YOLOv8-CB: Dense Pedestrian Detection Algorithm Based on In-Vehicle Camera

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

In this manuscript, the authors present YOLOv8-CB, using CFNet and a five-layer CBAM attention mechanism which improves multi-scale feature extraction and detection accuracy by 2.4%. These innovations have practical potential for precise object recognition. The proposed model not only boosts accuracy but also improves efficiency, with 0.5MB parameter reduction, 0.7GFLOPs reduction, and a 10.8ms inference time. It excels in high-density pedestrian areas, but future work should focus on occluded small target detection. 

Overall I think the article is great. I would recommend the authors to expand the introduction beyond 1 paragraph and fix all the reference errors for "figure not found". 

Author Response

Thank you for pointing this out. We agree with this comment. Therefore, we have expanded paragraphs 3, 4 and 5 of the introduction to make them well researched and the references cited relevant to the paper. All references have been properly cited and placed before punctuation, and legends and tables have been properly cited after the first citation in the text. All revisions have been updated into the manuscript for your review.

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript presents an improved lightweight multi-scale pedestrian detection algorithm called YOLOv8-CB. This algorithm aims to address challenges in dense pedestrian detection at intersections, such as high computational weight, complex models, and suboptimal accuracy for small and occluded pedestrians. I have some concerns that have to be addressed before this manuscript can be accepted.

1. Please highlight your contributions, including what improvements you brought to YOLOv8. Clarify which components of the networks are introduced by the authors.

2. I would like to see more results on different datasets. The current dataset is not enough to depict the performance.

3. More experiments on vehicle-embedded GPU are needed, e.g. Nvidia Jetson platform.

4. Figures 7 and 8 are blurred.

5. There are some obvious errors caused by the wrong reference of Figures.

 

 

Comments on the Quality of English Language

This manuscript is well-written with a clear structure and language skills.

Author Response

Comments 1: Please highlight your contributions, including what improvements you brought to YOLOv8. Clarify which components of the networks are introduced by the authors.

Response 1: Thank you for pointing this out. We agree with this comment. Therefore, we have expanded the introduction in paragraphs 3, 4, and 5 to make the research well researched and the references cited relevant to the paper. In the last paragraph of the introduction, I emphasize the improvements of the YOLOv8-CB algorithm, including the introduction of a lightweight cascaded feature fusion network (CFNet), replacing the traditional C2F block in the backbone network with an advanced Focal-NeXtF block, and integrating a four-layer CBAM channel spatial attention mechanism into the detection header, which enables the model to pay more attention to useful feature channels. A four-layer CBAM channel spatial attention mechanism is integrated in the detection header to enable the model to pay more attention to the useful feature channels. In the feature fusion section, a bidirectional weighted feature fusion structure (BIFPN) is superimposed to significantly improve the detection of pedestrians obscured by dense occlusions. Please refer to the last section of the introduction for detailed modifications.

Comments 2:I would like to see more results on different datasets. The current dataset is not enough to depict the performance.

Response 2: Agree. We have, accordingly modified the content in 3.2Comparative analysis by adding in the first part of 3.2 the experimental results of the model on the VisDrone and CrowdHuman datasets, which are authoritative in the detection set of small and occluded targets, and in this paper, we use the SSD (VGG), YOLOv3- tiny, YOLOv4-tiny, YOLOv5s, YOLOv7-tiny, YOLOv8n and YOLOv8-CB algorithms on each of the three datasets and the detection accuracy results obtained confirm the effectiveness of the improved algorithms, and all changes have been made to emphasize this point in 3.2.

Comments 3:More experiments on vehicle-embedded GPU are needed, e.g. Nvidia Jetson platform.

Response 3: Agree. Thank you very much for pointing out this revision, which is of great instructive significance and value for the subsequent research, while I will be more in-depth research to do more experiments on top of the embedded device e.g. Nvidia Jetson platform, but due to the fact that the current project team is unable to equip the hardware experimental conditions within 10 days due to insufficient funds, it is not possible to satisfy the need for deploying more Please forgive us.

Comments 4:Figures 7 and 8 are blurred.

Response 4: Agree. We have, accordingly modified the typographical layout of figures 7 and 8 and the font sizes of the titles and legends of the horizontal and vertical axes of the icons and of the figures to make them clearer and more readable, and all the modifications are shown in figures 7 and 8 in the main text to emphasize this point.

Comments 5:There are some obvious errors caused by the wrong reference of Figures.

Response 5: Agree. We have, accordingly modified revised all sections of the text that require references to be cited in accordance with the official dissertation reference template, all references have been correctly cited and placed before punctuation, and legends and tables have been correctly cited after the first citation in the text. All revisions have been updated into the manuscript for your review.

Reviewer 3 Report

Comments and Suggestions for Authors

The paper topic is of great importance for the visual automation and recognition of moving objects in traffic applications. Despite the importance of the content, there are some minor aspects that need to be corrected, namely:

- Please check all references to the images! Some references are missing or not correctly referenced. This is a minor error, but requires a thorough review of the article content.

- Some images should be redesigned so that they fit well into the article template. For example, Fig. 1 and similar are quite large, but a better look could be achieved by moving the function blocks.

Other aspects are good and understandable. Good work!

Author Response

Comments 1: Please check all references to the images! Some references are missing or not correctly referenced. This is a minor error, but requires a thorough review of the article content.

Response 1: Thank you for pointing this out. We agree with this comment. Therefore, we have revised all sections of the text that require references to be cited in accordance with the official dissertation reference template, all references have been correctly cited and placed before punctuation, and legends and tables have been correctly cited after the first citation in the text. All revisions have been updated into the manuscript for your review.

Comments 2: Some images should be redesigned so that they fit well into the article template. For example, Fig. 1 and similar are quite large, but a better look could be achieved by moving the function blocks.

Response 2: Agree. We have, accordingly, modified the layout of the relevant function blocks in Figures 1, 2, 4 and 6 to make them more reader-friendly by moving and adjusting the function blocks, as well as adjusting the position of the article's legends, formulas, and tables to be more in line with the thesis template requirements in order to emphasize this point.

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

I have no further concerns about this manuscript.

Comments on the Quality of English Language

The English presentation of this manuscript is adequate.