Equal Emphasis on Data and Network: A Two-Stage 3D Point Cloud Object Detection Algorithm with Feature Alignment
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsGeneral: 1. What is the main question addressed by the research? Lidar improvement with frameworks. 2. Do you consider the topic original or relevant in the field? Does it address a specific gap in the field? Yes. 3. What does it add to the subject area compared with other published material? New framework. 4. What specific improvements should the authors consider regarding the methodology? What further controls should be considered? See my comments. 5. Are the conclusions consistent with the evidence and arguments presented and do they address the main question posed? It can be improved. 6. Are the references appropriate? Yes. Additional: 1. It is suggested to have 5 keywords to improve paper's visibility. 2. Abstract - "...we found that the lidar reflection intensity...". Not appropriate sentence. Fix this. 3. Abstract - "To meet these challenges...". It is suggested to state "The objectives..." 4. Introduction is too long. Some contents can be summarized. 5. Introduction - previous works can be summarized. 6. One of your contributions is "optimization". You need to explain more about the method used. 7. 2.1 The Data is Imbalanced. The title is not appropriate. 8. Not enough explanation regarding this method (section 2.1, 2.2). 9. Please include and explain optimization method used in this experiment. 10. Good number of training, testing and validation set. 11. Other than IoU, it is suggested to add other segmentation evaluation metric such as F1 score, Jaccard, ROC curve, etc. 12. Conclusion - please include some results. 13. References - up-to-date.
Author Response
Response Letter
Thanks a lot for your professional and insightful comments which make this paper well revised and improved. And for the upcoming New Year, we wish you a happy New Year and all the best!
Response：
1. General: 1. What is the main question addressed by the research? Lidar improvement with frameworks. 2. Do you consider the topic original or relevant in the field? Does it address a specific gap in the field? Yes. 3. What does it add to the subject area compared with other published material? New framework. 4. What specific improvements should the authors consider regarding the methodology? What further controls should be considered? See my comments. 5. Are the conclusions consistent with the evidence and arguments presented and do they address the main question posed? It can be improved. 6. Are the references appropriate? Yes.
Answer: Thanks a lot for your thoughtful reading and valuable suggestions. We have also made modifications one by one according to your suggestions, the details are as follows.
2. Additional: 1. It is suggested to have 5 keywords to improve paper's visibility. 2. Abstract - "...we found that the lidar reflection intensity...". Not appropriate sentence. Fix this. 3. Abstract - "To meet these challenges...". It is suggested to state "The objectives..." 4. Introduction is too long. Some contents can be summarized. 5. Introduction - previous works can be summarized.
Answer: Thanks for your profound consideration and suggestion.We have added a fifth keyword "feature fusion" and modified the mentioned text content. We have also summarized the previous work, for details, please see lines 68-86, 90-112, 114-136.
3. One of your contributions is "optimization". You need to explain more about the method used.
Answer: Thanks for your professional advice. In the section 2, we have added the content expression of the proposed method on optimization, for details, please see the red content in the section 2.
4. 2.1The Data is Imbalanced. The title is not appropriate.
Answer: Thanks for your professional suggestion. We have changed the title to "Analysis of Data".
5. Not enough explanation regarding this method (section 2.1, 2.2).
Answer: Thanks for your professional advice. We have added more explanations of the proposed methods in section 2,.For details, please see the red text in section 2.
6. Please include and explain optimization method used in this experiment.
Answer: Thanks for your professional suggestion. In Section 3.1, we added the explanation of using optimization method in the experiment. For details, please see lines 431-439.
7. Good number of training, testing and validation set.
Answer: Thanks for your professional suggestion. In our study, all experiments were based on the available benchmark KITTI dataset, including 7481 training samples and 7518 test samples.
8. Other than IoU, it is suggested to add other segmentation evaluation metric such as F1 score, Jaccard, ROC curve, etc.
