An Intra-Class Ranking Metric for Remote Sensing Image Retrieval
Round 1
Reviewer 1 Report
Please see the attached pdf file.
Comments for author File: Comments.pdf
Minor editing of English language required
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
The authors proposed an intra-class ranking metric for remote sensing image retrieval. The presented mansucript is well written and organised. My concerns are as follows;
1. The Validation of all the parameters should be done on the 4 dataset used not just the UCMD and AID. We need to know the effectivenness of the proposed methids against the conventional methods.
2. The results (Ablation Study and Experimental Results) should be on a different section say section 5.
Good and understanding
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
Dear author,
The article is very interesting.
I have carefully read the article titled: "An intra-class ranking metric for remote sensing image retrieval".
The structure of the article seems correct to me.
I recommend the publication of this manuscript.
Best regards
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 4 Report
Clarify the research objective: Clearly state the specific objective of the paper, such as improving remote sensing image retrieval performance using a new sample generation mechanism and intra-class ranking loss function. This will provide a clear focus for the readers and help them understand the significance of the proposed approach. Provide a literature review: Include a comprehensive literature review that discusses the existing approaches and techniques related to remote sensing image retrieval, deep metric learning, and intra-class differences. This will help position the proposed method within the existing research landscape and highlight its novelty and contributions. Define the terminology: Define key technical terms and concepts, such as "embedding space," "intra-class differences," and "constrained positive samples." Providing clear definitions will enhance the reader's understanding of the proposed method and its underlying principles. Elaborate on the sample generation mechanism: Provide a detailed explanation of the new sample generation mechanism used to obtain quantifiable intra-class differences from real positive samples. Describe the constraints and considerations involved in the generation process and explain how it contributes to improving the discriminability of the embedding space. Explain the self-supervised approach: Elaborate on the self-supervised approach used to design the intra-class ranking loss function. Describe the specific methodology employed and how it addresses the issue of maintaining the ranking relationship of samples in the embedding space. Provide insights into how this approach enhances the discriminability of the generated embedding space. Present comprehensive experimental results: Provide detailed information about the experimental setup, including the remote sensing image datasets used, evaluation metrics employed, and comparison methods. Present a thorough analysis and discussion of the experimental results, highlighting the performance improvements achieved by the proposed sample-generated intra-class ranking loss function. Consider including visualizations or examples to support the findings and enhance the clarity of the results. Kindly refrain from using sources that were released before 2019. Cite recent studies that are highly relevant to your subject. The paper also doesn't have enough citations. Another key stage is to compare the topic of the article to other recent publications or works that are comparable to broaden the research's ramifications beyond the subject. Authors may cite and rely on these important works while discussing the subject of their article and the problems of the present. Heidari, A., Jafari Navimipour, N., Unal, M., & Zhang, G. (2023). Machine learning applications in internet-of-drones: systematic review, recent deployments, and open issues. ACM Computing Surveys, 55(12), 1-45.
Mehdi Darbandi; “Proposing New Intelligence Algorithm for Suggesting Better Services to Cloud Users based on Kalman Filtering”; Published by Journal of Computer Sciences and Applications (ISSN: 2328-7268), Vol. 5, Issue 1, 2017; PP. 11-16; DOI: 10.12691/JCSA-5-1-2; USA.
Vahdat, S., The role of IT-based technologies on the management of human resources in the COVID-19 era. Kybernetes, 2021.
Zadeh, F.A., et al., Central obesity accelerates leukocyte telomere length (LTL) shortening in apparently healthy adults: A systematic review and meta-analysis. Critical Reviews in Food Science and Nutrition, 2021: p. 1-10.
Rahhal, Mohamad M. Al, et al. "Contrasting Dual Transformer Architectures for Multi-Modal Remote Sensing Image Retrieval." Applied Sciences 13.1 (2023): 282.
Mehdi Darbandi; “Proposing New Intelligent System for Suggesting Better Service Providers in Cloud Computing based on Kalman Filtering”; Published by HCTL International Journal of Technology Innovations and Research, (ISSN: 2321-1814), Vol. 24, Issue 1, PP. 1-9, Mar. 2017, DOI: 10.5281/Zenodo.1034475.
Moderate editing of English language required
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 5 Report
This paper proposes an intra-class ranking metric loss for remote sensing image retrieval. The author performed preliminary experiments to analyze the effectiveness of the method. However, according to the comparison experiments provided by the authors, the retrieval performance improvement is less than 1%, what is the effect of this slight improvement? Besides, there are some issues in the paper need to be clarified.
(1) The authors should also review other types of remote sensing image retrieval methods, not just those based on metric learning. Besides, some metric learning-driven remote sensing image retrieval methods have been developed in recent years. Accordingly, it is hoped that the author can add the performance evaluation of these methods.
(2) The proposed positive sample generation method is built on the basis of embedding feature similarity measure. For images of the same semantic class, how to ensure the effectiveness of sample mining if the embedding feature distance is larger for images with higher similarity and smaller for images with lower similarity?
(3) The Proxy-Anchor loss also contains similarity constraints for anchor and positive samples, does this duplicate the function of the gen-anchor loss?
(4) The title of section 3.1 should be modified.
(5) The bolded values in Table 1 are not all optimal, so the conclusions of the analysis need to be modified as well.
(6) The bolded markings in tables such as Table 6 and Table 9 are similarly incorrect, and the author should double-check the raw data and revise the analysis corresponding to the results.
(7) The proposed method involves a large number of hyperparameters. Although this paper attempts to analyze the effect of each hyperparameter on the accuracy, in practical applications, a combination of different hyperparameters is required. How to determine or avoid the effect of the interaction of these hyperparameters on the results?
(8) This paper notes that the proposed loss can be easily integrated to other metric learning methods, but the effectiveness is yet to be verified. The improvement of the proposed loss combined with the Proxy-Anchor loss in the paper is tiny.
There are some suspected writing errors. For example, Line 524: The experiment analyzed the impact of different batch sizes on the intra-class ranking loss function retrieval results by.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Please see the attached file.
Comments for author File: Comments.pdf
Minor editing of English language required
Author Response
Thank you very much for your valuable comments and recognition. We invited a colleague who is well versed in English writing to check our manuscript and did our best to correct the grammatical errors.
Reviewer 4 Report
No further comment
No further comment
Author Response
Thank you very much for your valuable comments and recognition. We invited a colleague who is well versed in English writing to check our manuscript and did our best to correct the grammatical errors.
Reviewer 5 Report
The authors have responded and revised our suggestions, but there are still some problems here.
With less than 1% improvement in retrieval performance, I still don't think it's significant for the retrieval domain. Sometimes the error of repeated training can be more than 1%. Perhaps the authors could try more complex datasets, domain adaptation or scene classification application scenarios to demonstrate the effect of considering intra-class differences.
The authors have expanded on the related work as suggested, but transitional statements were missing, e.g., in Section 2.1, 2.2.3 and 2.2.5.
Experiment Setup à Experimental Setup
Experiment Results And Analysis àExperimental Results and Analysis
Comparison ExperimentalàComparison Experiment
Author Response
Please see the attachment.
Author Response File: Author Response.docx