Improve the performance of CenterNet through hybrid attention mechanism CBAM

Tianyang Li, Tibing Zhang, Jian Wang, Zhongjie Wang, Ting He, Yufeng Zhang

Abstract: The attention mechanism is one of the research hotspots in deep learning. The attention mechanism in computer vision aims to make the network ignore irrelevant information and pay more attention to the practical knowledge of the critical point, which has the advantages of fewer parameters, fast calculation, and sound effects. In this paper, we introduce the mixed attention mechanism (CBAM) into the CenterNet object detection model to improve the performance of network detection on small data sets. At the same time, we used the ReLU6 activation function in the model to ensure that the model could be used in low-precision equipment. We use four small data sets extracted from COCO2017 to verify the model's validity, and the experimental results show that our improved model has better detection performance than the original model on small data sets.