Review Reports

Response to Reviewer 1 Comments

1. Summary

Thank you for your suggestions on our manuscript entitled "CCBA-NMS-YD: A Vehicle Pedestrian Detection and Tracking Method Based on Improved YOLOv7 and DeepSort". All these suggestions are valuable and helpful for us to revise and improve the paper. We have carefully studied these suggestions and revised them, and we hope to get your approval.

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Can be improved

Is the research design appropriate?

Can be improved

Are the methods adequately described?

Must be improved

Are the results clearly presented?

Must be improved

Are the conclusions supported by the results?

Must be improved

3. Point-by-point response to Comments and Suggestions for Authors

Comments 1: Line 33……….’advanced1….”

Response 1: Thank you for pointing this out. I apologize for the difficulty in reading the references because I did not leave the statements in the original text blank when citing them, and I have now made the change.

Comments 2: Line 40, rewriet—not complete

Response 2: Thank you for pointing this out. I have rewritten that sentence.

Comments 3: Update the literature on attention mechanism

Response 3: Thank you very much indeed for your comments. I've updated the cited literature on attention mechanisms to those published within the last three years.

Comments 4: https://doi.org/10.1371/journal.pone.0264586

Response 4: Thank you very much for your reminder. I've added it to the paper.

Comments 5: Line 51 literatureError! Reference source not found

Response 5: Thank you very much for your reminder, I'm very sorry for the reading inconvenience caused by my negligence, I've now changed all the references.

Comments 6: Repeated use of “A design in the……..” in lines 51, 61, 64 and so on.

Response 6: Thank you very much for your reminder, I have used different sentences to describe it.

Comments 7: What is “MobileNetV3 lightweight module7”

Response 7: Thank you for pointing this out. Because I didn't space the statements in the original text when citing references, it made reading difficult, for which I apologize and have now made changes.

Comments 8: How self adversarial training and data augmentation work in YOLO?

Response 8: Thank you very much indeed for your comments, I will describe how they work in YOLO. Self Adversarial Training: Firstly, based on the original training data, the adversarial samples are generated by specific algorithms; then the model receives both normal samples and adversarial samples during the training process, forcing the model to learn more robust feature representations to distinguish between the normal samples and adversarial samples. Finally, the robustness of the model in practical applications is improved. YOLOv7 introduces the technique of self-adversarial training, which can effectively improve the performance of the model. Data Augmentation: First, according to the task requirements and data characteristics, choose the appropriate data augmentation methods. Common methods include image rotation, scaling, cropping, flipping, color adjustment, etc.; then, during the training process, these augmentation methods are applied to each sample of the original dataset to generate new data samples. In this way, the model can learn the features of the target under different transformation conditions, enhance its adaptability to new data, reduce over-fitting to specific training data, and ultimately improve the model's generalization performance on unknown data. I have added the above into the paper.

Comments 9: The deepsort flow chart has inconsistent notations.

Response 9: Thank you very much for your reminder. I apologize for the inconvenience caused by the inconsistent representation of the images. We have standardized the formatting of the images, including font size, formatting, and text boxes.

Comments 10: More detail on the object tracking module of the proposed architecture may be beneficial

Response 10: Thank you very much indeed for your comments. For the target tracking model DeepSORT-MNV3, I have provided more description.

Comments 11: Soft NMS is already an established method in object detection, then what is the modification proposed in this article?

Response 11: Thank you very much indeed for your comments. We use a linear decay formula instead of the exponential decay in Soft-NMS to make the decrease of scores smoother and more controllable, and this method reduces the over-suppression of scores at high IoU values, which preserves more detection information and helps to improve the accuracy of detection.

Comments 12: In line 292 “we utilize employs a fixed size…………”

Response 12: Thank you very much for your reminder. I'm very sorry for the inconvenience to your reading due to my oversight, I've revised the statement.

