Skip to main content
. 2022 Jun 26;22(13):4833. doi: 10.3390/s22134833

Table 4.

The performance of attentional mechanism in different backbone networks on BCTSDB dataset.

Backbone Params. FLOPs Head Attention AP (%)
MobileNetv3 × 1.0 5.4 M 220 M RetinaNet w/o. 51.2
ResNet50 25 M 3.8 G 59.7
ResNet101 46.3 M 7.6 G 64.8
MobileNetv3 × 1.0 5.9 M 231 M RetinaNet w/. 54.1
ResNet50 25 M 3.8 G 62.5
ResNet101 46.3 M 7.6 G 66.3