Visible–Infrared Person Re-Identification via Global Feature Constraints Led by Local Features

Round 1

Reviewer 1 Report

This article presents an approach for Person Re-Identification from Visible-Infrared images.

Following, the significant issues are listed.

1. The introduction should provide a deep overview of the study.
Instead, I think there is a lack in this sense: it is unclear what unique challenges
are associated with the task faced by the authors.
Therefore, the introduction should contain more details about the open research problems and
clarify the contributions of this work on how to address these research challenges.

Also, what is the reason why the combination investigated concerns ResNet-50? Please motivate.

2. I think that the SOTA is well introduced. However, I think the authors could consider to expand the
section to similar works that made use of non-infrared images for the same task.
Moreover, I think the authors should stress the research gap between this work and the limitations of other existing work.
Some (non-exhaustive) examples are:
    - https://link.springer.com/article/10.1007/s00521-020-05566-3
    - https://www.sciencedirect.com/science/article/pii/S0167865521000805?via%3Dihub
    - https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4156576
    - https://www.sciencedirect.com/science/article/pii/S0031320320302272?via%3Dihub
    - https://ieeexplore.ieee.org/document/9412485
    
3. Section 4 is generally ok, apart from the details regarding the datasets.
How were the training and testing images chosen?
What are their main characteristics and features? Do they need preprocessing steps?
Finally, and most important, were the images for the experimentations identical to Fig.4?
Please, clarify these aspects and add more details about the input images and the training/testing setup.

4. Also in Section 4, the results are well presented in Tables 1,2,3,4. However, there is a lack of discussion
comparing the results obtained with the state of the art of the methods exhibited.
How have the authors improved the state of the art?
What are the advantages of their proposal?

5. The ablation study gives much prestige to the work done by the authors.
However, again, it would be important and preferable for the authors to provide
justifying reasons for the results obtained and the improvements produced compared with the baseline.

6. Can the proposed methods be applied in a cross-dataset scenario?
Please provide a discussion of using the proposal as a general method.

7. In the conclusion section, the authors should emphasize more the real advantages
of their experimental results over existing ones in order to make them more valid
and clear to the audience.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

 

The authors proposed a Local Features Leading Global Features Network (LoLeG-Net) for Person Re-Identification in RGB-IR image pairs. A combination of ResNet50 and non-local attention blocks is used to obtain the modality-shareable features. The global feature constraints led by local features is utilized for intra-modality variation. The structure of the paper is fine and the results are decent. however, I have some concerns regarding the manuscript:

 

1-Figure: Please demonstrate the size of the layers in the proposed network.

 

 

2-Figure 3: Does the Non-local attention block inspired from "Non-local Neural Networks", wang et al, CVPR 2018? If yes, what are f(xi,xj) and BN in your implementation?

 

 

3-Line 167: Compared to original non-local neural networks, the multiplication is matrix multiplication not dot product. Whats the idea behind of changing this operation?

 

4-Figure 3: Please demonstrate the size of all signals.

 

5-The English should be improved. Here are some examples:

line 24: existing work==> existing works

line 159: two modalities are input into the network ==> two modalities are fed into the network

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I am satisfied with the improvements made by the authors. For me, the article can be published as is.

Best regards

Reviewer 2 Report

I have no more comments.