A Lightweight YOLOv5 Optimization of Coordinate Attention

Round 1

Reviewer 1 Report

In this paper, a lightweight network of the YOLOv5 is proposed to improve the detection speed with maintaining detection accuracy. This kind of network is built by reducing the number of network channels according to the size of the data-set and adding detection head and attention modules. It is an interesting thinking to design a lightweight object detection network. There are still some issues that need to be considered.

1.      Please indicate the meaning of horizontal and vertical coordinates in Figure 3.

2.      In Table 5, different models should be compared under the same data set.

3.      In comparison experiment, it is more convincing to use different data sets for comparison algorithms.

4.      I found several spelling mistakes throughout the paper. It is better to use grammar check software.

Author Response

Dear Professor:

Thank you for your precious comments and advice. Those comments are all valuable and very helpful for revising and improving our manuscript.

We have concerned these comments carefully and have made revision. Revised portion are marked in blue in the manuscript. The main adjustments and responses to the reviewer’s comments are as flowing:

 

Reviewer1’comments:

1. Please indicate the meaning of horizontal and vertical coordinates in Figure 3.

Response:

Thanks for your suggestion. We agree with the comment and analysis the result and meaning of horizontal with CA attention mechanism in different location.

(L319-L333)

 

2. In Table 5, different models should be compared under the same data set.

Response:

We agree with the comment and design 2 tables for VOC and Globalwheat2020 data set.

(Table 7,L405-L425)

 

3. In comparison experiment, it is more convincing to use different data sets for comparison algorithm

Response:

We are grateful for your suggestion. We have revise the experiment part, and built the result tables for comparison with other object detection models.( Table4,5,6,7)

 

4. I found several spelling mistakes throughout the paper. It is better to use grammar check software.

Response:

Thanks for your suggestion. We agree with the comment and check the grammar and spelling mistakes throughout the paper. 

 

Author Response File: Author Response.pdf

Reviewer 2 Report

Dear authors, this paper proposes a lightweight object detection method Based on lower channel optimization and coordinated attention. They propose to lighten the network by reducing the number of network channels according to the size of the data set and adding detection heads and attention modules. Articles related to computer vision have archival values, but before being considered for publication in this renowned journal, your work must be extensively revised.

1) The title is too long. Consider shorting it to less than ten words.
2) Please remove template instructions from your manuscript.
3) In the abstract, you state, "We reduce the network to 40% and 60% of 22 YOLOv5s". What were you able to reduce? Complexity? Please, elaborate on that.
4) Why are you considering YoLo's version 5 in this study? Please, state your reasons for not using, for example, YoLo's versions 6 or 7, which are more recent and efficient.
5) The introduction does not list the contributions of this study. Please, cite what the contributions of this work are.
6) The related work section is relatively poor. Please, consider rewiring that section with other works. The authors should describe and contrast works that are related to theirs. A table comparing and contrasting such works would be a significant contribution.
7) In figure 2, mention in the figure's caption what is shown in subfigures (a) and (b), respectively.
8) By looking at figure 2, it seems that your proposal is more complex than the original model. Explain why that is so.
9) The improvements proposed by the authors in section 3 are unclear. More figures should be added to elucidate the proposed modifications.
10) Figure 3 does not contain the name of the axis. What are you showing in that figure?
11) All tables should clearly indicate the entry presenting your model's metrics. It is somehow confusing and misleading.
12) Why don't you compare your results with YoLo's latest versions?
13) In the conclusion section, you should mention what your next moves regarding this work are, that is, what are the future research directions one should follow to improve on your work?
14) The number of cited works is relatively small. It should be improved by creating a related work section and extending the introduction.
15) Extensive and thorough proofreading is required since I've found several typos and grammar errors throughout my review.

Author Response

Dear Professor:

Thank you for your precious comments and advice. Those comments are all valuable and very helpful for revising and improving our manuscript.

We have concerned these comments carefully and have made revision. Revised portion are marked in blue in the manuscript. The main adjustments and responses to the reviewer’s comments are as flowing:

 

Reviewer2’comments:

1.The title is too long. Consider shorting it to less than ten words.

Response:

Thanks for your suggestion. We have changed as”A Lightweight YOLOv5 Optimization of Coordinate Attention”

(L2)

2.Please remove template instructions from your manuscript.
Response:

Thanks for your suggestion. We have revised.

 

3. In the abstract, you state, "We reduce the network to 40% and 60% of 22 YOLOv5s". What were you able to reduce? Complexity? Please, elaborate on that.
   Response:

We are grateful for your suggestion.We agree with the comment and revise the abstract.(L9-L19) 

 

4. Why are you considering YoLo's version 5 in this study? Please, state your reasons for not using, for example, YoLo's versions 6 or 7, which are more recent and efficient.
   Response:

We are grateful for your suggestion.We agree with the comment and analysis the comparison model. Because of the lightweight optimization, it is better to choose the light models.(L420-425)

 

5. The introduction does not list the contributions of this study. Please, cite what the contributions of this work are.
   Response:

We are grateful for your suggestion.We agree with the comment and revise the introduction.(L25-L65) 

 

6. The related work section is relatively poor. Please, consider rewiring that section with other works. The authors should describe and contrast works that are related to theirs. A table comparing and contrasting such works would be a significant contribution.
   Response:

We are grateful for your suggestion.We agree with the comment and revise the related work.(L67-L173) 

 

7. In figure 2, mention in the figure's caption what is shown in subfigures (a) and (b), respectively.
   Response:

We are grateful for your suggestion.We agree with the comment and revise the figure2.(L265) 

 

8. By looking at figure 2, it seems that your proposal is more complex than the original model. Explain why that is so.
   Response:

We are grateful for your suggestion.We agree with the comment and explain the reason.(L254-L262) 

 

9. The improvements proposed by the authors in section 3 are unclear. More figures should be added to elucidate the proposed modifications.
   Response:

We are grateful for your suggestion.We agree with the comment and revise the proposed modifications.(L236-L262) 

 

10. Figure 3 does not contain the name of the axis. What are you showing in that figure?
    Response:

We are grateful for your suggestion.We agree with the comment and revise the analysis of the figure3.(L319-L333) 

 

11. All tables should clearly indicate the entry presenting your model's metrics. It is somehow confusing and misleading.
    Response:

We are grateful for your suggestion.We agree with the comment and revise the table.(Table 5,6,7) 

 

12. Why don't you compare your results with YoLo's latest versions?
    Response:

We are grateful for your suggestion.We agree with the comment and give the reasons.(L418-L425) 

 

13. In the conclusion section, you should mention what your next moves regarding this work are, that is, what are the future research directions one should follow to improve on your work?
    Response:

We are grateful for your suggestion.We agree with the comment and revise the conclusion.(L4433-L446) 

 

14. The number of cited works is relatively small. It should be improved by creating a related work section and extending the introduction.
    Response:

We are grateful for your suggestion.We agree with the comment and revise the introduction.Also renew the reference(L67-L34，L465-L531) 

 

15. Extensive and thorough proofreading is required since I've found several typos and grammar errors throughout my review.

Response:

Thanks for your suggestion. We agree with the comment and check the grammar and spelling mistakes throughout the paper. 

 

 

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Accept in present form

Reviewer 2 Report

Dear Authors, all comments I left were appropriately addressed.