Cinematographic Shot Classification with Deep Ensemble Learning

Round 1

Reviewer 1 Report

This work is very similar to the paper [8] already published by the same authors. Although, [8] is published in the the proceedings of a conference, all the methodology is presented there is the same applied to a slightly different  problem. I do not think the results described in are new enough to deserve to be published in the journal. Besides, comparison with the existing state-of-the-art methods is catastrophically insufficient to see if the proposed (previously published method) brings an added value.    

Also, in my opinion, the topic is extremely distant from Electronics -- the main subject of the journal, so I do not find the paper suitable for publication in Electronics.

Author Response

The answers to the reviewer's suggestions are in the attached file.

Author Response File: Author Response.pdf

Reviewer 2 Report

The author presents a strategy to classify film shots based on Deep Learning VGG-16 and Ensemble Learning. The work looks relevant to the literature but should be severely improved to be published.

References

1. The literature review is very outdated. 10 of 16 references have more than 5-years old. 

Abstract

2. The author should add some significant results in the abstract. During the manuscript, many comparisons are made. What are the best models? What is their performance? Accuracy? F1? ...

Introduction

3. The author presents a short motivation and context introduction but is not supported. References and clarifications should be added to the text.

4. "Images and videos belong to different cinematographic shots according to..." - which are? Can the authors add some figures to clarify?

5. lines 17-18, which are these advantages? Can the authors add some relevant references?

6. lines 24-25, some references, please?

7. line 30, Figure -> Figure 1

8. line 30 "We gathered 10,545 images divided into 8 classes" -> This information looks to be the characterisation of the dataset. This is not relevant for this moment, but for the materials and methods - Dataset.

In this part of the section, the authors get into some details that may confuse the reader. An in-depth review is required to simplify the text and make it more transparent.

9. line36 "stacking learning technique" - what is this? Reference or place where is described in the text. 

Related work

10. The literature review looks outdated. Many references are old and do not characterise the work made. 

11. The authors briefly review previous work for approaching the problem but do not review the techniques. Why did the authors use deep learning? What may be the advantages against other strategies? 

12. What is ensemble learning? What are the main techniques? Why did the authors used ensemble learning and stacking learning technique? Other works that support this use.  

Dataset Composition

13. line 80 -> Some acronyms and abbreviatures are not well identified. Review all the abbreviatures. What is OD?

14. line 87 -> Clarify the selection criteria, please.

15. lines 92-97 and 110-116. Add some images to clarify the idea and the preprocessing made.

16. lines 120-123. I cannot understand why do the authors use 6 datasets. Can the author clarify this along with the manuscript, and where are they used?   

17. The dataset was not make public. Can you make it?

Experimental results

18. This section is very confusing. Can you clarify it?

Some additional notes can be found at the attached document.

Comments for author File: Comments.pdf

Author Response

The answers to the reviewer's suggestions are in the attached file.

Author Response File: Author Response.pdf

Reviewer 3 Report

This manuscript is founded in order to cinematographic shot classification with deep ensemble learning. Overall, I consider it worthy of publication in Electronics. Nevertheless, there are several issues that need to be fixed before I can recommend it for publication...

(1) The purpose of the research is not clearly stated anywhere in the manuscript. Usually, after the introduction and before entering the research method, the purpose of the research is mentioned.

(2) The novelty of the manuscript is not clear and it is not emphasized anywhere in the manuscript. In addition to the text of the manuscript, it is better to mention the novelty in the abstract.

(3) The abstract of the article has not been written comprehensively. The abstract should include numerical results and at the end present a practical result of the research.

(4) It is better to use words that are not in the title of the article when choosing keywords.

(5) In this study, the results obtained have not been compared with the results of other researchers, and this is one of the requirements.

I think that some additional information and clarification (mentioned above), should contribute to increase the impact of this work, which is in my opinion of good quality and is developed in the right direction.

Author Response

The answers to the reviewer's suggestions are in the attached file.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The arguments presented by the authors are not convincing to change the decision. 

Author Response

We have addressed the reviewer suggestions and comments in the attached pdf file

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors present a DL + Ensemble Learning strategy to classify cinematographic shots. The strategy present average good results. 

Congratulations! The improvements made to the manuscript are significant and improved the manuscript quality. 

Some minor reviews:

l201-206 In these lines you refer multiple datasets. In reality, you have a single set (or 3 sets at max), and divided the dataset into multiple set (train, validation and test sets). Can you adjust the text?

l218 "cration" to "creation"

l217-219 Can you explain what are hypercolumns?

l221 "Imagenet" to "ImageNet"

l245 "stacking-leaning" to "stacking-learning"

l254 "cnn" to "CNN". This abbreviature is not extended. I advise to use acro package to track the abbreviatures and assure all the abbreviatures were extended. 

section 4.1 Can you provide a diagram of the used VGG network. If you which, you can use PlotNeuralNet to provide it. 

Some references are from arXiV. Some authors publish the manuscripts in arXiV and in a journal. Please check this publications if they are not published anywhere else. 

Author Response

We have addressed the reviewer suggestions and comments in the attached pdf file

Author Response File: Author Response.pdf

Reviewer 3 Report

The respected authors addressed all of my concerns inside the revised manuscript. So, it can be accepted in the current form.

Author Response

We have addressed the reviewer suggestions and comments in the attached pdf file

Author Response File: Author Response.pdf