Answer: Thanks for your professional suggestion. We have added Frames Per Second (FPS) metrics for algorithms in Table 2 to evaluate real-time performance. Due to the lack of unified use of other segmentation evaluation metrics in the KITTI dataset, it is also difficult for us to achieve comparison, so we did not add other metrics.
9. Conclusion - please include some results.
Answer: Thanks for your professional suggestion. We have added some results to the conclusion. Please see lines 571-576.
10. References up-to-date.
Answer: Thanks for your professional suggestion.We have updated references. Thanks again for your help in our work and send our best wishes.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Comments and Suggestions for Authors1. Row 113 the sentence “In general, this paper is based on practical problems…” should be “This paper is based on practical problems…”
2. Rewrite the sentence on row 146 “In this section, we first introduce the relevant analysis….”. Don’t use first person pronouns "we"
3. Row 187 the word “Where” should be with small letter “w” and without intend. The same for rows 219, 266, 403
4. The symbols in the main text and in equations should be with the same style - Italic or not.
5. Reduce the text into the Figures captions. The first sentence to be the figure caption and the other text can be moved to the main text.
6. Check once again the formatting requirements for Table 1.
7. Rewrite the first sentence in the conclusion section
8. Extend the reference list with relevant publications, published in MDPI journals.
Author Response
Response Letter
Thanks a lot for your professional and insightful comments which make this paper well revised and improved. And for the upcoming New Year, we wish you a happy New Year and all the best!
Response：
1. Row 113 the sentence “In general, this paper is based on practical problems…” should be “This paper is based on practical problems…”
Answer：Thanks for your attentive reading and professional advice. We have revised this sentence.
2. Rewrite the sentence on row 146 “In this section, we first introduce the relevant analysis….”. Don’t use first person pronouns "we"
Answer：Thanks for your professional suggestion. We have rewritten these contents. For details, please see row 166-171.
3. Row 187 the word “Where” should be with small letter “w” and without intend. The same for rows 219, 266, 403.
Answer：Thanks for your attentive reading and professional advice. We have modified the relevant text.
4. The symbols in the main text and in equations should be with the same style - Italic or not.
Answer：Thanks for your attentive reading and professional advice. We have unified the relevant texts and equations.
5. Reduce the text into the Figures captions. The first sentence to be the figure caption and the other text can be moved to the main text.
Answer：Thanks for your professional suggestion. We have simplified the caption of figures.
6. Check once again the formatting requirements for Table 1.
Answer：Thanks for your attentive reading. We have revised Table 1.
7. Rewrite the first sentence in the conclusion section.
Answer：Thanks for your professional suggestion. We have rewritten the first sentence of the conclusion section. For details, please see row 537-539.
8. Extend the reference list with relevant publications, published in MDPI journals.
Answer：Thanks for your professional suggestion. We have referenced and cited relevant publications published in MDPI journals to expand the reference list. Thanks again for your help in our work and send our best wishes.
Author Response File: Author Response.docx
Reviewer 3 Report
Comments and Suggestions for Authors1. Introduction part does not explain the defects of the point-based method, and the work of this paper is a point-based method, the authors briefly explain the work of this paper at the end of the paragraph after introducing the point-based method, but after that, the paper introduces the multimodal method at a large length, thus there are some logical problems.
2. The contribution part is too verbose and needs to be condensed, and the contribution only mentions the proposal of each module. It is better to mention the advantages such as the accuracy improvement brought by the whole algorithm at the end.
3. The change of sparsity and reflection intensity of point clouds collected by LiDAR with distance is well known, and the relevant text can be appropriately simplified, fig1 and fig2 can make a graph to show the change of point cloud reflection intensity and sparsity with distance at the same time.
4. The first paragraph of Section 2 will be an inappropriate paragraph for the introduction of the structure of Chapter 2 and the introduction of the KITTI dataset, and should be divided into two separate paragraphs, and explain the difficulty classification of KITTI's training tasks, the official recommended indicators, what are the objectives, etc..
