DCS-TransUperNet: Road Segmentation Network Based on CSwin Transformer with Dual Resolution
Round 1
Reviewer 1 Report
Dear Authors,
Thank you for your interesting manuscript! In my opinion the work done is interesting and worth of publishing. I have few comments mainly in the presentation part of your work:
- please develop further the abstract, it is mention just in at its end that the overall aim is the road segmentation. I think it is better to be clarified somewhere in the beginning of the abstract.
- Please do a deep proof and language read of the whole manuscript, especially in the Introduction and Relate Work (even here just this title is wrong and suitable in this way).
- You are using a lot of abbreviations and almost nowhere they are not explained. Please add clarification for each abbreviation the first time you are mentioning it. There are also a lot of sentences that either do not make sense or a just not related to previous and next one... Please thoroughly check! (also the guidelines for citations)
- Results - please try to check/rewrite - most of it is a bit confusing to read.
- Discussion - ??? In my opinion your work should have exhaustive discussion! Please add
- Conclusion - should be a bit extended in my opinion.
THank you and regards.
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Abstract and conclusions are not supported with quantity results.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
DCS-TransUperNet: Road Segmentation Network based on 2
CSwin Transformer with Dual-resolution
The author design a novel remote sensing image segmentation network called Dual CSwin
Transformer UperNet (D-TransUperNet), which utilizes the advantages of CSwin Transformer.
Unlike the previous Transformer-based network, the designed DCS-TransUperNet first uses dual
subnetwork encoders with different scales to obtain the coarse-grain and fine feature representations.
Comment:
1. In academic work, comparing the obtained results to some related/recently published works under the same conditions (i.e., databases + protocols of evaluation) is necessary. The objective is to show the superiority of the presented work against the existing ones.
2. Explain all abbreviations in the paper, such as NLP, MRF, SAR, FCN, etc..
3. From Table 3 we can conclude that the method that the author proposes is only slightly different from the other methods. Please add more discussion about the benefits of your proposed method.
4. It is recommended to add more experiments on processing time/detection time for each model.
5. In table 5 D-resUnet achieves the highest precision value compared to others, please explain and discuss further.
6. The author states, In the next work, we will pay attention to building a lighter model based on Transformer to increase path extraction speed. Please discuss the path extraction speed in your proposed method.
7. Check all grammatical and English errors with English proof reading:
From sub-figure (a), there is obvious road false 437
detection in (4), (5) and (6). Compared with (7), (8), the segmentation result of (3) is cleaner
and more continuous. What is cleaner? or this is typo? please check every sentence in your manuscript.
8. Pleased add more related reference:
https://doi.org/10.3390/app11115050
DOI: 10.1109/ACCESS.2021.3094201
DOI: 10.1109/LGRS.2020.2976551
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
The revised version is clear enough, this paper can be accepted.
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.
