Enhanced Cascaded Object Detection Network Based on Indirect Self-Attention Mechanism

Object detection is one of the most common and widely applied sub-tasks in the field of computer vision. In recent years, object detection has also made significant strides spurred by advancements in deep learning. However, the challenge of how to simultaneously enhance the network’s detection performance while maintaining generalization has remained unresolved. Thus, We proposed a two-stage cascaded 2D object detection network equipped with a self-attention mechanism. Our proposed indirect self-attention machanism enhances performance while reducing computational overhead. Experimental results demonstrate that our approach effectively improves model performance.