Cross-Viewpoint Template Matching Based on Heterogeneous Feature Alignment and Pixel-Wise Consensus for Air- and Space-Based Platforms
Round 1
Reviewer 1 Report (New Reviewer)
This paper proposes a template matching method that to solve the problems of heterogeneous image sources, different scales and different viewpoints. The presentation is clear and the experimental results show that the algorithm has good effect and high computational efficiency.
However, the author described the different of the proposed, which is not systematic. And the innovation of the paper needs to be improved. The amount of data to test algorithm performance is small.
Author Response
We would like to express our gratitude to the reviewers for their valuable comments. Thank you for your review and dedication, which have been of great help in revising the paper. The revised content can be found in the attached file.
Author Response File: 
Author Response.docx
Reviewer 2 Report (New Reviewer)
. This paper has been well described and organized as essential for its proposed contributions.
Author Response
We would like to express our gratitude to the reviewers for their valuable comments. Thank you for your review and dedication. The revised content can be found in the attached file.
Author Response File: Author Response.docx
Reviewer 3 Report (New Reviewer)
The article proposes a deep learning based method using template matching for aerial images. The results on the dataset used by the authors are very good in comparison to that reported by other approaches. The article is discussing a very pertinent problem. It is also well written. Only concern is that the results are based on the users dataset, and it would be interesting if there can be some information obtained using a publicly available dataset. Also more information about the dataset images with some sample images and the templates will be helpful.
The results using data from other domains are also interesting. Can some images of these images be also included for better understanding.
Author Response
We would like to express our gratitude to the reviewers for their valuable comments. Thank you for your review and dedication, which have been of great help in revising the paper. The revised content can be found in the attached file.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report (New Reviewer)
The current version of this paper has significantly improved compared to the previous version. All the questions I raised have been well resolved. Therefore, I recommend accept.
This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.
Round 1
Reviewer 1 Report
This manuscript proposes a novel template matching method, which is very different to the traditional template matching techniques, for cross-view heterogeneous images. The authors present a novel algorithm framework and validate its performance on multi-modal image sets. In my opinion this manuscript can be accepted for publication after revision.
Detailed comments:
1. Page 5, line 179 and 181: The use of notation “=” in f^ds_s=(h,w,c) , f^ds_t=(m,n,c) , and S=(h-m+1,w-n+1) are not correct.
2. “gaussian” should be “Gaussian”
3. Eq. (10): The superscripts 1, 2, and 3 mean three scales? Clear description is necessary.
4. Eq. (27)-(28): Changing the subscripts to variables of the confidence map should be more appropriate.
5. Using epoch to determine positive and negative seems too empirical. If possible please use other common conditions; otherwise please provide detailed reasons and discussion.
Author Response
We greatly appreciate the suggestions and comments from the reviewer. Our response can be found in the attached document.
Author Response File: Author Response.docx
Reviewer 2 Report
This paper proposes a remote sensing image matching that is able to find the visual correspondences between two images at different altitudes. The authors proposed a cross-view heterogeneous remote sensing image-matching model, demonstrating that it is more adaptable to the template-matching task than other methods.
In general, the paper does an excellent job presenting the state of the art, the main concepts of template matching, presenting the architecture, and making a comparison of its results with respect to previous works.
I only have a few recommendations for this paper to be published.
- Check the English in the whole article. There are some minor flaws.
- In Figure 1.a and 1.b, indicate that sensors 1 and 2 are at different heights. Also indicate that the images acquired by satellites are called template images, and images acquired by drones are called searching images. Indicate which image corresponds to each data collection sensor in image 1.b.
- Re-design figure 2, so that there are stages in this figure that are in accordance with each sub-section that is dealt with in the materials and methods part. For example, we talk about Dense Feature Extraction, Heterogeneous Feature Alignment, and Multi-scale matching stages, where are these stages indicated in figure 2?
- On line 337: "Infrared to RGB: The Drone-view was used in this part, which is our self-built dataset." Where is the citation for this dataset?
- Check if the format of table 1 is in accordance with the guidelines of the journal. Is the use of included lines to indicate that there is no data allowed? Or preferably can N/A be used in empty spaces?
- I suggest a discussion stage where the most important findings of the article are discussed in detail, as well as the problems that arose within the development of this work, and the comparison of results with other methods.
Author Response
We greatly appreciate the suggestions and comments from the reviewer. Our response can be found in the attached document.
Reviewer 3 Report
The paper is identical to the priorly submitted paper machines-2074658 (revision 1).
The paper does still contain severe problems related to the presentation of the scientific concepts. The description is too close to the python implementation, as, for instance, raised by one of the primary reviewers, e.g.:
“? = (â„Ž − ? + 1, ? − ? + 1) is not mathematically correct. S is a score map while (â„Ž − ? + 1, ? − ? + 1) is a tuple. They are not equal.”
In a similar fashion, various descriptions are motivated by the data layout of the implementation than by the conceptual approach. The authors often respond by explaining the specific python command and the necessary reshaping of array sizes etc., which makes it unnecessarily hard to understand the underlying concepts.
I strongly advice to provide a conceptual explanation of the approach. Any implementation issue is only of interest, if it involves non-standard workflows or similar.
Author Response
We greatly appreciate the suggestions and comments from the reviewer. Our response can be found in the attached document.
Author Response File: Author Response.docx
Reviewer 4 Report
Overall, I am rather happy with this submission. The problem considered is relevant and it is motivated well. The technical approach proposed is sound, sufficiently novel, builds upon the existing work well, and adequately described. The experiments are also sufficiently convincing: sufficiently large in scale, consider the various issues of interest in sufficient detail, and provide good insight into the strengths and weaknesses of the author's method.
My only suggestion for improvement concerns language. There are many instances of incorrect article use, which makes the text somewhat difficult to understand exactly and creates confusion, and at various points the style is in need of greater rigour and better style. The manuscript really should be proof read by a native speaker. Please correct the use of articles first and foremost. Also, kindly do not use shortenings such as "it's" in formal writing; rather, always write "it is".
Author Response
We greatly appreciate the suggestions and comments from the reviewer. Our response can be found in the attached document.
Author Response File: Author Response.docx
Reviewer 5 Report
I don't see a change in the manuscript regarding my suggestions
