Next Article in Journal
Securing Edge Devices: Malware Classification with Dual-Attention Deep Network
Next Article in Special Issue
A Spectral Clustering Algorithm for Non-Linear Graph Embedding in Information Networks
Previous Article in Journal
The Interaction of Microplastics and Microbioplastics with Soil and a Comparison of Their Potential to Spread Pathogens
Previous Article in Special Issue
Lightweight Infrared and Visible Image Fusion Based on Nested Connections and Res2Net
 
 
Article
Peer-Review Record

Cross-Domain Person Re-Identification Based on Feature Fusion Invariance

Appl. Sci. 2024, 14(11), 4644; https://doi.org/10.3390/app14114644
by Yushi Zhang 1,†, Heping Song 2 and Jiawei Wei 3,*
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Reviewer 4: Anonymous
Appl. Sci. 2024, 14(11), 4644; https://doi.org/10.3390/app14114644
Submission received: 25 February 2024 / Revised: 12 May 2024 / Accepted: 20 May 2024 / Published: 28 May 2024

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript "Cross-domain person reidentification based on feature fusion invariance" provides a thorough overview of person reidentification, discussing the evolution of methods from traditional feature engineering to deep learning techniques, specifically convolutional neural networks (CNNs). It solves issues such as illumination fluctuations and body position. The topic is supported by the references, which include work on feature extraction and local information-based approaches. The research design is appropriate for cross-domain person reidentification, as it incorporates both supervised and unsupervised learning. The methods are well-documented and encompass feature fusion, feature memory, and loss functions. The results are clearly reported, indicating a considerable improvement in cross-domain individual reidentification. The findings support the conclusions stated, proving the efficacy of the proposed strategy.

The research comprehensively and rigorously addresses the problem of cross-domain person reidentification, presenting an innovative approach backed by solid results. The combination of supervised and unsupervised learning techniques, along with the introduction of elements such as feature memory and a cross-domain invariance loss function, represents a significant contribution to the field. Furthermore, the well-structured and scientifically sound methodology provides a solid foundation for future research in this area. 

Given the foregoing, the current reviewer decides to recommend the publishing of this work. 

Author Response

Thank you for your recognition of our work,please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript describes a novel method combining global and local features for cross-domain person re-identification. It is well written, but the following minor questions/comments should be addressed before publication:

1. The local feature extraction divided the person images into 6 regions. How was this number determined? Will more divisions further improve the performance of the model?

2. It figure 2(c), it is seen that the same person is identified and regions are generated despite the change of percentage area the person occupies in the full image. How will the performance of the model degrade if the person occupies a smaller and smaller area in the full image (i.e., the camera keeps zooming out and more background is included)?

3. When measuring similarity, the classical cosine similarity measure is used. Have the authors considered using other similarity measures commonly employed by the face recognition community? How will that affect the model accuracy?

Author Response

The anonymous reviewers are greatly appreciated for their detailed and constructive comments, which have significantly contributed to the improvement of our manuscript's quality. We have carefully examined the comments and made point-by-point revisions accordingly. For a detailed list of main corrections and our responses to the reviewers' comments, please refer to the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

It is desirable that more similar articles be included in the state of the art, but that they be more recent.

There are some minor grammar mistakes, like "2.2.2. local feature", which, should be written as "2.2.2. Local feature".

Some figures, like Figure 4 should be presented with a bigger size, to facilitate the visualization of the presented results.

The conclusion section must be improved, presenting the most important achievements during the research.

Comments on the Quality of English Language

There are some grammar issues in the paper. It is suggested to review and correct them.

Author Response

The anonymous reviewers are greatly appreciated for their detailed and constructive comments, which have significantly contributed to the improvement of our manuscript's quality. We have carefully examined the comments and made point-by-point revisions accordingly. For a detailed list of main corrections and our responses to the reviewers' comments, please refer to the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

The manuscript presents a cross-domain people re-identification method where the features of pedestrians are learned through supervised learning in the source domain and unsupervised in both domains. It also introduces a memory to store aligned features and designed a cross-domain invariance loss function. The results look impressive. However, I have some comments:

1. The paper is hard to understand, please improve the English.

2. I share a PDF with my comments.

 

Comments for author File: Comments.pdf

Comments on the Quality of English Language

The English is poor and needs extensive editing.

Author Response

The anonymous reviewers are greatly appreciated for their detailed and constructive comments, which have significantly contributed to the improvement of our manuscript's quality. We have carefully examined the comments and made point-by-point revisions accordingly. For a detailed list of main corrections and our responses to the reviewers' comments, please refer to the attachment.

Comments 1: The paper is hard to understand, please improve the English.

Response 1:

We extend our deepest apologies for the oversight encountered. Prompt corrective action has been taken to rectify this matter, and we have meticulously reviewed and revised the entire document to guarantee the absence of any grammatical inaccuracies.

Comments 2:  I share a PDF with my comments.

Response 2:

We will provide a PDF file that has been modified according to your suggestions, and the modified part will be highlighted

Comments 3: You mentioned in your comments whether the code can be made public

Response 3:

In response to your inquiry, we would like to express our apologies as we do not have plans to publicly release the code at this moment, as our research based on this work is still ongoing. rest assured, we will make the code available at an appropriate time in the future when our research progresses further.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Comments and Suggestions for Authors

Thank you for considering all the comments in my previous review. I think the manuscript is ready to be published, just consider another minor English language review.

  Comments on the Quality of English Language

The authors should consider another minor review to correct some grammar errors.

Reviewer 4 Report

Comments and Suggestions for Authors

Thank your for attending the comments.

Back to TopTop