Figure 2.

The architecture of our DA-SHT Dense 3DCE Network, using our Novel Dense 3DCE R-FCN as the backbone framework. The training of the regressor and the detector are consecutive steps. The regressor is trained first, and then the detector is trained.