Deep Descriptor Learning with Auxiliary Classification Loss for Retrieving Images of Silk Fabrics in the Context of Preserving European Silk Heritage
Round 1
Reviewer 1 Report
This manuscript proposes an approach for descriptor learning for retrieving images of silk fabrics. The paper is well-organized and easy-to-read. Some findings of the authors look interesting, e.g. an auxiliary classification loss. There are some incremental improvements in comparison with the previous works of the authors in the scope of EU SILKNOW project. In general, it worth to be published to my mind.
Nevertheless, this paper has one significant shortcoming: an aim of the investigation is declared as image retrieval, but there is no results demonstrated performance of image retrieval directly; the paper shows outcomes of the classification on semantic classes, that serves as indirect measure of an effectiveness of the proposed approach. Authors try to explain this fact by complexities of manual labeling and experiments with evaluation results by experts. However, at least some qualitative results/examples of image retrieval based on proposed descriptor should be done. Personally I do not insist on obtaining of new more relevant experimental results, however it will not surprise me if other reviewers request major revision.
Also, there are several small issues:
- "constrastive" instead of "contrastive" in line 130.
- seems, something wrong in the phrase "whereas the in our work" in line 171.
- there is more valid reference for ILSVRC-2012-CLS dataset instead of [60], please see https://image-net.org/challenges/LSVRC/2012/#cite
- double word "equation" and "respectively" in Italic in the capture of Table 2.
- knn and kNN are used; it is preferable to use identical terms.
- seems "Insurprisingly" in line 685 should be changed to "Unsurprisingly".
Author Response
Thank you very much for your review. Please see the attachment for details.
Author Response File: Author Response.pdf
Reviewer 2 Report
The article is interesting and focused on the problem of retrieving images to search for records in a database of silk fabrics. We evaluate the approach on a dataset of tissue images in a kNN classification, showing promising results, demonstrated also to other digital collections.
Nevertheless, the paper is not ready for the publication in its present forms. I recommend the authors to rewrite the papers addressing at least the main issues, listed below:
(i) adequately describe the possible applicative repercussions of the research at an economic, technical, functional, cultural and symbolic level.
For example, what can be the economic advantages of industries and in the world of the creative and design industry.
Also mention how issues of cultural heritage strengthen the dialogue between different cultures.
And not only will they make it possible to study and preserve Europe's cultural heritage, but for example, they will open up new research in other fields. Indicate any new application fields of the research.
(ii) Better describe the contents of Art 500k, OmniaArt
(iii) Improve the bibliography
I recommend the authors to address these points and re-submit the paper.
Author Response
Thank you very much for your review. Please see the attachment for details.
Author Response File: Author Response.pdf
Reviewer 3 Report
The manuscript presents a novel approach of image retrieval using deep learning approach. The importance and novelty of this study consists in technical evaluating of the two datasets with pictures of silk using SILKNOW project of the European Initiative and WikiArt dataset. The paper excellently combines machine learning, image classification, pattern recognition and cultural heritage approaches of AI (deep learning) which presents a multi-disciplinary study well deserved to be published.
I summarized my detailed review in the report attached and recommend to publish the manuscript in the present form. I did not find any errors or mistakes that would require corrections. Therefore, I recommend to accept the paper and proceed with final editing for publication.
Comments for author File: Comments.pdf
Author Response
Thank you very much for your review and for your very positive assessment.