A Novel Method to Generate Auto-Labeled Datasets for 3D Vehicle Identification Using a New Contrast Model

Round 1

Reviewer 1 Report

General Comments:

The paper deals with automated object detection and presents a deep-learning approach to generate 3D-based auto-labeled datasets for Autonomous Vehicles (AV) from boundary conditions. The proposed approach is independent of the point of view in which the sensors (cameras or LiDAR’s) are placed. The work has been verified by experiments. The topic is important for AV industry and the proposed approach is useful.

 

 

Specific Comments:

1.    There should be a block diagram of the whole system.

2.    The paper lacks algorithmic / mathematical structure. For example, it is unclear how occlusion level has been handled.

3.    There should be a clear analysis and testing to show that the performance of proposed system is independent of the sensors’ point-of-view.

Author Response

Dear Reviewer,

Thank you for taking the time to review our submission. We appreciate your feedback and have addressed all of the issues you raised in our revised manuscript.

We have included a PDF document with our response that outlines the changes we have made and provides detailed explanations for each revision. Please let us know if you have any further questions or concerns.

Thank you again for your valuable feedback, and we look forward to hearing from you soon.

Best regards,

Author Response File: Author Response.pdf

Reviewer 2 Report

Authors contributions:

According to the authors, auto-labeling is one of the main problems in vehicle detection. To identify objects in LiDAR data, auto-labeled datasets can be used. The large amounts of data is one of the challenge in this task.

To solve this problem, the authors have proposed propose a methodology to generate new 3D based auto-labeling datasets with a different point of view setup than the one used in the most recognized datasets.

After a comparative analysis with other solutions in this study area, the obtained accuracy is above 80%.

 

I have some reviewer notes:

Format the paper according to the journal requirements.

Abstract. Present your results with values, not only text descriptions how well your methodology works. Also, how your work can be implemented in the practice.

Introduction. The aim of this study is not clearly presented. Also, at the end of this part, you can present the structure of your paper and what are your contributions.

Table 1. If the values presented in this table are dimensionless, you have to note it in the text description. Also, F-score have to be defined. What is its meaning?

2.3.Validation models. How do you determine your sample size? It is not clear.

Line 304. OS1-64 (manufacturer, city, country of origin).

Line 311. Nvidia Jetson: Model (manufacturer, city, country of origin).

Line 317. What level of sensor fusion is applied?

Table 3. Also, other tables with same comparisons. What are the measurement units of the presented accuracy?

Discussion part is missing. You have to compare your results with minimum 3 other papers.

Conclusion part. What are the limitations of your work? How the work will be continued? How your results can be implemented in the practice?

 

I have some suggestions:

Improve the presentation of your results. Make more comparative analyses with other papers. These suggestions will improve your contributions.

Author Response

Dear Reviewer,

Thank you for taking the time to review our submission. We appreciate your feedback and have addressed all of the issues you raised in our revised manuscript.

We have included a PDF document with our response that outlines the changes we have made and provides detailed explanations for each revision. Please let us know if you have any further questions or concerns.

Thank you again for your valuable feedback, and we look forward to hearing from you soon.

Best regards,

Author Response File: Author Response.pdf

Reviewer 3 Report

 1.     What’s the unique advantage for your study concerning the statement “The proposed method is based on the YOLO model trained with KITTI” As far as I am concerned, YOLO model was proposed about 4 years ago. Now we have a lot of new model why may perform better.

2.     The statement “Depending on the occlusion level of the object, set 3 levels of difficulty, and the accuracy…” was difficult to follow. Please try to rephrase the statement.

3.     For the line 75, “Traditionally, most of the work has been developed under same positions…” Please add more evidences.

4.     Why YOLO V5 was selected in your study for the model training, instead of YOLO v4, etc. Did authors choose the version only due to its latest version?

5.     More details about Ultralytics, as shown in line 209, were in need.

6.     The following studies were recommended to be properly cited: [1] Exploring influence mechanism of bikesharing on the use of public transportation-a case of Shanghai. Transportation Letters-the International Journal of Transportation Research. 2022. [2] Sensing Data Supported Traffic Flow Prediction via Denoising Schemes and ANN: A Comparison, IEEE Sensors Journal, vol. 20, pp. 14317-14328, 2020.

Author Response

Dear Reviewer,

Thank you for taking the time to review our submission. We appreciate your feedback and have addressed all of the issues you raised in our revised manuscript.

We have included a PDF document with our response that outlines the changes we have made and provides detailed explanations for each revision. Please let us know if you have any further questions or concerns.

Thank you again for your valuable feedback, and we look forward to hearing from you soon.

Best regards,

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The Authors have included clarifications and a “Discussion” section to support analyzing the proposed approach. Block diagrams have also been added to compensate for the algorithmic structure. The current version has better presentation and is suitable for publication.