Semantic–Structural Graph Convolutional Networks for Whole-Body Human Pose Estimation

Round 1

Reviewer 1 Report

The manuscript is devoted to a novel semantic-structural graph convolutional network (SSGCN) for whole-body human pose estimation tasks, which leverages the whole-body graph structure to analyze the semantics of the whole-body keypoints through a graph convolution network. At the same time, a novel heat map-based keypoint embedding module, which encodes the position information and feature information of the key points of the human body, is proposed.

The article's structure is missing at the end of the Introduction.

It is highly appreciated to share a link to the code.

The Discussion should cover the system's performance compared to a full extent.

The Conclusion can be extended focusing on the outcomes obtained (some more benefits based on the provided tests). 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors propose a novel semantic-structural GCN to accurately estimate whole-body pose. The paper is well-written with few issues that i describe below and it is of particular interest to the research community. The manuscript requires minor proofreading, especially after the Related Work section. Other comments include:

1) Please revise equations (1) and (2). They are not clear.

2) It is not clear how m_ki is computed in equation (3).

3) The Group representations in Figure 5 should be better clarified on what joints include.

4) In lines 184-185, L_i and L_fully are not clarified. L_1D, L_2D and L_overall are used instead.

5) The ablation study should appear before the experiments withe the state-of-the-art methods.

6) The lambda parameter in equation (11) has been defined empirically or experimentally? Why the value of 0.1?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf