Serialized keypoint estimation using body part segmentation

Human pose estimation is a topic of interest in the field of computer vision. Once we precisely predict where the human body is, we can further use that information to perform high-level actions such as action recognition or behavior prediction. In this paper, we focus on finding keypoints of human along with body part segmentations that surround keypoints. After roughly finding body part segmentations, we hope to refine accurate keypoint from it. We used Stacked Hourglass model, which is often used in pose estimation problems, as the backbone and further attached model to predict body part segmentation. We also tested several networks to reduce unwanted side effect that occurs when using keypoints and body part segmentation together.