Saliency Guidance and Expansion Suppression on PuzzleCAM for Weakly Supervised Semantic Segmentation

Round 1

Reviewer 1 Report

1-     Keywords are written with an uppercase as the first letter.

2-     Correct this sentence by removing the Extra square bracket as follows: “The basic annotation methods that 68 are commonly seen in weakly supervised semantic analysis tasks are bounding boxes [9– 69 11], scribbles [12,13], and image-level annotation [5,6,14–26].”.

3-     Figure 1's title is ambiguous. Please select a title that reflects the great resolution of the structure provided.

4-     It is advantageous to rewrite the section on related works as a single section rather than the parts into which it was divided, and to add weaknesses to each study that was mentioned, even if in total, to the most important work mentioned. Instead, you can include a table that shows the study's strengths and weaknesses.

5-     Adopting a single style in writing the main titles of the paper.

6-     Use a single template to refer to figures, such as fig 1 or figure 1.

7-     Figure 2 is not clear.

8-     Refer to the equations mentioned in the study paragraphs.

 

9-     Writing the proposed work algorithms in a more organized manner and best discussing the results to reflect the level of research presented first and the points of future development second.

Author Response

Please find the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

For weakly supervised semantic segmentation tasks, this study adopted a supression module to suppress strong points of interests and resolve the problem of miclassifying scenes as objects, and designed the saliency map guidance module to help obtain background. The segmentation mIoU results outperform state-of-the-arts. The article was well organized and written. Accept is suugested.

Author Response

Thanks for your positive comment and acceptance. Please check the response sheet for other reviewers. 

Author Response File: Author Response.pdf