4. Response to Comments on the Quality of English Language

Point 1: Extensive editing of English language required

Response 1: Thank you for pointing this out. I am very sorry for the inconvenience of your reading, I have rewritten the article to improve the quality of the English language.

Response to Reviewer 2 Comments

1. Summary

Thank you for your suggestions on our manuscript entitled "CCBA-NMS-YD: A Vehicle Pedestrian Detection and Tracking Method Based on Improved YOLOv7 and DeepSort". All these suggestions are valuable and helpful for us to revise and improve the paper. We have carefully studied these suggestions and revised them, and we hope to get your approval.

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Must be improved

Is the research design appropriate?

Can be improved

Are the methods adequately described?

Must be improved

Are the results clearly presented?

Must be improved

Are the conclusions supported by the results?

Must be improved

3. Point-by-point response to Comments and Suggestions for Authors

Comments 1: "Error! Reference source not found." - several references are not present

Response 1: Thank you for pointing this out. Thank you very much for your reminder, I'm very sorry for the reading inconvenience caused by my negligence, I've now changed all the references.

Comments 2: "Redmon et al. proposed" please add citation here

Response 2: Thank you for pointing this out. I have added the citation.

Comments 3: "evaluate the probability of each category8." what is a category8?

Response 3: Thank you for pointing this out. I apologize for the difficulty in reading the references because I did not leave the statements in the original text blank when citing them, and I have now made the change.

Comments 4: "one of the more practical versions available." what authors mean by "more practical versions available"?

Response 4: Thank you very much indeed for your comments. Since this article was started in January 2023, the most practical model at that time was YOLOv7, which outperformed YOLOv5. YOLOv8 had just been proposed, and its performance was not yet stable, so it was not considered. That's why I said "one of the more practical versions available."

Comments 5: Equation (1),(2),(3),(4),(5) - please introduce all symbols used in those two equations.

Response 5: Thank you for pointing this out. I have explained the symbols that appear in these formulas.

Comments 6: What is "frame14", "one15"?

Response 6: Thank you for pointing this out. I apologize for the difficulty in reading the references because I did not leave the statements in the original text blank when citing them, and I have now made the change.

Comments 7: Equation (8) - it is not clear why there are constants "3" and "6". How they were chosen?

Response 7: Thank you very much for your reminder. 3 is added to the input x. This is done by shifting the value field to the right to transform otherwise negative values into positive values, avoiding the ReLU6 function from setting some values directly to 0 at the output.The ReLU6 function is then used to ensure that the result is constrained to be in the range of 0 through 6. Finally, the output of ReLU6 is divided by 6 and multiplied by the original input x to ensure that the output of the h_swish function is consistent in scale with the original input.

Comments 8: "Dataset analysis" you mention several datasets, however none of them is https://data.csail.mit.edu/graphics/fivek/ which is reported in "Data Availability Statement".

Response 8: Thank you for pointing this out. I have updated the links to these datasets

4. Response to Comments on the Quality of English Language

Point 1: I am not qualified to assess the quality of English in this paper

Response 1: I am very sorry for the inconvenience of your reading, I have rewritten the article to improve the quality of the English language.

Response to Reviewer 3 Comments

1. Summary

Thank you for your suggestions on our manuscript entitled "CCBA-NMS-YD: A Vehicle Pedestrian Detection and Tracking Method Based on Improved YOLOv7 and DeepSort". All these suggestions are valuable and helpful for us to revise and improve the paper. We have carefully studied these suggestions and revised them, and we hope to get your approval.

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Can be improved

Is the research design appropriate?

Can be improved

Are the methods adequately described?

Can be improved

Are the results clearly presented?

Can be improved

Are the conclusions supported by the results?

Can be improved

3. Point-by-point response to Comments and Suggestions for Authors

Comments 1: The introduction could be strengthened by providing a more detailed discussion of the limitations of current pedestrian detection and tracking technologies, specifically how they fail in complex scenarios.

