Learning Delicate Pixel-Level Representations for Bottom-Up Human Pose Estimation | IEEE Conference Publication | IEEE Xplore