Partial-to-Partial Point Cloud Registration by Rotation Invariant Features and Spatial Geometric Consistency
Round 1
Reviewer 1 Report
The paper indtroduces rotation invariant and spatial geometry consistency for registration of partially overlapping point clouds. I think, it does not have any new method for the registration of partial overlapping point clouds. It only intruduces the application of point-to-point matching of known methods. The results and matching procedure should be given with details. I can say that the literature review is enough. But, some known methods such as “4 point congruent sets-4PCS” (Dror Aiger et al., 2008. 4-Points Congruent Sets for Robust Pairwise Surface Registration) was ignored. It should be usefull to selecting distinquished details from point clouds, especialy for noisy data. Some comments about the paper given below:
1. DCP, GNN and (…..) Please give abbreviation in the first location on the text.
2. It should be usefull to be given abbreviation list.
3. Eq.2,3,4: Please give detail about weight (W) and learnable weight (α)
4. Page7/Line 210: Actually, your matching definition is very similar to SIFT keypoint matching (Lowe, 2004). What do you offer to difference between d and Md.
Lowe, D.G. Distinctive Image Features from Scale-Invariant Keypoints. Int.J. of Comp. Vis. 60, 91–110 (2004). https://doi.org/10.1023/B:VISI.0000029664.99615.94
5. Page 7/Line 204: Can the RANSAC be tried for matching?
6. Please give units on Table1, Table 2, … You mentioned from the unit on line 247, but it is not informal. It should be better to given on each Table.
7. Table 5 and Table 6 : What is your results fort his experiment. Please Show your method and indicate of your results as bolt type.
As an overal comment about your paper, it gives usefull details about deep learning methods on point cloud registration. The detailed results make it better and convincing. Please take into account mentioned points above.
Author Response
Thank you for your valuable advice. Please see the attachment
Author Response File: 
Author Response.docx
Reviewer 2 Report
In this paper, we propose a new learning-based pipeline to address this issue and 11 deal with partially overlapping 3D point clouds. Specifically, we utilize rotation-invariant local fea-12 tures to guide the point matching task, and utilize cross-attention mechanism to update the feature 13 information between the two point clouds to predict the key points in the overlapping regions.However, there are problems with the paper, both in terms of innovation and quality of presentation. I therefore recommend that significant revisions should be made before this manuscript is considered for acceptance.
1. Revise the formulas in the manuscript according to the template format.
2. Is it possible to add the dataset collected by the author for experimentation to enhance credibility.
3. Specify more details on how the GNN model was trained.
4. Consider simplifying the language and providing more context for technical terms.
5. In the DISCUSSION section, the authors provide a discussion of the problems with the proposed approach, which is certainly useful. But, in my view, 'discussion' is more than just throwing up questions. We can allow for the shortcomings of the 'new' or 'optimised: method', but it is more important to give specific indicators or results here to quantify them.
6. In the Conclusions section, the authors aim to present a new point cloud processing method that is different from the previous ones, and perform several sets of experiments and training to distinguish it from other methods. However, the Conclusions section does not reasonably state what is innovative about g this work. In my opinion, the 'conclusion' is as important as the 'abstract', and the author's presentation in this section gives me concerns about the quality of the presentation of the manuscript, which seems to fall short of the 'remote sensing This does not seem to meet the standard of "remote sensing". It is therefore suggested that this section be rewritten.
In the manuscript, a number of formatting as well as grammatical errors were seen. Some of the errors are even very low-level, so I would suggest that the presentation and grammarj be carefully revised throughout the revised manuscript. Apart from the content of the manuscript, presentation and language are also important components of a scientific article.
Author Response
Thank you for your valuable advice.
Please see the attachment
Author Response File: Author Response.docx
Reviewer 3 Report
Dear Authors,
I have the following minor comments:
1. In the literature review, please also consider least-squares based point cloud co-registration (e.g. https://doi.org/10.1016/j.isprsjprs.2005.02.006)
2. The results given in Table 1 seem unrealistic. Please check, and if correct, please discuss.
3. Conclusions are too brief and do not reflect the outcomes of the study sufficiently.
Author Response
Thank you for your valuable advice.
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
You said that 3d data has no units, thus units are not giving on the Tables for tranlastion (response 6). But is not acceptable situation. The registration is not possible for any 3d data with arbitrary unit. Distance threshold can not applicable without units, and other resgistration stesps.
The paper give information about ANN in point cloud registration but the presentation needs more revision.
Author Response
Thank you for your valuable and thoughtful comments.
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 2 Report
The author has carefully reviewed my suggestions and made changes accordingly, and although they do not fully answer all my questions, the changes are largely complete. Thus, in my personal opinion, the manuscript seems to have met the criteria for "remote sensing". I hope that the scientific editors will decide further.
Minor editing of English language required.
Author Response
Thank you for your valuable and thoughtful comments. We have revised and polished the entire manuscript. All content can be viewed through “Track Changes” function. In this minor revision, we have focused on modifying the methods and results sections to facilitate readers' understanding. The method not only adds more details, but also provides more context for technical terms.
