Vision-Based Quadruped Pose Estimation and Gait Parameter Extraction Method
Round 1
Reviewer 1 Report
- The abstract section needs to be improved and concise.
- The information from the literature has not been appropriately and sufficiently acknowledged in the introduction section.
- Please explain how to set the confidence threshold of the detection box. (Section 2.1, Point no-1)
- RFB (Receptive Field Block) should be defined when used for the first time in the abstract section.
- The proposed idea developed in section 2 should be elaborated in a logical sequence.
- How the numerical evaluation is done and results are obtained? Please explain in detail.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The authors propose a method for processing images and frame sequences with quadrupeds for their pose and gait estimation. The authors say that it is necessary in many researches including creation of quadrupedal robots. I have several concerns:
1) The authors say that they present a method for pose estimation, but they neither describe it nor present the experimental results for it. So, the authors should either add this method to their article or rework the article.
2) When the authors describe their experiments firstly they say that they have a dataset with 50 animal species, then they say that they use 28 animal categories, and finally they claim that they tested 10 different kinds of animals. It is completely unclear why the number was reduced from 50 to 10. The authors should explain it.
3) In Tables 3 and 4 the authors say that their network has 63.972 parameters, but then in text we find that, in fact, it is 63,972,000 parameters. The authors should correct the numbers in Tables 3 and 4.
4) In Table 5, the authors present their results compared with "Manual method", but then in the text say that they performed a comparison with "artificial statistical method". The authors should present the full comparison in Table 5. Also, it is unclear how many animals of each species were used in this experiment. As a result, it is unclear how important is the shown difference in average error. The authors should describe their experiment in more detail.
5) In Section 3.2.4 the authors say that they use "the above method", but no method for gait duty cycle extraction was presented.The authors should add the description of the method. In Table 6, the authors compare their method with "manual calculation", but in the text they write about "artificial statistical method". The authors should describe the comparison that they performed.
6) In Figure 2, the authors present the architecture of their RFB-HRNet, and in page 6 they also describe it. The given numbers do not correspond. For example, according to Fig. 2, the input size for "Improved RFB" is 72x96x48, but according to the text it is 384x288 (the channels are not mentioned). The authors should check their architecture and describe it as it is.
7) In Sec. 2.4 the authors provide the normalization factors of the key points. How were they calculated?
8) Throughout the article, there are errors in Figures numbering. For example, in Section 2.3 the authors say that the module structure is shown in Fig.5, but in fact it is shown in Fig. 3. The same thing happens with Sections. For example, Sections 2.3 and 2.4 have the same title. Finally, the paper contains several translation mistakes and typos, e.g., "avgerage accuracy" in Table 2. or "By investigating the network structure" in page 5. All in all, the authors should thoroughly check their text.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
The study on this topic is fascinating. The structure is clear and logical, and challenging. The research is timely and worthwhile.
Authors should follow the style of a structured abstract based on the IMRAD structure of a paper. The abstract should briefly state the purpose of the research, the principal results, and significant conclusions. An abstract is often presented separately from the article, so it must be able to stand alone.
I hope you find the following observations helpful:
Materials and methods: I found this section very important for the paper's readability. Methods should be described in detail. I think the research procedure could be more clearly described using a diagram, highlighting its potential and limit.
Authors should consider more previous works (e.g., theoretical, conceptual, and empirical reviews) published in the literature. Authors should discuss the results and how they can be interpreted from the perspective of previously published studies. I strongly recommend adding these works to the list of references:
- Izonin I., Tkachenko R., Fedushko S., Koziy D., Zub K., Vovk O. (2021) RBF-Based Input Doubling Method for Small Medical Data Processing. Advances in Artificial Systems for Logistics Engineering. ICAILE 2021. Lecture Notes on Data Engineering and Communications Technologies, vol 82. Springer, Cham, 2021. pp 23-31. https://doi.org/10.1007/978-3-030-80475-6_3
I strongly recommend adding these works to the list of references.
Authors should discuss the results and how they can be interpreted from the perspective of previously published studies.
The diagrammatic presentation of the study research will be the most substantial section of this work. I suggest adding a visual presentation of obtained outcomes in section Results.
I also suggest a grammar and spelling review.
The conclusion is thorough.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The authors have revised the paper carefully as per the reviewer's suggestions. The paper can be accepted in the present form.
