Deep Learning and Vision-Based Early Drowning Detection

Round 1

Reviewer 1 Report

The topic is relevant, and the use case is an important problem to study. However, some major concerns about the paper.

Interesting use case, not recurrent but important problem to study

1.     the authors test several good SOTA models, but didn’t justify the choice of each one (Why exactly those 5 and not others?)

2.     In L46, the authors mentioned "We developed" while they transferred learning from pre-trained models.

3.     Total Dataset used 148 images extremely low compared to the ones used by related works. Unfortunately, it’s not enough data to learn o deep learning model. I think Data augmentation helped to accurately extract the features, but most drowning (based on presented samples) contain high hands, maybe there are other drowning positions to detect. how many images are used after a data increase?

4.     the comparison made in table 8 is unfair because on the one hand you are comparing detection work with classification work and generally the IoU or M average precision metrics are less good than the accuracy for classification. On the other hand, you compare your work to other works that use enough data while your work uses only 148 images. I invite you to redo this comparison by using several images comparable with that used in other works.

 

5.     Some phrases structure (L138 : comparable to …) , (A CNN with 50 layers is called ResNet50 ?)

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

§  The paper's premise interesting and important.

§  The methods used appropriate.

§  The data support the conclusions.

§  The research is publishable in principle and deserves a carefully detailed reading for the reader to find scientific interest and enjoyment.

§  The title properly reflects the subject of the paper.

§  The abstract provides an accessible summary of the paper.

§  The keywords accurately reflect the content.

§  The paper an appropriate length.

§  The key messages short, accurate and clear.

§  The text’s meaning is clear.

§  A well-written the introduction.

§  Sets out the argument ....

§  Summarizes recent research related to the topic ...

§  Highlights gaps in current understanding or conflicts in current knowledge ...

§  Establishes the originality of the research aims by demonstrating the need for investigations in the topic area.

§  Original and topicality can only be established in the light of recent authoritative research.

§  This research has enough data points to make sure the data are reliable.

§  The results seem plausible, the trends you can see support the paper's discussion and conclusions, There are sufficient data.

§  The references relevant, recent, and readily retrievable.

§  The author has a deep understanding of the paper's content.

§  The research has most interesting data.

§  Abstract highlight the important findings of the study.

§  The authors presenting findings that challenge current thinking.

§  The evidence they present strong enough to prove their case.

§  The correct references cited.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

General thoughts

The article raises very important issues related to increasing swimming safety, which is why its publication is purposeful, and the proposed solutions are very interesting, original, and possible to be implemented in many facilities, especially swimming pools.

The article should consist of generally agreed parts, such as:

Introduction

Material and research methods

Results

Discussion

Conclusions

I suggest that the authors make appropriate corrections to their text, clearly separating individual parts of the article, placing the relevant content in them. Particularly important part are Research material and Methods and Discussion. Also, in the case of sub-points, pay attention to their numbering.

It is also worth verifying the content of the Conclusions  so that they provide an answer as to how the main objective of the research has been achieved and how the achieved results can be implemented and by whom.

Detailed notes

Change in line 20 downing - to drowning

Verify numbering 4.1. in line 93, 116

In table 7, in the 3rd row, there is number 97,9166, other % results are with 2 decimal places - should be standardized, corrected.

Similarly, you should correct this number in line 304.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

 For Response 4:

The reference [11] ‘An Improved Detection Method of Human Target at Sea Based on Yolov3’ is adressing a detection issue with bounding boxes not classification, to differentiate sea and land persons which does not determine if the person is swimming or drowning, so neither the task or the use case is comparable to the study of this article.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf