Human Pose Estimation Based on Lite HRNet with Coordinate Attention

Human pose estimation is an essential task in computer vision, applied in motion recognition, motion capture, augmented reality, etc. The emergence of Lite HRNet balances computational complexity with high precision, using channel attention mechanism instead of complex convolution while maintaining high resolution, thus possessing less computation and higher precision. The use of the channel attention mechanism results in the loss of the position information that is crucial to generating spatially selective attention maps. Accordingly, in this paper, we implement an improved pose estimation method based on the coordinate attention mechanism and Lite HRNet. Our method adds the coordinate information embedding to the original approach instead of expensive point-by-point convolutions. In this way, more information can be reserved in the case of less computation. The proposed method is evaluated on the COCO dataset, and the experiments show that this method can achieve a better balance in terms of accuracy of calculation quantities.