Meta-Knowledge Guided Weakly Supervised Instance Segmentation for Optical and SAR Image Interpretation
Round 1
Reviewer 1 Report
Contributions:
Inspired by human visual perceptual habits, this paper proposes an MGWI-Net for weakly supervised instance segmentation of optical and SAR images. The algorithm mainly designs three different weakly supervised losses for exploiting fundamental, appearent, and detailed knowledge, respectively. The algorithm achieves approximate performance of fully supervised methods on experimental datasets using box-level supervision, saving the cost of labeling for instance segmentation.
Questions：The following issues need to be explained：
（1）The algorithm is inspired by the human visual perception habits of “from coarse to fine”, which is a gradual process, implying a sequential relationship of perception. However, the proposed method utilizes the three types of knowledge in parallel and does not reflect a gradual correction process, which may not match the motivation.
（2）In the comparison experiments, details of the weakly supervised + fully supervised and weakly supervised setups of other comparison methods as well as their differences with the proposed method need to be elucidated.
（3）The "YOLACT", "Mask R-CNN" and "CondInst" used in the comparison experiments are all tailored for fully supervised instance segmentation. Surpassing these methods will not reflect the advantages of the algorithm in this paper. The authors should add comparison experiments with dedicated weakly supervised instance segmentation methods, such as Ref. 36 and Ref. 40 mentioned in this paper.
（4）Line 460,
、
and 
are very important parameters and authors need to clarify the reason for setting them to 1, 1, 0.1 respectively. There are the same problems with 
and 
in line 369 .
（5）In Figure 3, the integration operation needs to be explained clearly.
Author Response
Dear reviewer,
Thank you for the comments concerning our manuscript, “Meta-Knowledge Guided Weakly Supervised Instance Segmentation for Optical and SAR Image Interpretation” (ID: remotesensing-2276725). It is a pleasure to have your approval of our work, and we thank you for the time and effort you have put into reviewing the previous version of the manuscript. Those comments are all valuable and helpful for revising and improving our paper and the essential guiding significance of our research. We have studied the comments carefully and have made corrections which we hope meet with approval. Our detailed responses to your comments are attached. The responses are marked in red, with the critical area bolded. Any revisions to the manuscript have been marked up using the “Track Changes” for review. In the following pages are our point-by-point responses to each of the comments. The line numbers in this cover letter correspond to the “Track Changes—Simple Markup” status in the revised manuscript.
We would like to express our sincere gratitude to the esteemed experts for providing valuable and significant feedback. We have made every effort to improve the manuscript according to your suggestions, and your warm feedback is of great importance to advancing our research.
We appreciate your hard work and valuable feedback and hope these revisions meet with your approval. Thank you once again for your time and consideration.
With best regards,
Sincerely,
Zhisong Pan
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This study is well-written and clearly presents their research methods, results, and potential impact. However, it could also benefit from minor revisions.
1. There are some minor details in the manuscript. Some formatting needs to be corrected in the manuscript, for example, the SAR-CNN has more space on line 70. The format of the references needs to be consistent, and some references have errors, such as the year in [63] appearing twice.
2. The authors claim that the SSDD dataset includes SAR images which cover various polarization methods. What are the effects of different polarization modes on SAR image segmentation?
3. Considering the speckle noise characteristic and geometric distortions of the SAR image which can affect the results. Please provide some details about the effect of noise and geometric distortion on SAR image segmentation.
4.I suggest that the authors provide a table listing the specific structure of the network in the experimental section, which would allow the readers to understand it better.
5.Regarding the 5. Discussion, please discuss potential challenges and limitations. While the advantages of the proposed method are mentioned, it would be valuable to discuss potential challenges and limitations of the approach. Are there any limitations in terms of the types of objects or scenes that the proposed method may struggle with? Are there any potential issues related to the generalization or scalability of the method to different datasets or domains?
6. Authors could provide some suggestions for future research directions based on the findings of this work would be beneficial.
Author Response
Dear reviewer,
Thank you for the comments concerning our manuscript, “Meta-Knowledge Guided Weakly Supervised Instance Segmentation for Optical and SAR Image Interpretation” (ID: remotesensing-2276725). It is a pleasure to have your approval of our work, and we thank you for the time and effort you have put into reviewing the previous version of the manuscript. Those comments are all valuable and helpful for revising and improving our paper and the essential guiding significance of our research. We have studied the comments carefully and have made corrections which we hope meet with approval. Attachment is our detailed responses to your suggestion. The responses are marked in red, with the critical area bolded. Any revisions to the manuscript have been marked up using the “Track Changes” for review. In the following pages are our point-by-point responses to each of the comments. The line numbers in this cover letter correspond to the “Track Changes—Simple Markup” status in the revised manuscript.We would like to express our sincere gratitude to the esteemed experts for providing valuable and significant feedback. We have made every effort to improve the manuscript according to your suggestions, and your warm feedback is of great importance to advancing our research. We have also used “Track Changes” in the manuscript to provide detailed annotations of the modifications for your convenience.
We appreciate your hard work and valuable feedback and hope these revisions meet with your approval. Thank you once again for your time and consideration.
With best regards,
Sincerely,
Zhisong Pan
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
All my concerns are well responded.