Response 1: Thank you for pointing this out. We have discussed in detail in the introduction the limitations of current pedestrian detection and tracking techniques, especially the shortcomings in complex scenarios.

Comments 2: It would be beneficial to include more detailed explanations of the improvements made to the NMS algorithm and the integration process of the MobileNetV3 module within DeepSort.

Response 2: Thank you very much indeed for your comments. We detail an improvement in the NMS algorithm that uses a linear decay formula instead of the exponential decay in Soft-NMS, allowing for a smoother and more controlled decrease in confidence scores and reducing the excessive suppression of scores at higher IoU values. In the meantime, For the integration process of the MobileNetV3 module in DeepSort, I have provided instructions in the article.

Comments 3: More related works could be reviewed such as "Modelling of destinations for data-driven pedestrian trajectory prediction in public buildings".

Response 3: Thank you very much for your reminder. This paper presents a novel destination-driven pedestrian trajectory prediction method, DDPTP, that emphasizes the importance of destinations in predicting pedestrian movements by identifying destinations that pedestrians are likely to travel to and training a specialized trajectory model for each destination to predict the future movements of pedestrians.At the core of DDPTP lies the decomposition of the pedestrian trajectory prediction problem into multiple destination-specific sub-problems, which simplifies model complexity and improves prediction accuracy. The core of DDPTP is the decomposition of the pedestrian trajectory prediction problem into multiple destination-specific sub-problems, which simplifies model complexity and improves prediction accuracy.DDPTP is able to automatically identify destinations through unsupervised learning techniques, enabling an end-to-end data-driven solution that not only reduces the reliance on manually labeled data, but also improves the model's resilience in unknown or dynamic environments.

Comments 4: There is a need for a more detailed discussion on the implications of the improved accuracy and efficiency. How do these improvements impact practical deployments in real-world scenarios?

Response 4: Thank you very much indeed for your comments. I have provided a more detailed discussion of the impacts of improved accuracy and efficiency, and the implications of those impacts for real-world practical deployments, in the concluding section.

Comments 5: What specific challenges in the existing YOLOv7 and DeepSort models prompted the development of the CCBA-NMS-YD algorithm?

Response 5: Thank you very much indeed for your comments. Although YOLOv7 has achieved great improvement in detection accuracy compared to other algorithms, it still faces challenges in some complex scenes, such as small-scale target detection, targets being occluded, changes in object viewpoints and poses, and poor lighting conditions, which are difficult to detect. For the tracking algorithm DeepSort, it may be difficult for the tracking algorithm to maintain accurate tracking of each pedestrian when the distance between targets is too close or when crossings and occlusions occur, and the tracking algorithm may not be able to update the tracking state in a timely manner if the target is moving quickly or changing direction in the scene, leading to tracking drift or loss of the target. In addition, algorithms such as YOLOv7 and DeepSort, although optimized for speed, are still unable to complete processing in a short period of time when dealing with high-resolution images or complex scenes.

Comments 6: How do the modifications in the NMS algorithm specifically contribute to reducing false positives in high-density scenarios?

Response 6: Thank you very much indeed for your comments. In high-density scenarios, the NMS may incorrectly suppress some detection frames that actually represent different targets, resulting in some targets not being detected. Figure 10 in the paper illustrates this scenario.

Comments 7: Please fix the misprints such as "Error! Reference source not found" in the manuscript.

Response 7:Thank you for pointing this out. Thank you very much for your reminder, I'm very sorry for the reading inconvenience caused by my negligence, I've now changed all the references.

4. Response to Comments on the Quality of English Language

Point 1: English language fine. No issues detected

Response 1: Thank you for your encouragement, I will check the English quality of the article again and hopefully make it a better reading experience for you.

1. Summary

Thank you for your suggestions on our manuscript entitled "CCBA-NMS-YD: A Vehicle Pedestrian Detection and Tracking Method Based on Improved YOLOv7 and DeepSort". We apologize that the last revision did not meet your requirements. We have carefully studied the suggestions and made changes that hopefully will meet with your approval.

