Semantic Attention and Structured Model for Weakly Supervised Instance Segmentation in Optical and SAR Remote Sensing Imagery

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

A semantic attention enhancement and structured teacher guided multi-scale weakly  supervised instance segmentation network (SAST-Net) is studied.  This research provides a low-cost, high- quality solution for the instance segmentation task in optical and SAR RS imagery .

Please supplement the analysis of the processing speed of this method.

Author Response

Dear reviewer,

We would like to express our gratitude for your letter and the valuable comments provided by the reviewers regarding our manuscript titled "Semantic Attention and Structured Teacher for Weakly Supervised Instance Segmentation in Optical and SAR Remote Sensing Imagery" (ID: remotesensing-2536647). We sincerely appreciate your approval of our work and the time and effort you dedicated to reviewing the previous version of our manuscript. Your comments have been incredibly valuable and helpful in revising and improving our paper, emphasizing the significance of our research.

We have carefully studied the comments and made the necessary corrections accordingly. Please find the attached document containing a point-by-point response to your comments. We kindly request you to download and review it. In our response, the revised sections are marked in red font, with critical areas highlighted in bold. We have also utilized the "Track Changes" feature in the revised manuscript to facilitate the review process.

Once again, we extend our gratitude for your invaluable feedback and guidance. We eagerly await your evaluation of the revised manuscript.

Sincerely,

Zhisong Pan

Author Response File: Author Response.docx

Reviewer 2 Report

Comments and Suggestions for Authors

This paper proposes a new network for instance segmentation. The network constructs weakly supervised constraints by modeling the spatial domain relationships through the segmentation branch. It includes a semantic attention enhancement module (SAE module), a structured teacher guidance module (STG module), and a multi-scale feature extraction module (MCFE module), which can achieve the instance segmentation task of remote sensing imagery without pixel-level labeling. the real data is used for validation. overall paper are fine. The comments are given below.

 

1. how does color similarity loss can be applied to SAR data?

2.the feature map should be given

3.the english should be futher improved.

Comments on the Quality of English Language

the english should be futher improved.

Author Response

Dear reviewer,

We would like to express our gratitude for your letter and the valuable comments provided by the reviewers regarding our manuscript titled "Semantic Attention and Structured Teacher for Weakly Supervised Instance Segmentation in Optical and SAR Remote Sensing Imagery" (ID: remotesensing-2536647). We sincerely appreciate your approval of our work and the time and effort you dedicated to reviewing the previous version of our manuscript. Your comments have been incredibly valuable and helpful in revising and improving our paper, emphasizing the significance of our research.

We have carefully studied the comments and made the necessary corrections accordingly. Please find the attached document containing a point-by-point response to your comments. We kindly request you to download and review it. In our response, the revised sections are marked in red font, with critical areas highlighted in bold. We have also utilized the "Track Changes" feature in the revised manuscript to facilitate the review process.

Once again, we extend our gratitude for your invaluable feedback and guidance. We eagerly await your evaluation of the revised manuscript.

Sincerely,

Zhisong Pan

Author Response File: Author Response.docx

Reviewer 3 Report

Comments and Suggestions for Authors

Revision suggestions are shown below:

1. Line 46-Line 48: What are the differences among object detection, semantic segmentation and instance segmentation ?  More explanations about object detection and semantic segmentation are suggested.

2. Line 167-Line 170:  This could not be regarded as a single contribution. Suggest to combine with the first contribution.

3. Ablation Study (Line 682) is not well designed. Try to show the function of each module.

4. A discussion section is suggested before Conclusion.  

Comments on the Quality of English Language

Some revision suggestions on writing:

(1) Captions of the figures and tables are too long. Move some of the texts in the figure captions to the main body of the paper.

(2) Line 139 is confusing.

(3) Line 172:  2.1 Change ‘Instance Segmentation’ to  ‘2.1 Supervised Instance Segmentation’.

(4) Line 722: caption of Figure 6 is wrong.

(5) There are some errors in the language, e.g. Line 735, Line 739. Please check the language of the entire paper.

Author Response

Dear reviewer,

We would like to express our gratitude for your letter and the valuable comments provided by the reviewers regarding our manuscript titled "Semantic Attention and Structured Teacher for Weakly Supervised Instance Segmentation in Optical and SAR Remote Sensing Imagery" (ID: remotesensing-2536647). We sincerely appreciate your approval of our work and the time and effort you dedicated to reviewing the previous version of our manuscript. Your comments have been incredibly valuable and helpful in revising and improving our paper, emphasizing the significance of our research.

We have carefully studied the comments and made the necessary corrections accordingly. Please find the attached document containing a point-by-point response to your comments. We kindly request you to download and review it. In our response, the revised sections are marked in red font, with critical areas highlighted in bold. We have also utilized the "Track Changes" feature in the revised manuscript to facilitate the review process.

Once again, we extend our gratitude for your invaluable feedback and guidance. We eagerly await your evaluation of the revised manuscript.

Sincerely,

Zhisong Pan

Author Response File: Author Response.docx

Reviewer 4 Report

Comments and Suggestions for Authors

Overall, the paper is well-motivated. It proposes a weakly supervised method to tackle the problem of expensive manual pixel-level annotation that prevents scability and generalization. The authors present a semantic attention enhancement module (SAE) to alleviate cluttered backgrounds’ interference, a structured teacher guidance module (STG) to guide the SAE module via edge information, and a mutli-scale feature extraction module (MSFE) to be more robust to scale variants in remote sensing. The results show improvements over state-of-the-art models.

 

While the introduction and experiment sections are well-written, the method section lacks coherence. For example, given the holistic diagram of the framework, I think it would be easier for the readers to follow the network design in a natural way, i.e. from input to output layers. Instead, the authors start with the segmentation branch, which lies at the end of the network, work backward to the SAE module, then move forward to the STG module and conclude the method with MSFE, which handles the input. As much as I am experienced in neural network architecture design, I feel a bit lost reading the method section; any reader would have a difficult time understanding the framework as they have to work back to the holistic diagram frequently. The order of the modules should be adjusted accordingly anywhere mentioned.

 

Apart from a quite convoluted method, this paper can be accepted after addressing the presentation issues in the method section since it is well-motivated. The experiments are comprehensive and able to show significant improvements in performance.

 

Example typos:

 

• Equation (4) and (8), is it $F_{fusi.}$ or $F_{fuse.}$ ?

• Equation (13), is it $M_{goru.}$ or $M_{grou.}$?

Author Response

Dear reviewer,

We would like to express our gratitude for your letter and the valuable comments provided by the reviewers regarding our manuscript titled "Semantic Attention and Structured Teacher for Weakly Supervised Instance Segmentation in Optical and SAR Remote Sensing Imagery" (ID: remotesensing-2536647). We sincerely appreciate your approval of our work and the time and effort you dedicated to reviewing the previous version of our manuscript. Your comments have been incredibly valuable and helpful in revising and improving our paper, emphasizing the significance of our research.

We have carefully studied the comments and made the necessary corrections accordingly. Please find the attached document containing a point-by-point response to your comments. We kindly request you to download and review it. In our response, the revised sections are marked in red font, with critical areas highlighted in bold. We have also utilized the "Track Changes" feature in the revised manuscript to facilitate the review process.

Once again, we extend our gratitude for your invaluable feedback and guidance. We eagerly await your evaluation of the revised manuscript.

Sincerely,

Zhisong Pan

Author Response File: Author Response.docx