A Dense Small Object Detection Algorithm Based on a Global Normalization Attention Mechanism
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors- In general, the manuscript focuses on an interesting topic of study adopting deep learning framework, the organization of the paper is good and acceptable. Theoretical background, basics, equations are well explained. Comparative results presented validate the proposed framework. However,, I have the following comments:
- The mauscript need extensive proof reading as there are many typos (e.g., line 15 space needed before "(GNAM)" to be consistent either you insert space or not across all the abbreviations in the manuscript). Similarly the space before the citation number and text (e.g., [5] on line 42), line 72 should be new paragraph?
- You should compare the results of your work to the original works by Liu et al. "NAM: Normalization-based Attention Module"
- There needs to be more details about the experiments being conducted. How was the classification conducted and how was the data split?
- Table 2, does the performance gain versus overhead justify the new method? What was the backbone network used in the other yolo methods?
- The table of abbreviations is needed as per the template.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe manuscript presents the GNYL (Global Normalization Attention You Only Look Once) technique for detecting dense tiny objects in the air based on the global normalization attention mechanism.
Some major points should be addressed before final acceptance:
The title needs to be reformulated to be more concise and informative.
No sufficient background is built; therefore, the reader cannot infer the motive or even the contribution.
The contribution statement should be derived after the revised background with justification from the literature for selecting the proposed solution technique.
The authors need to refer to deep learning advances made in Residual Neural Networks for Origin–Destination Trip Matrix Estimation from Traffic Sensor Information and Variance-based global sensitivity analysis for rear-end crash investigation using deep learning.
In addition to the MAP, the authors should report other goodness of fitness measures such as confusion matrix metrics, AUC, etc.
The achievement accuracy should be justified with an illustration of the selected identification components for the different identified objects.
Comments on the Quality of English LanguageModerate editing of English language required
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsThe authors did not respond to my comments accurately. The response letter should follow the common format of responding to reviewers' comments. Mention each comment of the review report with its arguments accompanied by referencing the action done in the manuscript to cure the raised concern.
In the current case, the article could not be accepted for publication.
Comments on the Quality of English LanguageModerate editing of English language required
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 3
Reviewer 2 Report
Comments and Suggestions for AuthorsNo further comments.
Comments on the Quality of English LanguageNone.