5. Table1 has a high accuracy, but it seems to be tested on its own divided test set. Other SOTA methods should be experimented with and the performance of other SOTA methods on this test set should be introduced. Is Table2 tested in the KITTI online test set? If so, it should be noted that this is the result of the experiment obtained in the online test set after submission.
6. Real-time is very important in autonomous driving tasks, this paper does not show and compare the FPS indicators of the algorithms. The point-based method has shortcomings in real-time, does the improvement of this paper improve the defect of baseline? If not, are the sacrifices of timeliness and the improvement of accuracy complementary?
7. It is recommended to make the best results bold in each table.
Comments on the Quality of English LanguageModerate editing of English language required
Author Response
Response Letter
Thanks a lot for your professional and insightful comments which make this paper well revised and improved. And for the upcoming New Year, we wish you a happy New Year and all the best!
Response：
1. Introduction part does not explain the defects of the point-based method, and the work of this paper is a point-based method, the authors briefly explain the work of this paper at the end of the paragraph after introducing the point-based method, but after that, the paper introduces the multimodal method at a large length, thus there are some logical problems.
Answer: Thanks for your attentive reading and professional advice. We have revised the text content mentioned and explained the defects of the point-based method. For details, please see lines 90-112.
2. The contribution part is too verbose and needs to be condensed, and the contribution only mentions the proposal of each module. It is better to mention the advantages such as the accuracy improvement brought by the whole algorithm at the end.
Answer: Thanks for your attentive reading and profound advice.We have revised the relevant content and summarized the advantages of different methods. For details, please see lines 68-86, 90-112, 114-136.
3. The change of sparsity and reflection intensity of point clouds collected by LiDAR with distance is well known, and the relevant text can be appropriately simplified, fig1 and fig2 can make a graph to show the change of point cloud reflection intensity and sparsity with distance at the same time.
Answer: Thanks for your profound consideration and suggestion. We have simplified the relevant text and combined fig1 and fig2 as a graph showing the change of reflection intensity and sparsity of point cloud with distance.
4. The first paragraph of Section 2 will be an inappropriate paragraph for the introduction of the structure of Chapter 2 and the introduction of the KITTI dataset, and should be divided into two separate paragraphs, and explain the difficulty classification of KITTI's training tasks, the official recommended indicators, what are the objectives, etc..
Answer: Thanks for your profound consideration and suggestion. We have revised and added relevant contents according to your suggestions. For details, please see lines 172-189.
5. Table1 has a high accuracy, but it seems to be tested on its own divided test set. Other SOTA methods should be experimented with and the performance of other SOTA methods on this test set should be introduced. Is Table2 tested in the KITTI online test set? If so, it should be noted that this is the result of the experiment obtained in the online test set after submission.
Answer: Thanks for your attentive reading and professional advice. The data applied in the experiment is the test set officially divided by KITTI, without other modifications. At the same time, our experiment is also tested on the open platform OpenPCDet to ensure the fairness of the experiment. In addition, we also added the experimental results of the baseline method PointRCNN in Table 1 to show the improvement of the proposed method through comparison. For details, please see lines 458-466.
6. Real-time is very important in autonomous driving tasks, this paper does not show and compare the FPS indicators of the algorithms. The point-based method has shortcomings in real-time, does the improvement of this paper improve the defect of baseline? If not, are the sacrifices of timeliness and the improvement of accuracy complementary?
Answer: Thanks for your profound consideration and suggestion. We have added FPS metrics for the comparison algorithms in Table 2. In terms of real-time performance, due to the data optimization and filtering before network training, the efficiency of our method is increased by 53% compared with the FPS index of the benchmark method PointRCNN. For details, please see Table2 and lines 477-479, 487-491.
7. It is recommended to make the best results bold in each table.
Answer: Thanks for your professional suggestion. We have bolded the best results in each table.
8. Moderate editing of English language required
Answer: Thanks for your professional suggestion. We have improved the English language. Thanks again for your help in our work and send our best wishes.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsGood work.
