DCAT: Dual Cross-Attention-Based Transformer for Change Detection
Round 1
Reviewer 1 Report
Article looks well written, interesting and promising with results. However, if in the article (in abstract) link to github will be provided, it would be a good to double check the solution with described process during review process. Now there is only "The code will be released soon".
Author Response
Response to Reviewer 1 Comments
Point1: Article looks well written, interesting and promising with results. However, if in the article (in abstract) link to github will be provided, it would be a good to double check the solution with described process during review process. Now there is only "The code will be released soon".
Response 1: Thank you very much for your positive comment. We ensure that the code will be released after the paper is accepted.
Reviewer 2 Report
The work is interesting and the authors seem to have put their efforts to make the paper understandable and clear for readers.
However, I would suggest they should add a section about and important stage of their work and the process of image analysis - SEGMENTATION. They have given information about that in many places and whenever required, but without details on how. In my opinion this is not a must, but rather a way of improving the work and making it more interesyting for image processing applications.
Author Response
Response to Reviewer 2 Comments
Point 1: However, I would suggest they should add a section about and important stage of their work and the process of image analysis - SEGMENTATION. They have given information about that in many places and whenever required, but without details on how. In my opinion this is not a must, but rather a way of improving the work and making it more interesting for image processing applications.
Response 1: Thank you for your advice. We think the current version is enough for understanding the algorithm. For the space limitation, unnecessary details are not added.
Reviewer 3 Report
This study develops a method using a dual cross-attention transformer for addressing the ability of fast-changing high frequencies or similarly slow-changing low frequencies to effectively represent complex dual-time features and to make full use of the cross-attention mechanism. The results of the study appear to be positive, but there are still issues that need to be clarified.
1. Are any other state of the art methods using similar ideas? If yes, please add some comparisons. If not, please discuss why (it could be that this idea is totally new, and this should be pointed out).
2. The traditional multi-feature-based change detection methods should be reviewed in the introduction section, these studies can be used to introduce your research points. Some literatures need to be supplemented.
[1] Change Detection Based on a Multifeature Probabilistic Ensemble Conditional Random Field Model for High Spatial Resolution Remote Sensing Imagery, IEEE Geosci. Remote Sens. Lett., 2016.
[2] Integrating change magnitude maps of spectrally enhanced multi-features for land cover change detection, IJRS, 2021.
[3] An improved change detection approach using tri-temporal logic-verified change vector analysis, ISRPS P&S, 2020.
3. The experiments are very comprehensively done, but some of the comparative methods do not seem to be particularly new in time.
4. Can you explain how you came up with the idea of using the dual cross-attention for change detection.
5. There are many methods for extracting high and low frequency information extraction from the perspective of signal processing, why not use other high and low frequency extraction methods? What are the advantages of the proposed method compared to them?
6. This is not the first time that cross-attention has been proposed, please explain the difference between the proposed cross-attention and other cross-attention.
Author Response
Thank you for your constructive and sincere comments, please refer to the word attachment.
Author Response File: 
Author Response.docx
