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.
Thank you for your revised work, entitled "Fine-art recognition using convolutional transformers", in PeerJ Computer Science. Authors have shown great effort in addressing reviewer's concerns about this work.
[# PeerJ Staff Note - this decision was reviewed and approved by Xiangjie Kong, a PeerJ Section Editor covering this Section #]
The authors have addressed all of my concern. Therefore, I recommend to accept this paper.
No comment.
No comment.
No comment.
The revised manuscript addresed all my concerns. The language is professonal. The manuscript work is fine with good structures. I’m pleased to recommend this work is now ready for publication.
Authors conducted many additional experiments to support their conclusion. All tables and figures are well-presented.
Statistical evidence supported the validity of their findings. Model performance across multiple trials varies in a small range, confirming the model’s stability.
Thank you for your recent summited work, entitled "Fine-art recognition using convolutional transformers", in PeerJ Computer Science. It has been examined by expert reviewers whose reviews are enclosed. The reviewers have expressed serious concerns about this work that make it unsuitable for publication in its current form.
However, if you feel that you can correct these problems, we would be happy to consider a revised manuscript as a new submission (assuming no related data is published in the interim).
[# PeerJ Staff Note: It is PeerJ policy that additional references suggested during the peer-review process should only be included if the authors are in agreement that they are relevant and useful #]
no comment
Technical novelty and soundness: the authors should clearly state the difference between their proposed method and ViT [1] to support their claim about “proposed model architecture”. Because from Figure 3, it is similar to the original ViT paper.
[1] Dosovitskiy, A.; Beyer, L.; Kolesnikov, A.; Weissenborn, D.; Zhai, X.; Unterthiner, T.; Gelly, S. An Image is Worth 16 × 16 Words:Transformers for Image Recognition at Scale. In Proceedings of the ICLR 2021: The Ninth International Conference on LearningRepresentations, Vienna, Austria, 3–7 May 2021.
• Experiment:
o The authors should report the avg results of the proposed method from Table 2 to Table 1, instead of choosing the best performance. Then, for other methods, the authors should at least run 5-10 times or even same as what they have done in Table 2 for other methods. Then, it will be more reasonable and persuasive.
o The authors should consider evaluating the proposed method on another dataset, such as WikiArt Dataset [2].
• Writing and presentation: I suggest the Table 2 should be converted to a plot, which will make it clearer and easier to interpret information.
[2] W. R. Tan, C. S. Chan, H. E. Aguirre and K. Tanaka, "Improved ArtGAN for Conditional Synthesis of Natural Image and Artwork," in IEEE Transactions on Image Processing, vol. 28, no. 1, pp. 394-409, Jan. 2019.
This work proposed using convolutional transformers to tackle fine-art recognition problem. The background about digital arts and relevant studies is sufficient. In general, this work is fine with good structures and language used. To be considered for publication, this work needs to futher improved.
The proposed method is a combination of a residual network, convolution, and a transformer, specified by an attention mechanism. The model was benchmarked against traditional transfer learning models. The proposed model was demonstrated to have higher performance compared to state-of-the-art models.
There are several issues below that need revisions:
- All of these models for benchmarking are transfer-learning-based. After extracting features from pre-trained models, why didn’t authors use conventional machine learning algorithm to develop models. If not, please conduct these experiments to compare with them.
- On which criteria did authors select the attention mechanism? Randomly or by any preliminary research?
- Did residual shortcut connection really help to improve the model performance? Please prove that by experiments
- What is the shortcoming of the proposed method?
- Grid-background of the Figure 2 should be removed to improve visualization.
- Adam optimizer (line 205) needs a citation.
Authors provided sufficient evidence to support the validity of their findings. Small variations in measured metrics confirms the model’s stability.
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.