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Must be improved

Is the research design appropriate?

Must be improved

Are the methods adequately described?

Must be improved

Are the results clearly presented?

Must be improved

Are the conclusions supported by the results?

Must be improved

3. Point-by-point response to Comments and Suggestions for Authors

Comments 1: there are still sentences starting with "Literature...."

Response 1: Thank you for pointing this out. I apologize again, and I have now corrected the above errors in accordance with the common writing style in English-language journals (lines 45, 60, and 65 of the article)

Comments 2: There are many typos which needs professional proofreading.

Response 2: Thank you very much indeed for your comments. I apologize for the inconvenience. With the help of our three English major colleagues, we have corrected the typos that appeared in the article.(Lines 36 on page 1, 179 on page 4, 204 on page 5, 342 on page 9, 410 on page 12, and 445 on page 13 of the article)

Comments 3: I suggest the author provide the response to review with line and page references to the updated manuscript to highlight the changes made during the revision.

Response 3: Thank you very much for your reminder.

4. Response to Comments on the Quality of English Language

Point 1: English very difficult to understand/incomprehensible.

Response 1: Thank you very much indeed for your comments. We corrected grammatical errors that existed in the article and changed some statements to make them more fluent. In addition, keywords were changed to lowercase. The language of this article is American English.(Lines 10, 31, 38, 159 on page 4, 220 on page 5, 387 on page 11, and 481 on page 15 in the article.)

1. Summary

Thank you for your approval and we are truly grateful for your dedicated work.

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Yes

Is the research design appropriate?

Yes

Are the methods adequately described?

Yes

Are the results clearly presented?

Yes

Are the conclusions supported by the results?

Yes

Response to Reviewer 1 Comments

1. Summary

Thank you for your suggestions regarding the paper. We have carefully studied your suggestions and the referenced papers , and have made changes based on them, which we hope will meet with your approval.

2. Questions for General Evaluation

Reviewer’s Evaluation

Response and Revisions

Does the introduction provide sufficient background and include all relevant references?

Can be improved

Is the research design appropriate?

Can be improved

Are the methods adequately described?

Can be improved

Are the results clearly presented?

Can be improved

Are the conclusions supported by the results?

Can be improved

3. Point-by-point response to Comments and Suggestions for Authors

Comments 1: In the sentence "traditional target detection algorithms and target detection algorithms based on deep learning", what exactly does traditional algorithm refer to, is it referring to traditional ML (better to name that as Traditional ML) or Image processing?

Response 1: Thank you very much for your reminder. It refers to traditional machine learning, which I have modified in my paper (line 39).

Comments 2: Regarding hand-designed feature extraction methods in image processing, more clarification/information is needed. You may refer to "Fusion of multi-scale bag of deep visual words features of chest X-ray images to detect COVID-19 infection" for reference. 

Response 2: Thank you very much indeed for your comments. I have carefully read and cited this paper(line 45).

Comments 3: Abbreviation "MOTDT", "CCBA" ,"SAN" and so on in first place without spelling it out.

Response 3: Thank you very much for your reminder. I have revised it in thesis(Line 15, page 2 line 91, page 13 line 448).

Comments 4: Please introduce how attention is the major contribution here for detection. You may refer to " Fruit classification using attention-based MobileNetV2 for industrial applications" for how attention can contribute to image recognition.

Response 4: Thank you very much indeed for your comments. The contribution of attention mechanisms has been added to the paper on page 8, line 314.

Comments 5: rewrite the deep sort section in a standard algorithmic way.

Response 5: Thank you for pointing this out. I've added an introduction to the DeepSort pseudo-code(Listing 1, Line 200 on page 6).

Comments 6: The ablation study needs more clarification on which part of a network was improved with the given block such as CCBA, or SAN or any other.

Response 6: Thank you for pointing this out. I have indicated on page 13, line 450 of the paper which part of the network is to be used with the given block.