Semi-Supervised Crowd Counting Method Based on Attention Mechanism

Aiming at low counting accuracy caused by less labeled data in the field of crowd counting, we propose a new semi-supervised counting method, which introduces mixed attention and information enhancement method. Firstly, the spatial attention mechanism and channel attention mechanism in mixed attention are added to the network structure, so that the model can adaptively learn the attention weights of different regions, focus on the foreground regions with more information, and ignore the unimportant background regions; Secondly, we further improve the expression ability of features and the generalization ability of the model through the GRN layer and Drop out layer, enhancing dimension processing. Extensive experiments on three challenging crowd counting datasets show that our method largely achieves comparable performance to current state of the art semi-supervised counting methods.