This repository provides training and inference code for human pose estimation based on the lightweight OpenPose approach. It detects human skeletons consisting of 18 keypoints (ears, eyes, nose, joints) and their connections in images. The model is optimized for real-time CPU inference with minimal accuracy loss compared to the original OpenPose. The implementation achieves 40% AP on the COCO 2017 Keypoint Detection validation set using single-scale inference.