YOLOSA: Object detection based on 2D local feature superimposed self-attention

Weisheng Li, Lin Huang

We analyzed the network structure of real-time object detection models and found that the features in the feature concatenation stage are very rich. Applying an attention module he…