All reviews of published articles are made public. This includes manuscript files, peer review comments, author rebuttals and revised materials. Note: This was optional for articles submitted before 13 February 2023.
Peer reviewers are encouraged (but not required) to provide their names to the authors when submitting their peer review. If they agree to provide their name, then their personal profile page will reflect a public acknowledgment that they performed a review (even if the article is rejected). If the article is accepted, then reviewers who provided their name will be associated with the article itself.
Thanks for addressing all the comments and this version is ready to be published. I have reviewed your response and found it appropriate.
[# PeerJ Staff Note - this decision was reviewed and approved by Xiangjie Kong, a PeerJ Section Editor covering this Section #]
Please address the comments as highlighted in the reviewers comments in the next version. Thanks
**PeerJ Staff Note:** Please ensure that all review, editorial, and staff comments are addressed in a response letter and that any edits or clarifications mentioned in the letter are also inserted into the revised manuscript where appropriate.
**Language Note:** PeerJ staff have identified that the English language needs to be improved. When you prepare your next revision, please either (i) have a colleague who is proficient in English and familiar with the subject matter review your manuscript, or (ii) contact a professional editing service to review your manuscript. PeerJ can provide language editing services - you can contact us at copyediting@peerj.com for pricing (be sure to provide your manuscript number and title). – PeerJ Staff
This paper introduces a road scene detection algorithm named SDG-YOLOv5, which addresses issues related to low detection accuracy and poor real-time performance. The algorithm integrates SIoULoss, Decoupled Heads (DHs), and Global Attention Mechanism Group Convolution (GAMGC) into the YOLOv5 network structure. Experimental analysis conducted on Udacity Self Driving Car, BDD100K, and KITTI datasets show promising improvements, with the algorithm achieving mAP@.5 increases of 2.2%, 3.4%, and 1.0%, respectively, compared to the original YOLOv5, while maintaining a detection speed of 30.3 FPS.
Strengths:
- The integration of SIoULoss significantly improves the regression and convergence speed of the prediction bounding boxes, enhancing detection accuracy for categories with fewer labels.
- The incorporation of Decoupled Heads (DHs) leads to notable increases in recall rate and mAP@.5, while also reducing the number of parameters and floating-point operations, which enhances computational efficiency.
Weaknesses:
Although this work has a very attractive starting point, it also has some obvious limitations：
- The writing and presentation should be improved. First, the authors should elaborate more on the underlying intuition or motivation behind the SDG-YOLOv5.
- Why use YOLOV5? Why not the newer YOLO-V8?
- The references in this article do not conform to the standard format, with some formally published works being cited as arXiv versions.
- The compared methods are very old. The authors will need to consider recent methods.
- The initial loss function, Complete-IoU Loss (CIoU Loss), was found to converge slowly and degrade to Distance-IoU Loss (DIoU Loss) under certain conditions, potentially affecting convergence speed and accuracy.
- The detection head in the original network couples the classification and regression tasks, which may compromise the accuracy of each task.
- The algorithm utilizes more CBS and C3 structures, leading to a large number of redundant feature maps without adequate focus on important features within the network.
- While the integration of attention mechanisms such as GAM, GAMGC, and GAMDW enhances accuracy, each method comes with its trade-offs in terms of parameter count, floating-point operations, and frame rate, potentially complicating network optimization.
- Although GAMGC demonstrates notable precision and frame rate improvements with fewer parameters, its performance in combination with SIoULoss and the improved decoupled head is slightly lower compared to GAMGC alone, indicating potential compatibility issues.
N/A
N/A
N/A
No Comment
Although the topic seems valuable, this paper has several potential gaps. This paper listed a few selected approaches, but they should have discussed how they selected those methods for their datasets. Authors should explain the reason behind choosing these methods.
Authors have described the results but also have to discuss validation techniques to validate their numbers. The paper is trying to address the critical issue, but it is vital to justify its approach and the results.
Instead of discussing the literature survey papers in the introduction, authors can create a separate section for the research work they have reviewed.
All text and materials provided via this peer-review history page are made available under a Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.