ECBAPose: Enhanced CNN Backbone with Attention for LiteWeight Human Pose Estimation
Jiahui Li, Deyuan Zhang, Xiangbin Shi · 2024
Human pose estimation has important applications in human-computer interaction in electromechanical automation. Existing human posture estimation models are designed to be more complex in order to obtain high accuracy. This paper focuses on lightweight network, using lightweight convolutional neural network as the backbone network, enhancing the global modeling capability of the model by introducing modified transformer module while ensuring the lightweight nature of the model, and further improving the performance of the network by incorporating a lightweight channel attention module. Experimental results on the COCO2017 human posture dataset show that compared with alphapose, it is able to reduce the number of parameters and computation by 30.16% and 34.18% respectively with comparable AP, this allows our human pose estimation network to be better deployed in electromechanical automation systems, reducing the computational cost of the system.