A New Workflow for Instance Segmentation of Fish with YOLO
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsJMSE (ISSN 2077-1312); Manuscript jmse-3060912. Authors Zhang, Wang.
Dear Authors, your work is interesting, but your report needs some updating. My remarks below are meant to help you to improve the paper. First I give two important remarks and below these, I have a number of small remarks/textual remarks.
Your paper would profit from the visualization of your new Workflow announced in the title. It is now introduced very briefly in text Lines 194 – 198, while it is an important feature of your paper. I recommend to have a visual display of your Workflow, for easy reference for the reader. In Line 280 you say that your Workflow method is even self-iterative, so a diagram of this workflow is barely needed. Did you happen to rediscover the adversarial competing network method?
Also, your paper announced in the title: using and testing Yolo. You forgot to say what it is in the text. Yolo has to be explained YOLO (You Only Look Once) is a real-time object detection algorithm developed by Redmon and Farhadi in 2015. It is a single-stage object detector that uses a convolutional neural network (CNN) to predict the bounding boxes and class probabilities of objects in input images. You do not explain why you did not use Yolo9000: https://openaccess.thecvf.com/content_cvpr_2017/papers/Redmon_YOLO9000_Better_Faster_CVPR_2017_paper.pdf .
I think your paper should be published, so below I give additional remarks about your text, to help improving it:
Suggestions for improvements
1. In your Title you use ‘instance’. This seems unclear to me. I come back to it below in item 6. Should you use, in the title, the common word (for the reader): ‘object’?
2. The word fish in your chosen title of the paper is unclear. You do not segment fish in the flesh. It should read ‘images of fish’, or ‘fish images’.
3. The Abstract has too much of the style of an introduction. You should summarize in it the content + output of the paper. Please omit an explanation of the field as if it is your introduction. Lines 8 – 13 should move to chapter 1, this could also help in explaining Yolo, about which I come back below. Your Abstract could better start with ‘In this study, we applied …’.
4. Line 12, ‘ushered in’ should read ‘entered’ I think.
5. Lines 30, 33, 'annotation' is a word for human remarks in texts and for types of writing. I think that you should use the word 'labeling' everywhere to replace ‘annotation’. You already use labeling nine times, for instance in Line 287 and then you shift in Line 305 again erroneously to ‘annotation’.
6. Line 55, I do not understand ‘separate’ here. ‘but also to separate different object instances’. Do you mean ‘distinguish’? To distinguish different multiple objects with the same label?
7. Line 65, 'some research works' needs a small list of references here.
8. Line 77, 'base network' seems incorrect. Do you mean 'basic network'?
9. Line 82, wrong sentence. 'with a density-based regression approach was integrated’ should read ‘with an integrated density-based regression approach’.
10. Line 83, you put RMP-Net between quotes and elsewhere you do not use quotes: PFFS, PIF. Please be consistent.
11. Line 85-86, 'complicated terrain on the seabed' should read 'complicated seabed terrain'.
12. Line 88, ‘of acoustic waves and the ambient noise’ is not clear. Do you mean that wireless communication only is capable of transmission of slowly oscillating waves?
13. Line 113, what does ‘indicators of the same form’ mean? Is it a similar number format?
14. Lines 119-123, should be put before Line 113, the performance indicators.
15. Line 170 -182, tells about your new ‘workflow for fish recognition’. Could you relate this for the reader to changes in your method in Lines 194 – 198?
16. Lines 320 – 330, to me it seems that you forgot to explicate how your workflow defined in Lines 194 – 198 changes by your achieved optimization?
17. Line 126-127, do not explain why you selected Yolo. Is a Genetic Adversarial Network better? Might Yolo be a bad choice? The adversarial technique is to let two NNs compete against one another. You can find an explanation at https://en.wikipedia.org/wiki/Generative_adversarial_network .
18. Line 201-202, the sentence is unclear ‘many standards usually with variations in pixel accuracy’. Should it read ‘standards vary in pixel accuracy’?
19. Line 203, ‘were’ should read ‘are’.
20. Line 208, the remark ‘though they are both the sum of False Positive and False Negative.’ Is unclear. Could you explain how ‘False Negative number and False Positive numbers’ are both the sum of themselves?
21. Line 224, ‘metrics’ should read ‘metric’.
22. Lines 227-233, use the acronyms TP and FN. These should be introduced before in Lines 207-208.
23. Line 229, ‘under’ should read ‘of’.
24. Line 242-247, figures 2 and 3 are too small and therefore unreadable. Please enlarge the fonts.
25. Line 232-233, the IoU intersections are 3D and the AP threshold is 2D. Is this correct?
26. Line 269, figure 5 is missing.
27. Line 278, ‘Due to the lack of a large amount of available training data’ should read ‘Due to insufficient training data’.
28. Lines 310-311, I fail to understand your conclusive remark here. Could you make it clearer?
Comments on the Quality of English Languagesee many remarks above.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe manuscript proposes a new workflow for segmentation of fish datasets, by using YOLO-related methods. The results obtained show that their fish identification and segmentation workflow can achieve better performance in the segmentation task, and could effectively improve the efficiency of fish survey.
The manuscript is clear, and relevant for the field. The discussion addresses sufficiently the main research question posed and is consistent with the contributions presented. A large part of the cited references are mostly recent publications and relevant to the subject.
Overall, the topic is interesting, and the quality of the current version of the paper is about average.
Just a few comments.
In Introduction, related work and relevant methods were investigated, however, a comparison is missing against the authors’ proposed method and work (e.g., any advantages/contributions of the current work).
On line 153, the font size needs correction.
Figure 1B is not very clear.
I would prefer to have more details on YOLOv5 and YOLOv8 methods, particularly for those readers that are not so familiar.
In addition, it is not clear the software tool/platform upon which this YOLO methods were implemented and used.
Conclusion seems to be rather short, but anyway, Discussion is indeed lengthy.
In Conclusion, on the phrase “… than the original methods ….”, it will be more comprehensive to state explicitly some of these original methods. In addition, perhaps a brief summary of the paper’s contributions could be appropriate in Conclusion.
Comments on the Quality of English LanguageMinor editing.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsNo further comments.
