Correlating Edge with Parsing for Human Parsing

Round 1

Reviewer 1 Report

The presented article proposes a human analysis method based on MCEP network, which can promote analysis results by filtering background information while jointly learning edge analysis and human analysis. A large number of experiments on single and multiple human analysis datasets show that this method has major advantages.

The article is well written and organized. The implemented methods are correctly described. Research is important in the area and is of great interest for future applications.

Some suggestions for improving the text:

- care throughout the text with the hyphenation of words;

- Check the lack of white spaces between words throughout the text;

- In the tables, the values must all have the same number of decimal places;

- Sometimes capital letters appear in the middle of sentences. It is convenient to search the entire text in greater detail.

- In the abstract the sentence of the last 2 is repeated;

- The acronym LIP is not described in the text;

- Sections 2.1, 2.2, etc should not have an "s" before them;

- In section 2.1, 3.1, 3.3, the title starts with a lowercase letter;

- Variables or mathematical formulas described in the middle of the text should be written in italics to be more noticeable;

- Sometimes appears 2D other times two-dimensional. They must choose one of the forms throughout the text;

- On page 6, right at the beginning the sentence was broken;

- In section 4 the text is not justified;

- In equations (5) and (6) remove the word "the" at the beginning;

- Correct in the denominator of equation (5) Pji by Pij;

- Replace "*" by "x" in the equation on page 7;

- Define miou in the same way throughout the text (miou (page 7), mIoU (table 1), MIoU (table2));

- In the conclusions it is missing to present which are the projects/interests for the future in relation to the developed method.

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The idea is very similar with the paper [31] Correlating edge, pose with parsing. Therefore, the novelty seems very limited. A detailed discussion of the difference between the two works is definitely necessary. 

The philosophy of the SMP module is not clearly presented. It is not enough to plainly give the technical details without describing the motivation of the designs.

The related work is far from comprehensive. Many recent human parsing and semantic segmentation works should be included in Sec. 2, for example, Differentiable Multi-Granularity Human Parsing-IEEE TPAMI, Hierarchical human semantic parsing with comprehensive part-relation modeling and  Rethinking Semantic Segmentation: A Prototype View-CVPR.

Limitation analysis or failure cases should be presented. 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The revision has addresses my concerns.