A Human Body Part Semantic Segmentation Enabled Parsing for Human Pose Estimation

ADITI VERMA, Vivek Tiwari, Mayank Lovanshi, Rahul Shrivastava

Abstract:

Human Body Part Semantic Segmentation and Human Pose estimation are considered to be essential for understanding human
behaviours. Both of these tasks are correlated with each other. Employing them together in a unified framework to perform two distinct Human Centric Visual Analysis tasks simultaneously allows
benefiting from each other. Taking advantage of the correlation between Human Body Part Semantic Segmentation and Human Pose
Estimation, this paper proposes a unified framework that explores
efficient context modelling. The framework simultaneously predicts
the human body part semantic segmentation and pose estimation
with high-quality results. The results extracted from the segmentation are used to predict the pose estimation task. An experimental
analysis of the proposed framework is done on the benchmark LIP
Dataset. The analysis of the results shows that the proposed framework outperforms the state-of-the-art by 7.3% when evaluated on
mean IoU. Moreover, Mean Accuracy, Pixel Accuracy and PCKh
are the other metrics used for the evaluation of the framework.