Use of GAN to Help Networks to Detect Urban Change Accurately

Round 1

Reviewer 1 Report

GAN-based change detection methods have been proposed by many scholars in recent years, the innovation of the paper is not high enough, the writing quality is mediocre. Especially lacking of sufficient experiments and theoretical analysis.

 Here are some detailed questions for this paper:

(1)   There are far too many errors in grammar, punctuation, and English usage, the authors are encouraged to very carefully revise the writing in this paper.

(2)   In the first contribution, the authors claim that existing methods ignored the pixel-to-pixel relation-ship. But, the solution to this problem is not mentioned in this paper.

(3)   What changes need to be detected? The paper didn’t mention it.

(4)   Fig.6, please use “concatenation” instead of “connection”.

(5)   How do you train your Central-surround model? Finetuning?

(6)   Central-surround Architecture, Attention Fusion Module. How do them work together as a joint model?  Fig. 5,6,7 do not show how they are used.

(7)   P11, “I7-11700 @ 2.50GHz, a gpu of 3060ti- 319 8g, and a memory of 16g,” Please use capital letters for technical terms.

(8)   There are many remote sensing datasets for change detection. The authors are suggested to use open-source datasets to further verify the effectiveness of the proposed method.

(9)   “FC-Siam-diff” was proposed by Daudt et al. in 2018, I don't think the experimental comparisons are convincing. In addition, the parameter size of the model is not the same, at least the time comparison or parameter size comparison should be provided.

 

 

 

 

 

 

 

 

 

 

 

 

 

Author Response

We sincerely that you for providing us with such a valuable revision opportunity. Thus, we can futher improve and present our studies. The comments form you were highly insightful and enable us to greatly improve the quality of our manuscript.We have carefully revised the manuscript, which is in the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper presents a novel application of GAN for detecting urban changes from remote sensing images. The proposed method is based on an improved generator network and a discriminator network. The detailed change detection (CD) results are shown, analysed, and compared  for the blocks of urban images of the two periods. 

The paper is interesting and technically sound while it needs few minor revisions as follows: 

1. The words 'Generative' and 'Generator '  are mixed up somewhere in the paper. Please use either Generator or Generative all through.

2. It is hard for the reader to find the difference between UNET-CD-G and UNET-CD-GAN. Please explain it clearly in the paper mentioning  their function, role and performance. Does UNET-CD-GAN performs better?

3. What is ASPP?

4. What is rationale for the use of block 8 and block 12 for the demonstration and analysis of the results?

5. For the data preprocessing part in page 4 (end of the paragraph), it is mentioned that no data enhancement is performed if the number of change pixels in the reference change image is less than 200 pixels. What is the basis for this assumption? Please clarify it.

Author Response

We sincerely that you for providing us with such a valuable revision opportunity. Thus, we can futher improve and present our studies. The comments form you were highly insightful and enable us to greatly improve the quality of our manuscript.We have carefully revised the manuscript, which is in the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

This manuscript presents the detect urban change using GAN. 

 

In general, this application using GAN is good and can be applied in many real applications.

 

I have several comments as,

1. Please write concise the main contribution (lines 90 to 108),

2.  Explain more detail in the caption of each Figure,

 

3. Read and check carefully Grammarly and English style, (many Grammarly problems)

4. Edit and name all equations such as in lines 190, 268, 277,...

5. Table 4, 5, ... were written in Chinese

6. How can you create the dataset and GT? (was the dataset published?)

7. Please, compare time computational with other methods.

Author Response

We sincerely that you for providing us with such a valuable revision opportunity. Thus, we can futher improve and present our studies. The comments form you were highly insightful and enable us to greatly improve the quality of our manuscript.We have carefully revised the manuscript, which is in the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Thank you for your efforts,

I have no further comments.