GaitAE: A Cognitive Model-Based Autoencoding Technique for Gait Recognition
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors1-The paper structure should be included at the end of the introduction.
2-With regard to the proposed method (Fig. 2. The GaitAE framework), I have identified many ambiguities. The paragraph explaining this method does not present it enough, nor does it highlight the real contributions made.
2. 1 Can you explain your method in detail?
2.2 What is the encoding sequence h used to extract features at frame level, and how does it influence the quality of the extracted features?
2.3 How does operation G aggregate video features from different frames, and what alternative methods could be used for this aggregation?
2.4 Why are operations G and H referred to as approach refinement, and how do they contribute to the overall process of approach recognition?
3 regarding section 4, I have a few questions:
3.1 Why was a batch size of P ∗ K = 8 ∗ 12 used, and how does this configuration influence learning?
3.2 Why are the numbers of training iterations different for CASIA-B, SUSTech1K (160k) and OU-MVLP (550k), and how do these differences affect the model results?
3.3 What is the significance of reducing the learning rate to 1e-5 after 450k iterations for training on OU-MVLP?
4. I note that the interpretation of results relating to GaitAE's robustness to Gaussian noise could be improved. I invite the authors to provide a detailed explanation of the results, especially when the Gaussian noise parameter is set to 0.200.
Comments on the Quality of English Language
Moderate English language skills are required.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThe paper presents Gait analysis using autoencoder method. The work is interesting and the results presented are reasonable. However, it would be helpful to the readers if the authors provide further details and the suggestions are as follows.
1. In the introduction, include size of the problem, applicability and significance of the work. Elaborate more on current method in practice.
2. In Figure 2, a block with label Novelty is shown. Highlight the novelty in text by explaining the significant contribution to the existing approach.
3. Is there any novel term in the equation 2? Also, highlight how the presented method is different in line with equation 3.
4. Why did you change the training iterations and learning rate for different datasets? If the intention is to compare the effectiveness, all the datasets should be trained with same parameters.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsHere are my comments:
1. The issue description and the ways in which the suggested strategy overcomes the drawbacks of the current methodology must be made explicit in the introduction. Although the overview of earlier research, such GMSN and GQAN, is instructive, it should be more closely related to the suggested approach in order to emphasize its originality.
2. A more thorough comparison of the suggested method's computational efficiency with that of current approaches should be included in the related work section, taking into account the introduction of the HOR strategy.
3. While creative, the explanation of the autoencoder process and HOR approach needs further explanation. More specifically, a more thorough explanation of the reasoning behind the design decisions (such as the particular parameters employed in the HOR approach) should be provided.
4. The given formulae (such the RMSE computation) are understandable, but a more detailed explanation is needed for how the triplet loss is integrated into the total loss function. In contrast to utilizing RMSE alone, how does this integration enhance the model's performance?
5. It is important to talk about how the suggested approach would affect computational complexity during the testing stage. Although the advances, according to the authors, do not introduce additional factors during testing, it is still important to be aware of any potential trade-offs in terms of processing time or resource needs.
6. Although the experiments are well-designed, the way the data are presented may use some work. For instance, a more thorough discussion of the CASIA-B dataset performance under various situations (NM, BG, and CL) is warranted. What particular difficulties does each ailment bring, and how is GaitAE able to help?
7. The ablation research is essential to proving how each part of the suggested approach contributes. The study should, however, provide a more thorough examination of the findings, especially with reference to the HOR strategy's efficacy.
8. While the comparison with cutting-edge techniques is praiseworthy, the research ought to address the statistical significance of the findings. Are the reported benefits substantial enough to support switching from current approaches to GaitAE?
9. The conclusion should more briefly restate the paper's major contributions and offer potential avenues for further investigation. It's also important to emphasize the possible real-world uses for GaitAE.
10. Although the statistics are helpful, they may be improved by adding more thorough captions. For instance, the parts of Figure 2 that are surrounded by the red dotted lines have to have unambiguous labels in the caption.
11. To help visualize the performance variability, tables that summarize the experimental data should contain confidence intervals or standard deviations.
Comments on the Quality of English LanguageMinor editing of English language required.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsAfter a thorough review of the manuscript, I could see that the authors had not only addressed all the questions raised in a clear and precise manner, but had also provided exhaustive and well-argued answers. Their explanations are coherent, and the proposed solutions are both innovative and relevant to the subject at hand. Because of the quality of their work and the rigor of their scientific approach, I strongly recommend the publication of this article. I am convinced that it will make a significant contribution to research in this field.