Animal Pose Estimation Based on 3D Priors
Round 1
Reviewer 1 Report
Following are my comments and suggestions:
1. Highlight the research motivation and contribution in Introduction.
2. Include flowchart of the model.
3. More mathematical background for the model is required.
4. Rest of the content is satisfactory.
5. The authors can review some of the recent works on image analysis as:
a) https://doi.org/10.1016/j.eswa.2021.115637
b) https://doi.org/10.1016/j.knosys.2021.107432
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 2 Report
There is no significant scientific improvements from the BMVC publication.
New datasets, adding noise and evaluation the response of the algorithms cannot be considered scientifically relevant, in my opinion.
Furtermore, references should be reordered such that reference [1] is the first to appear on text.
It is not clear how many keypoints are required to achieve goos results.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
The research paper titled “Animal Pose Estimation Based on 3D Priors” presents the 2D animal pose estimation using prior 3D knowledge. The work is well presented but there are several confusions that need to be clarified.
1. The authors have explained the gaps in 2D estimation, which is obvious that the performance will be better in 3D than in 2D. Can the proposed technique be interpreted as something between 2D and 3D identification? If yes what significant improvement it can present if we use 3D estimation alone?
2. The 3D dictionary will be derived from that specific animal 3D data images.
3. The proposed methodology section deals with three steps 1) pose dictionary 2) pose estimation and 3) pose refinement. The authors didn’t explain the main task of pose estimation in methodology, the method explanation etc?
4. The linear combination step mentioned in figure 4 is not explained anywhere in the manuscript. How the authors performed the step must be clearly explained in methodology.
5. Also, the optimization step mentioned in figure 4 is not clear or explained?
6. The authors must read some latest research articles from reputed journals on the 3D pose estimation. And discuss some 2021-2022 papers on 3D pose estimation and their problems in related work section.
7. There are some typos in abstract, please correct them
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
There are some improvements.
Author Response
Point 1: There are some improvements.
Response 1: Thank you very much for your positive comments.
Reviewer 3 Report
The authors have improved the manuscript a lot. Please clarify few points mentioned below.
1. Line:224-228: The placement of residual block is unclear. Figure 5 shows the overall proposed network. The shown residual block lies somewhere in figure 5 as discussed in these lines. It would be better if the residual block is highlighted in the figure 5. As the input and output of figure 6 is not clear too.
2. Line 336: The authors claimed that they have constructed two datasets Cat and Amur. However, they have mentioned only Cat dataset in section 3 [dataset collection] and have annotated poor quality images of Amur dataset. Please avoid using contradictory statements in the manuscript Moreover, section 5.1: is also a section of datasets probably used for evaluation. Please make the section title more clearly to avoid confusion for the readers.
3. Section and subsections numbers are not assigned properly in the manuscript like section 4, 5,6.
Author Response
Please see the attachment.
Author Response File: Author Response.docx