Reverse Knowledge Distillation with Two Teachers for Industrial Defect Detection
Round 1
Reviewer 1 Report
Good day, Dears! Recommended for publication! the study contains formulas. calculations, tables, figures!
Author Response
Thank you for reviewing our work. Best wishes to you.
Reviewer 2 Report
The paper looks interesting but needs to be improved.
1. More technical details with novelty are suggested
2. TPR, FPR are only performance criteria considered, however researchers are of the opinion that these may provide biasness, justify. Discuss with any other such criteria for soundness of the algorithm proposed
3. The authors are suggested to go for statistical significance test for validation of the results
4. Comparison with other works in Table should state the ref. Paper number for better understanding of the readers.
5. Use proper grammer checking with plagiarism check for the paper
Author Response
Thank you for reviewing our work. Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
The paper proposes a new method for industrial defect detection using reverse knowledge distillation with two teachers. The method employs an attention mechanism and iterative attentional feature fusion to improve anomaly scores at both image and pixel levels. The proposed method is evaluated on several datasets and shows promising results.
Proposing a new method for industrial defect detection using reverse knowledge distillation with two teachers.
Employing an attention mechanism and iterative attentional feature fusion to improve anomaly scores at both image and pixel levels.
Designing a student network to prevent overfitting.
Conducting extensive experiments on Mvtec and BTAD datasets to evaluate the proposed method.
The paper does not compare the proposed method with other state-of-the-art methods for industrial defect detection. Including such a comparison could provide a better understanding of the proposed method's strengths and weaknesses.
I think it would be nice for the authors to further improve the conclusion of this paper.
Unless the authors will disclose the code implemented in this paper, sufficient information should be included to reproduce the proposed method. (e.g., hyperparamter, etc.)
Author Response
Thank you for reviewing our work. Please see the attachment.
Author Response File: Author Response.docx