DFE-YOLO: A Multi-Scale-Enhanced Detection Network for Dense Object Detection in Traffic Monitoring

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Dear authors,

Here are some comments that will guide you in manuscript imporvement:

1. Table 4: In the last column, you bold the wrong number. The bold font should be for CenterNet. CenterNet is better by APsmall. Please try to comment and argument is it important for your comparative analysis.
2. The title leads to expectancy of traffic analysis. However, there are no mention of "Complex Traffic Monitoring Scenarios". Hence, the title is misleading. You should change the title or to introduce real traffic scenarios analysis in the manuscript.
3. What do you mean with "traffic object detection scenarios"? What is the difference between object detection and traffic object detection in the context of what you presented in the paper?
4. Can you train your network to detect other objects, such as ships?
5. What is minimum and maximum resolution for the reliable operation of the developed ANN?
6. What are limitations of the proposed network?
7. References style and data are not according to the instructions to the authors.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This manuscript proposes DFE-YOLO, a modified object detection framework based on YOLOv8, targeting improved accuracy in dense, small-object, multi-scale traffic monitoring scenarios. There are some major concerns that need to be addressed before this work can be deemed publishable.

 

1. Insufficient Novelty Beyond Engineering Integration

While the paper introduces a compelling combination of FASFF, DySample, and EIoU loss, these components are incremental improvements or adaptations of existing ideas (e.g., ASFF, dynamic sampling, IoU variants). There is limited theoretical novelty or innovation in architecture design.

 

2. Limited Real-Time Performance Evaluation

The method is intended for traffic monitoring, yet no inference speed (FPS), model size, or computational cost (FLOPs) is reported. This makes it difficult to assess deployability on edge devices, which is critical for traffic applications.

 

3. Ablation Study Scope is Too Narrow

The ablation study focuses only on the three proposed modules, but does not include comparisons:

• With other feature fusion strategies (e.g., BiFPN, PANet variants)

• Between different upsampling methods (bilinear, transposed conv, CARAFE, etc.)

• With other IoU losses under different scene conditions (e.g., SOTA regression losses like SIoU/WIoU in traffic vs. non-traffic)

 

4. Clarity and Writing Issues

The manuscript suffers from some grammatical and structural issues, with frequent awkward phrasing and inconsistent terminology. Please perform a through language revision. Below are a few examples:

• “objection” instead of “object detection,”

• “dySample” vs. “DySample”

• “In the A4 experiment, the integration of…” There was no prior mention of “A4.”

• “The Dysample sampling strategy collaborates with FASFF, resulting in…” “Collaborates with” is awkward for describing algorithm components. Use something like “combined with” instead.

Comments on the Quality of English Language

See the main letter.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

Integrate answer to my comment no. 5 in the manuscript text (not all, but table and some conclusion would be fine).

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Thank you for the revision and response. The reviewer’s previous comments have been addressed.

Author Response

Comment1: Thank you for the revision and response. The reviewer’s previous comments have been addressed.
Response1: We sincerely appreciate the reviewer’s recognition of our efforts in revising the manuscript. Thank you for your constructive feedback during the review process, which has significantly improved the quality of this work.