CoDerainNet: Collaborative Deraining Network for Drone-View Object Detection in Rainy Weather Conditions
Round 1
Reviewer 1 Report
The submission proposed a Collaborative Deraining Network (CoDerainNet), which simultaneously and interactively trains a deraining subnetwork and a DroneDet subnetwork to improve the accuracy of Rainy DroneDet. Overall, the submission is well-structured and the experimental experiments can sufficiently support conclusions. I do not find any major issues with this work. I suggest a major revision to further improve its quality. Here are some limitations and suggestions:
1 A table should be provided to summarize the advantages and disadvantages of similar studies mentioned in Section related work.
2 Please add the reason why CoDerainNet is designed like this. It is more important to explain the design decisions.
3 Please add some pseudocodes for the proposed CoDerainNet to further describe adequately.
4 There is not enough discussion of the experimental results. Please add more discussion on the experimental results of ablation studies.
5 Although the article is written well, the authors should also check the article for typo errors and English grammar.
6 Please check the style and format of references.
7 What about the experimental results when the image's contrast is low or in low light?
Author Response
Dear Sir or Madam,
Thank you very much for allowing major revisions of our manuscript, with an opportunity to address your comments. We appreciate very much your valuable suggestions and comments. We have revised our manuscript accordingly.
We are uploading the following materials for the second-round review of our revised manuscript:
- The item-by-item response to the comments (below) (response to reviewers),
- A revised manuscript with all changes highlighted in blue,
- A clean revised manuscript without highlights (PDF main document).
Your favorable consideration on our revision would be most appreciated. Thank you!
Best regards,
Yue Xi, on behalf of all co-authors
Author Response File: 
Author Response.docx
Reviewer 2 Report
Summary: In order to increase the precision of Rainy DroneDet, the authors of this study suggest a Collaborative Deraining Network, or "CoDerainNet," which simultaneously and interactively trains a deraining subnetwork and a droneDet subnetwork. In order to eliminate rain-specific interference in features for DroneDet, they also suggest a collaborative teaching paradigm termed "ColTeaching" that makes use of rain-free features retrieved by the Deraining Subnetwork and teaches the DroneDet Subnetwork similar features. Suggestions and Comments: - The paper is well-written and well-structured. - In addition, the considered topic is very interesting, and the obtained results are very promising. - It is better to avoid including formulas and equations in the introduction. - Is it possible to share the different datasets and the obtained AI-Model? - What about the computation time? - Is it better to make calculations for "cleaning" the rainy photos inside the drones or remotely on external servers having higher computation capacities? - The authors need to add a short paragraph about the security aspects of drone communications. May this influence the proposed approach? - For this purpose, they may include the following references (and others) in their study: 1. https://ieeexplore.ieee.org/document/9842403 2. https://www.mdpi.com/1424-8220/21/6/2057 - Is it possible to extend the same approach to the case of foggy photos? - Similarly, is it possible to extend the proposed approach to the case of autonomous cars running in bad weather conditions?The authors are asked to identify the limitations of their approach and to propose more work directions for the future.
Author Response
Dear Sir or Madam,
Thank you very much for allowing major revisions of our manuscript, with an opportunity to address your comments. We appreciate very much your valuable suggestions and comments. We have revised our manuscript accordingly.
We are uploading the following materials for the second-round review of our revised manuscript:
- The item-by-item response to the comments (below) (response to reviewers),
- A revised manuscript with all changes highlighted in blue,
- A clean revised manuscript without highlights (PDF main document).
Your favorable consideration on our revision would be most appreciated. Thank you!
Best regards,
Yue Xi, on behalf of all co-authors
Author Response File: Author Response.docx
Reviewer 3 Report
The article gives a general description of the collaborative Deraining Network for Drone-view Object Detection in Rainy Weather Conditions algorithm.
1. There are many optimizers for neural networks, the authors would have to enter at least one sentence explaining why the Adam optimizer is used (276)
2. The authors write about super resolution, the type and implementation of which is not explained (191).
3. The article uses an image pyramid. The pyramid can be implemented in various ways, for example, a Gaussian pyramid, a wavelet tree, a wavelet decomposition, or anothers (213).
4. Significantly small amount of real data compared to the amount of synthetic data. (Table 1). There is also a comparison of the final network with seven SOTA networks trained on the same data sets. It can be seen from the results that the network works with less accuracy on real data, which is explained by the insufficient number of image pairs in the dataset on which the network was trained, compared to synthetic data.
5. In the plans for further work, the authors have the adaptation of the network to other weather conditions and experiments with unsupervised learning, the position raises questions. Therefore, it would be good to prove or show the possibilities of unsupervised learning with examples.
Author Response
Dear Sir or Madam,
Thank you very much for allowing major revisions of our manuscript, with an opportunity to address your comments. We appreciate very much your valuable suggestions and comments. We have revised our manuscript accordingly.
We are uploading the following materials for the second-round review of our revised manuscript:
- The item-by-item response to the comments (below) (response to reviewers),
- A revised manuscript with all changes highlighted in blue,
- A clean revised manuscript without highlights (PDF main document).
Your favorable consideration on our revision would be most appreciated. Thank you!
Best regards,
Yue Xi, on behalf of all co-authors
Author Response File: Author Response.docx
Reviewer 4 Report
This manuscript proposed Collaborative Deraining Network for Drone-view Object Detection in Rainy Weather Conditions. Firstly, it proposed CoDerainNet, a light object detector for Rainy DroneDet. In addition, it proposed ColTeaching, which could transfer intrinsic degradation knowledge from Deraining Subnetwork to DroneDet Subnetwork to block rain-specific interference in features for Rainy DroneDet. Finally, it built three drone-captured datasets, including two synthetic datasets and one real dataset for experiments.
In general, the manuscript is well structured and the experiments are abundant. I suggest a Major revision of the manuscript to further improve its quality. There are some a little bit confused problems in this submission.
Here are some comments.
(1) Please add the organization of this submission in Section Introduction.
(2) In Section 2, the related works is reviewed from the two aspects, general object detection under rainy weather conditions and DroneDet. Please add some discussion for why these existing methods can not solve the problems in this submission.
(3) In Section 4.2, please give the detailed description of the Deraining Subnetwork structure.
(4) In Section 4.3, please give the concrete formula of the loss function for training DroneNet Subnetwork.
(5) In Section 5.2.2, this evaluation metrics have been widely used. Its description is a little bit redundant, please reduce the length of this part.
(6) In Table 4, how did you improve the AP50 on RainVisDrone of the baseline from 56.75% to 62.15%. The authors should introduce the improvement step by step.
Author Response
Dear Sir or Madam,
Thank you very much for allowing major revisions of our manuscript, with an opportunity to address your comments. We appreciate very much your valuable suggestions and comments. We have revised our manuscript accordingly.
We are uploading the following materials for the second-round review of our revised manuscript:
- The item-by-item response to the comments (below) (response to reviewers),
- A revised manuscript with all changes highlighted in blue,
- A clean revised manuscript without highlights (PDF main document).
Your favorable consideration on our revision would be most appreciated. Thank you!
Best regards,
Yue Xi, on behalf of all co-authors
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
The authors considered all my comments and suggestions. Good luck.
Reviewer 4 Report
The authors have responded carefully to my comments. All the questions and concerns have been answered and addressed appropriately. I therefore would recommend to accept the manuscript for publication.
