Table 2.
The ablation study on the BCTSDB dataset.
| Methods | Parameters (M) | FLOPs (G) | AP (%) | AP50 (%) | AP75 (%) |
|---|---|---|---|---|---|
| RetinaNet baseline | 37.74 | 95.66 | 59.7 | 89.4 | 71.2 |
| +ClassDecoder | 35.03 (−2.71) | 70.30 (−25.36) | 61.6 (+3.7) | 91.8 (+2.4) | 75.8 (+4.6) |
| +Global Extract Encoder | 36.95 (+1.92) | 90.45 (+20.15) | 63.4 (+3.4) | 93.9 (+2.1) | 80.6 (+4.8) |
| +Attention | 37.45 (+0.5) | 90.65 (+0.2) | 65.2 (+3.1) | 95.1 (+1.2) | 84.2 (+3.6) |
| +Anchor-free | 37.31 (−0.14) | 89.95 (−0.7) | 65.8 (+2.1) | 95.7 (+0.6) | 87.4 (+3.2) |
| +Data augmentation | 37.31 (+0) | 89.95 (+0) | 76.1 (+4.1) | 97.6 (+1.9) | 91.4 (+4.0) |