An Improved YOLOv5s Model for Building Detection
, Jing Cao 1, Yutai Gu 1, Yuanze Wu 1, Chong Chen 1,* and Yingying Wang 2Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsIn the manuscript the authors proposed a lightweight building detection model based on YOLOv5s. Data augmenting and pruning are conducted on the YOLOv5s. After pruning, the author discussed the influences of different activation functions. However, in my opinion, several key issues should be addressed before the consideration of publication.
(1) Grammar problems should be checked carefully;
(2) The author should try to explain the reason why activation functions have such effects on the results;
(3) The author should try to explain how to conduct the pruning and why conduct the pruning this way, may consider specific features of the dataset or any other aspects.
(4) This is a special issue of "Advances in Autonomous Vehicle: Motion Planning, Trajectory Prediction and Control", so the author should explore how does the proposed model works in autonomous vehicle, because as I can see, building dection in this manuscript is detecting buildings from images, how about in videos?
Comments on the Quality of English LanguageGrammar problems should be checked carefully.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThis paper proposes a lightweight building recognition model based on YOLOv5s, which improves recognition efficiency through enhanced data augmentation and model pruning techniques, achieving higher accuracy.
1.In the abstract (particularly in comparative experiments), model parameter sizes are typically described in MB. For precision in describing detection accuracy, it is recommended to replace "0.823" with "82.3%".
2.The authors should supplement additional statistical information regarding the proposed dataset, such as image size, the number of target boxes, and the results of normalizing target boxes in terms of length and width.
3.Combining data collection and dataset creation, it is essential to clearly outline the specific practical application background and significance of the task in the introduction.
4.The literature review on lightweight methods in the introduction is not sufficiently comprehensive. For instance, mainstream lightweight studies such as MobileNetV3 and GhostNet are not addressed.
Recommended references:
Sure, here are some recommended reference articles related to lightweight YOLOv detection method:
1). Guo Y, Chen S, Zhan R, et al. LMSD-YOLO: A Lightweight YOLO Algorithm for Multi-Scale SAR Ship Detection[J]. Remote Sensing, 2022, 14(19): 4801. DOI：https://doi.org/10.3390/rs14194801.
2).Xu K, Zhang H, Li Y, et al. An ultra-low power tinyml system for real-time visual processing at edge[J]. IEEE Transactions on Circuits and Systems II: Express Briefs, 2023. DOI：https://doi.org/10.1109/TCSII.2023.3239044.
3).Dang C, Wang Z, He Y, et al. The Accelerated Inference of a Novel Optimized YOLOv5-LITE on Low-Power Devices for Railway Track Damage Detection[J]. IEEE Access, 2023, 11: 134846-134865. DOI：https://doi.org/10.1109/ACCESS.2023.3334973.
5.While the authors conducted numerous comparative experiments, but did not analyze the optimization results of bn layer weights in sparse training. Please supplement the weight changes of the batch normalization (BN) layer during the training process.
Comments on the Quality of English LanguageThe overall narrative flows smoothly, but there are still some grammatical issues that need to be addressed.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsDear authors,
Your proposed paper may be an interesting one, but you have some issues to work on.
1. Be careful when using abbreviations! Even familiar terms should be explained when they first appear in the text.
2. Pay attention to the Bibliography - you have items older than 5 years!!! Out of 34 items, more than half are older than 5 years.
3. After redoing the Bibliography, you must redo the introduction to match the bibliographic items.
4. From the abstract, your contribution appears to generate an improved GridMask data augmentation method - but in the paper, you have not highlighted your proposed method.
5. Beware of typos or errors of expression, for example:
- line 31, page 1 - "In the realm of autonomous driving [1-3]"
- line 37, page 1: "Recent years, people have conducted extensive research on building recognition."
6. line 123, page 3, says: "Additionally, we use the Labeling tool for image annotation." But then I didn't find in the paper explaining this labelling tool
7. In "Material and method," you discuss GridMask (from the literature), but your proposed modifications are not highlighted.
8. In Chapter 2, you describe a general theory or algorithm taken from the literature. I think it would be a good idea to introduce your own issues and the adaptations of the algorithms so that the reader can understand what you have done and the differences with the algorithms proposed in the literature.
9. Also in chapter 2 it would be good to introduce a flowchart of the workflow.
10. In chapter 3, say: "We optimized model parameters using Adaptive Moment Estimation (Adam), with 278 an initial learning rate of 0.001 and a final learning rate of 0.0001". Which model, yours or the one taken from the literature?
11. Try to rewrite the interpretation of the results because they seem to be results taken "from somewhere". Unfortunately, it is not understood that they are your own issues.
Please make these changes!
It would be a pity to lose so much work!
I wish you good work and look forward to seeing the material in a better form!
Comments on the Quality of English Language
Minor editing of English language required
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
Comments and Suggestions for AuthorsAccept in present form
