All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
Reviewers are satisfied with the revisions, and I recommend accepting this manuscript.
[# PeerJ Staff Note - this decision was reviewed and approved by Xiangjie Kong, a PeerJ Section Editor covering this Section #]
-
-
-
-
**PeerJ Staff Note:** Please ensure that all review, editorial, and staff comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
**Language Note:** When you prepare your next revision, please either (i) have a colleague who is proficient in English and familiar with the subject matter review your manuscript, or (ii) contact a professional editing service to review your manuscript. PeerJ can provide language editing services - you can contact us at [email protected] for pricing (be sure to provide your manuscript number and title). – PeerJ Staff
I carefully reviewed the author's latest version of the manuscript and found that the previously raised concerns have not been fully addressed. Significant issues remain, and the current version does not yet meet the publication standards of the journal. I strongly recommend that the author undertake a thorough revision. The specific comments are as follows:
1. The figure captions remain overly simplistic, particularly for Figure 1 and Figures 3 through 7. The author must provide more detailed and informative captions that clearly explain the content and relevance of each figure. Captions should be self-explanatory and enhance the reader’s understanding without requiring reference to the main text.
2. As noted in the previous review, the manuscript contains substantial issues with writing quality and punctuation. Several sentences lack proper punctuation, which affects readability and clarity. For example:
o A period is missing after Formula 1 and Formula 3.
o There is no punctuation following Formula 4.
o Lines 467 to 474 also lack appropriate punctuation and should be revised accordingly. The author must carefully proofread the manuscript and correct all such issues throughout the text.
3. The manuscript still lacks citations to recent and relevant literature. The author should update the reference list to include current studies that support and contextualize the findings.
-
-
**PeerJ Staff Note:** Please ensure that all review, editorial, and staff comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
This paper proposes a person re-identification framework based on open attribute recognition, combining NLP and computer vision technology to retrieve target persons in surveillance videos through text queries. The method includes NLP module (extracting text attributes), pedestrian detection module (YOLOv9+RetinaFace) and open attribute recognition module (BLIP-2). The experiment achieved 84.8% Rank-1 accuracy on the PA-100k dataset, which is better than the existing methods and is suitable for security, missing persons search and other scenarios. However, this paper still has some problems in paper writing, method innovation and drawing. The specific opinions are as follows:
1. This paper lacks mention of the privacy issues of open attribute recognition in surveillance, and it is recommended to add relevant discussions.
2. Some recent references are missing.
3. There are big problems with the language of this paper. Some sentences are too long and some lack punctuation. For example, there is a missing period before "Experiments demonstrate the effectiveness of our proposed framework, achieving a rank-1 accuracy of 84.8% in identifying and retrieving individuals from video data These results outperform state-of-the-art methods, thus validating the potential of the approach in real-world applications."
4. Abbreviations should be used for the names of the methods in Tables 1 and 2 as much as possible.
5. The figures and tables in this paper need to be significantly revised. The figures in Figures 11 and 12 are deformed, especially in Figure 12, which does not need to be arranged vertically in three figures, occupying a large amount of space. Some legends and annotations need to be further enriched.
no comment
no comment
no comment
1. The paper is well stated in English professionally, without evident typos or ambiguity.
2. The paper lacks some necessary references to previous works. For instance, the datasets illustrated in Table 1 have no citations, as these contribute to ReID community considerably. In Line484, there is no reference to literature of WIDER FACE benchmark. In Line 511, what does the one study refer to? It's not clear and confusing.
3. The structure of this article is much below the accepted. There is two much blank space where tables , equations and figures insert, especially in Page 8, 9, 13 and 17, which makes the paper ill-structured and unpleasant-looking. The order of inserted figures are inappropriate, as Figure 4 is ahead of Figure 3 in Page 10 and 11. The figures are depicted in low resolution with evident blurry areas, such as the texts in Figure 3, the graphs in Figure 8, 9 and 10, and the image attributes in Figure 11 (I can't even see what the attributes of the person are ). Some figures are little relevant to the content of the article. The Figure 4 has nothing to do with the NLP module introduced in Section 3.1, as all of the steps (keyword extraction, attribute verification, and synonym generation) are not mentioned or illustrated. Instead, Figure 4 seems like the whole workflow. In Figure 3, the illustration makes it a little confusing. What's the image query attributes? It should be image gallery attributes from the context.
4. The description of Line422-433 is unclear and needs more details. The attribute fusion is not stated clearly? How can two attributes take element-wise multiplication? Does it take in space of their hidden states or textual embeddings ? What's the shallow neural network used to assign a similarity score ? How does it relates to Equation 4?
5. It would be better if the workflow of Figure 4, 5, 6 is illustrated with more graphics than texts.
1. The person detection module totally separates from other modules. The whole experiments are not implemented in an end-to-end manner, since YOLOv9 and RetinaFace model is just evaluated in WIDER FACE benchmark, not in real-world scenarios. It contradicts to the workflow of Figure 3 and 4, where the detected persons are sent to open attribute recognition (POAR) module. Also, the detection module has nothing to do with person body and it actually detects person face from the contexts. I can't understand why this part exists, since a lots of mistakes.
2. The NLP module in Section 4.3 contradicts to Section 3.1. The ChatGPT API model is used to generate synonyms for recognized attributes In Section 3.1, while the ChatGPT API processes the whole input textual query in Section 4.3.
3. The experiments can't demonstrate that the proposed framework can be applied in real-world scenarios. First, the components of the whole workflow are implemented separately. There is no demo to demonstrated its application in real world. Second, in open attribute recognition (POAR) module, BLIP2 and LLM are used to generate open attribute for every gallery image. But the computation cost of these two models are extremely high, which can't be applied in real-time video surveillance.
1. The paper contributes null findings. The proposed methodology consist of several existing works. Also , it benefits little to real-world applications because it can't be implemented in real-time or deployed in resource-limited hardwares.
2. From the experiments, the proposed methodology can't be applied in real-time and real-world scenarios. The computation cost of the whole workflow is huge and unacceptable.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.