Efficient and Lightweight Human Pose Estimation Algorithm
Jie Zhang, Mai Yan, Haiyong Zheng · 2023
This study examines a straightforward and effective human posture estimation technique based on a SimpleBaseline network. A simple and effective network model is created, and a multi-feature fusion module is implemented on the foundation of SimpleBaseline to fuse high-resolution feature maps and low-resolution feature maps, in an effort to address the issue that the present human pose estimation task model is too complex. The effective Shuffle module's functions include replacing the original network's basic module, reducing unnecessary parameters, and enhancing model inference. A coordinate attention mechanism is presented to embed location information into channel attention to increase the accuracy of the lightweight network and make the network more lightweight and efficient in order to address the issue of prediction accuracy decline after model light weighting.