.5% | Params(M) |
|---|---|---|---|---|---|
| Fast RCNN7 | 84.4 | 54.9 | 66.5 | 56.5 | 143.7 |
| Libra RCNN43 | 83.7 | 60.0 | 70.0 | 61.4 | – |
| Dynamic RCNN36 | 87.0 | 58.3 | 69.8 | 60.0 | – |
| Sparse RCNN37 | 94.1 | 52.6 | 67.6 | 59.7 | – |
| SSD30 | 86.5 | 27.4 | 42.0 | 49.2 | – |
| YOLOv311 | 84.6 | 42.7 | 56.8 | 50.0 | – |
| YOLOv444 | 76.2 | 52.5 | 62.2 | 51.7 | – |
| YOLOv7-tiny33 | 89.8 | 74.9 | 81.7 | 80.9 | 6.2 |
| YOLOv5-s | 91.2 | 76.8 | 83.3 | 81.6 | 7.2 |
| SC-YOLO38 | 93.8 | 76.8 | 84.5 | 84.3 | 6.1 |
| FocusDet(ours) | 92.2 | 76.9 | 83.9 | 87.8 | 9.26 |
The bolded performance is the best one.