Baru-Net: Surface Defects Detection of Highly Reflective Chrome-Plated Appearance Parts
Round 1
Reviewer 1 Report
In this paper, the authors propose an approach for detecting surface defects of highly reflective chrome-plated appearance parts. They present Baru-Net, which is based on Unet with the addition of a CBAM module and an ASPP module. Thanks to the authors for their research, but this paper raises some questions that need to be clarified:
-
Why is the network called Baru? I would like to see an explanation for the choice of name.
-
The applicability of the study is not clear. The authors establish a limitation to the illumination scheme. And the images presented by the authors are not difficult to detect. The defects are clearly visible and can be found using OpenCV. If Baru-Net only works for images like this, why can't authors just use OpenCV? I would like to see an explanation. It would also be great if the authors add examples of detection in difficult conditions, where OpenCV or classic object detection models can not cope. In general, the advantages of Baru-Net with the proposed input data are not clear.
-
The procedure for creating artificial data is not entirely clear. Are there any limitations to randomness? What combinations of transformation functions are used? How is the defect inclusion area selected? How is the defect type selected and are there combinations of them? Or do all transformations occur under conditions of uncontrolled randomness? I would like to see more detailed material.
-
There is no discussion section. Does Baru-Net have limitations? What are the main points of your work? Also would like to know about further plans of the authors, will the current research be continued?
-
The conclusions are primitive. It is worth expanding the Conclusion. What is the novelty of the study or the authors' network? What were the main findings? The applicability of the authors' study?
Minor editing of English language required
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
This paper claims to propose a novel approach to improve prediction accuracy and deal with beam light and measured surface that extracts features from time-series workload data, which improves the forecasting accuracy. The problem is of interest. However, the reviewer finds that paper does not contribute enough to improve the existing literature. It simply suggests to include artificial detect image models such as CNN. Already there are papers which advocate of using such features. Moreover, the results are not enough and the authors are failed in justifying the improvements they are getting. Therefore, the reviewer feels that the paper does not qualify the criteria of getting recommendations for publications (at least in its current form).
I suggest that the authors try to develop other image prediction model such as SSD, Yolov8, .. and that compare with their proposed model. Also the dataset should be described in the paper. Quality of figures must be improved. I propose to detail the novelty in the proposed approach
some errors must be corrected
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
It is recommended for publication after review by an editor.
Minor editing of English language required
Reviewer 2 Report
The authors have well improved their work, all suggestion have been taken on consideration. I think that the paper can be accepted in its present format.
