Unsupervised Few Shot Key Frame Extraction for Cow Teat Videos
Round 1
Reviewer 1 Report
- Some references are old. Please update the manuscript with the latest articles.
- Lines 414-424 on page 16, many L1 and L2 formatting inconsistencies, please correct them. Also, there are also many inconsistencies between the L1 and L2 formats in the full text. Please correct them all.
- This article proposes a novel unsupervised few-shot learning model which extracts key frames from large (∼21,000 frames) video streams. The existing few-shot learning methods can in the following: metric learning; meta-learning; transfer learning; and data augmentation. This article uses the metric learning method, which learns metrics/similarity of few-shot samples through deep networks. It is best to describe the existing few-shot learning in the related work section.
Suggested reference:
Wei, R., & Mahmood, A. (2021). Optimizing Few-Shot Learning Based on Variational Autoencoders. Entropy, 23(11), 1390.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The manuscript presents a methodology for identifying the key frames from a large video stream to be extracted in order to perform analysis on the health condition of cow teats.
This document is very well written, resulting in a text that is easy to read and to understand. It is also well organized, and it includes a Simple Summary, an Abstract, a first Section containing an useful Introduction, a second Section on Related work followed, a third Section on the Methodology, a fourth Section on Experiments, a fifth Section showing the Discussion, a sixth Section including the Conclusions of the work, and finally an Appendix and the References used.
As it is, In my opinion the manuscript proposal would be of medium value, offering an interesting contribution to identification of the necessary video frames to be extracted for monitoring the cow teats' health condition.
Aiming to improve the quality of the work and without any intent to underrate neither its accuracy nor its contributions, I would like to make the following suggestions:
1. Lines 26 and 36: Consider replacing '1000s' with '1,000s'.
2. Line 35: Consider replacing " cow's " with " cows' ".
3. Figure 1: Do the two cameras shown on the left correspond to the "video camera" shown in the mid-bottom of the image?
4. Figure 2: This figure seems too simplistic in my opinion, and does not provide enough value to justify its incorporation in the document.
5. Figure 4 label: Please check the need for the 'Best view in color' comment at its end.
6. Line 171: Correct the 'the the base' inclusion.
7. Line 188: Modify 'KF image' to 'KF images'.
8. Figure 6 label: Replace 'triangles' with 'rectangles'.
9. Line 230: Change '(Fig. 6a)' to 'Fig. 6a'.
10. Line 232+2: Replace 'function,' with 'function:'.
11. Line 239: Please provide key specifications on the camera used.
12. Line 244: Please extend the description on the lighting used in the experiments.
13. Line 247: Change 'cow' by 'the cow'.
14. Line 252: Change '500 frame differences' with '500-frame differences'.
15. Line 255: Change 'brief' with 'short'.
16. Line 272: Please provide key specifications on the graphic card used.
17. Line 357: Consider changing 'candidates followed' to 'candidates, followed'.
18. In the Conclusions section, some attention to future works could be of value. Can the process be performed in real time? Which changes would be necessary in order to make this possible?
19. Figure A4 label: Correct 'SUFR' to 'SURF'.
20. Line393+4: Change '3200' to '3,200'.
21. Line 399: Please clarify the meaning of 'features are mixed together that we cannot find a clear pattern'.
22. In the reference sections, please review the use of bold type for some of the year references.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors wrote in Simple Summary and Abstract what they did for the research. These are very well-formulated elements of research methodology, although this scope of considerations could be called a methodology. In the mentioned parts of the article, however, I clearly miss an unambiguously formulated research goal, namely the sentence: "The aim of the research / study was ...". Even if the authors proposed a new method of monitoring the health of dairy cows, it could be written that ... The aim of the research / study was to propose (and verify?) A new method of monitoring ...
In general, when presenting the purpose of the research / study, I suggest that the Authors write what was the cognitive (scientific) goal and what was the utilitarian (useful) goal. Before presenting the purpose (s) of the research / study, it would also be worth formulating the research problem and the gap in the current state of knowledge that inspired the research. In the Introduction, the authors provided a lot of information on the basis of which a research problem can be formulated. The term problem was even used in line 71, only this research problem needs to be formulated in the form of one or two sentences pointing to the research problem.
I think line 6 should say "dairy cow's ..." instead of "diary cow's ...".
Why on line 98 the Authors included citation [10,10]? Was the citation written as [10] not enough?
I do not know how to interpret the position of the teat surface [130…] mentioned in footnote (2) on page 8. In which units is the considered parameter given? It would be worth supplementing the information at this point.
Has the camera lens been systematically cleaned during measurements in the milking parlor because of the risk of contamination with animal faeces or other contamination. It is worth mentioning in the article as a possible element disturbing the taking of pictures / measurements.
I think it would be worth giving more details about the carousel milking parlor where the research was carried out. At what speed was the platform with 60 milking stalls rotating? Did the speed of rotation of the milking parlor platform affect the accuracy of the camera measurements? What was the productivity of the milking parlor, i.e. how many cows were milked per hour? What was the milking capacity expressed in the number of cows per milker and per man-hour? How many milkers worked in the milking parlor? How long did it take to milk the 1,600 cows mentioned in the article? How many people were additionally supporting the herd during milking, i.e. how many people led the cows to the waiting (holding) area? What was the milking parlor's capacity in liters (kilograms) per hour and man-hour? It is also worth writing if the milking of the cows was carried out in accordance with the principles that determine the high-quality handling of cows and the quality of the obtained milk. Were dipping and pre-milking used? What was used to clean the teats before milking: paper towels or individual material towels? Were the teats of the cows covered with a chemical preservative at the end of each cow's milking and recordings? Was it applied to the surface of the teats by hand (using a dipping canister) or with a mechanical sprayer (manually or automatically operated)? I think that providing this information will significantly increase the value of the data supporting the experiment, characterizing the place of research. I think that a large part of the information I asked for would help a lot in meeting the condition of repeatability of the experiment carried out on real data.
Regardless of the information and parameters regarding the operation of the milking parlor, in the part describing the research material, it would also be worth providing more information about the herd of milked cows. The authors only provided information about the breed of cows (Holstein breed) in the article, but in my opinion, it would also be worth providing information on the DIM of cows, age, and the percentage of cows in each lactation. Or maybe it would be worth adding information about cow problems with teats? How many cows were there with these problems and what were the problems?
In conclusion, it would be worth pointing to the limitations of the research study carried out, primarily regarding the minimum number of cows in the herd, at which it is profitable to apply the proposed model approach. Is there an upper limit to the dairy herd size to which the proposed test method is applicable? The authors have already written about the limitations of the use of the model in the Discussion chapter, but I was thinking about developing the issue of possible other limitations.
Taking into account the classic approach to the discussion of research results, it could be expected to confront the results of own research with the literature. The authors developed a new research method / model, so it may be difficult to confront with other publications. Therefore, perhaps it would be worth rewriting the title of the chapter Discussion to the following wording: Discussion of the results of own research and the limitations of the model.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
