ALFPN: Adaptive Learning Feature Pyramid Network for Small Object Detection

Figure 2

The overall architecture of ALFPN. It mainly includes two parts: (1) the AFI module devote to find a better adaptive fusion way and change the transfer coefficients at different levels; (2) the CAS module is a feature alignment supervision module, which aims to further weaken the inconsistency of different levels; (3) the Att module represents channel attention, which can further make better use of features at different scales.