Unsupervised Deep Embedded Clustering for High-Dimensional Visual Features of Fashion Images

Round 1

Reviewer 1 Report

quick google search pointed to https://link.springer.com/content/pdf/10.1007/978-3-030-21451-7.pdf pages 85-96 paper "Unsupervised Deep Clustering for Fashion Images" by some overlapping author names, images, algorithm description, result tables, and sentences (another link to same paper https://www.researchgate.net/publication/333716329_Unsupervised_Deep_Clustering_for_Fashion_Images). The paper under review cites that 2019 conference paper

Author Response

We confirm that the references mentioned in comments belong to our previous work under the same project. Our previous study focused on enhancing the joint learning approach, which motivated us to continue exploring this approach through further experiments. Therefore, the current work is advanced research with more effective novelty and contribution. We have performed further experiments (shown in table 4 and figure 8) to evaluate the performance of our method with additional baseline methods. Similarly, we added new datasets Fashion-MNIST for comparison.

In contrast to previous work, our current work's methodology and techniques have improved the overall model's performance. Meanwhile, the presentation of the work has been updated based on feedback from the previous work. Additionally, we confirm that there is no crucial overlapping with our prior work.

Author Response File: Author Response.docx

Reviewer 2 Report

In this paper, the authors present an unsupervised clustering approach developed for fashion images. The authors base their clustering on a convolutional neural network architecture. They should clarify the novelty and originality of their contributions. Moreover, they should improve some technical aspects of the paper. More detailed comments follow next.

1. The authors should clarify the technical aspects of their original contribution. In this sense, they should clearly state which technical novelties contain their work, compared with competing approaches in this area. The authors should clearly state which novel technical features contain the proposed clustering approach compared to other clustering approaches and neural network architectures. What is specific about the proposed architecture?
2. The mathematical notation used in the paper should be considerably revised. To this aim, the authors should use conventional notation, explaining in detail the dimensions of each variable, and carefully introducing each variable and function appearing in their equations.
3. Regarding the optimization step, the authors should elaborate on how different optimization approaches influence the results.
4. Why do the authors consider a dataset with handwritten digits besides the Amazon-fashion dataset? Does this imply that their method outperforms the state-of-the-art even for non-fashion image datasets? Please elaborate.
5. There are many errors in punctuation around equations, errors in line 339, etc. Please, double-check the paper.

Author Response

Word file attached

Author Response File: Author Response.docx

Reviewer 3 Report

The paper proposes a fashion images-oriented deep clustering (FIDC) method utilizing 4096-dimensional feature vectors, followed by applying a DL-based dimension reduction (DR) tool. I have the following CRITICAL comments regarding this paper before further review:

----------------------------------------------Specific Comments -----------------------------------------------------

1- What is new in your technique? Is it utilizing the technique for fashion images? Please clarify.

2- Author(s) try to do DR using the DL-based technique. However, I think that adding this block also can increase the whole computational complexity.

3- The introduction section is not complete. Authors could discuss the literature in this field more extensively. It should be MODIFIED in the new version.

4- what is CNN-F? what is ?̃ in Line 166? You didn’t define this parameter, but how you are discussing it?

5- Authors should study the generalization capability of their method.

6- The sensitivity of the proposed approach to the parameter settings should be evaluated.

7- Authors should discuss the computational complexity of the proposed approach.

8- What is the limitation of the proposed methods? The authors should critically discuss the results.

-------------------------------------------Secondary Comments----------------------------------------------------------------

- I think reporting figure in the introduction section is not common.

- The writing of the paper is not acceptable. Some sentences are not fluent and confusing. The paper has many grammatical issues.

- Line 155: ‘’fashion images-oriented deep clustering FIDC method ….’’. You have provided the ’fashion images-oriented deep clustering as (FIDC) before, but here you brought both one after another. Please check your paper’s writing before sending it to one journal.

- The figures of poor image quality.

- The paper suffers from different issues regarding the notations. For example, Matrix and Vectors should be in boldface.

- Can you see your legends in Figure 11? It also does not have any axes’ names. I can see too many issues throughout the whole paper. It is not the reviewer’s task to tell the authors of these kinds of mistakes.

- ensure that you have defined the total parameters in Equations (14-16). 

Author Response

Word file attached

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The authors have corrected the paper following my review. I have no further comments or suggestions for improvements.

Author Response

please fine attach file

Author Response File: Author Response.docx

Reviewer 3 Report

Thanks for addressing my comments. However, still the paper needs further improvements including: 

 - I can see some grammatical issues. Please double  check. 

- The quality of figures is not acceptable. 

- The pseudocode is not well written. The predefined commands should be in boldface. 

- The best values in Table 4 should be presented in boldface. 

Author Response

please find attach file.

Author Response File: Author Response.docx