Feature Relation Guided Cross-View Image Based Geo-Localization
Round 1
Reviewer 1 Report
Reviewer’s Report on the manuscript entitled:
Feature Relation-Guided Cross-View Image-Based Geo-Localization
The authors proposed a feature relation-guided cross-view image-based geo-localization to consider feature relation between notes and to solve the effect of geometric distortion. They successfully applied it to a popular cross-view dataset. In my view, the method and results are interesting and promising, but the presentation can be further improved. Please see below my comments.
The title is complex and has 4 hyphens. I suggest revising the title. It does not have to mention the name of your model specifically and should not have any abbreviation.
Line 165. Please add some recent references with a sentence description for ResNET50.
For example, please include the following article that shows the robustness of ResNET50 for land cover image classification:
https://doi.org/10.3390/s21238083
And the following article which shows the robustness of ResNet50 for measuring human perception of urban landscape:
https://doi.org/10.1016/j.jag.2022.102886
Reference [5] by Shi et al. in line 581 seems an unpublished document, so you may remove it. Similarly, for Reference [3]. Please check and include the publishers name if any.
Please define all the parameters/variables in Eqs. (1)-(4).
Section 4.2. Have you utilized techniques such as early stopping in your model to prevent over-fitting and speed up computation? Please elaborate. You may refer to the deep transfer learning article above. Please also comment on the computational time of your method vs others listed in Table 1.
In Tables 4,5 and Figures 10,11, like your previous tables and figures, please replace “MC_GRA” with “Ours”.
Line 478. It is not clear what different colors in the heatmap represent. I suggest adding the color bar with label in Figure 12 and elaborate more on what those colors represent.
Lines 542 and 543. You wrote “cross-view image-based geo-localization” twice. Please rewrite this sentence better. Line 551. You can use the abbreviation here, similarly for some other places in the manuscript.
The font size of some of the Figures are small. I suggest improving the figure quality and front size.
Please add the limitations of your method and future direction.
Adding an acronym table at the end of the manuscript would be useful.
Thank you!
Regards,
There are some typos/style/grammar issues that need to be checked and corrected. The title of the manuscript needs to be revised too.
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
1. Page4,line164: "Similiar to most existing cross-view image-based geo-localization methods" The source of the paper on Similiar methods mentioned in the text is not found, please specify if there is, if not, please add it in the literature.
2. Page18,line520: the discussion and analysis of the ablation experiment results are too shallow and one-sided, and the experimental results are not fully understood and explained. It is recommended to carefully analyze the deep-seated causes and supplement the analysis and discussion of the experimental results.
3. The size and layout of Figure 4 on page 7 are slightly prominently observed in the article, please adjust it appropriately.
4. The format of uppercase z in formula 14 and 15 on page 9 is not very consistent with other English letter formats in the article, please check and choose the correct formula editor.
What was the rationale for the topK in line 464 on page 16, please elaborate on the connection between this indicator and the recall@K indicator described in the 4.3 evaluation metric.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
