Optimized and Efficient Color Prediction Algorithms Using Mask R-CNN

Round 1

Reviewer 1 Report

This paper proposed two new algorithms namely average windows (AVW) and pixel skip (PXS) that can extract dominant colors accurately in less time.

The reviewer suggests some points that can improve the paper's quality:

1. Please improve the quality of the figures. The class name of the predicted result cannot be recognized.

2. On figure 7-14, please change to tables that contain the statistical report (mean, std deviation, and so on) so that the readers can easily to know recognize which model is better.

3. To write the pseudo code (line 324-335 and line 412-424) more beautifully, it can use \begin{algorithm} on latex version (please referred to Algorithms - Overleaf, Online LaTeX Editor.).

Good luck.

Author Response

"Please see the attachment."

Author Response File: Author Response.pdf

Reviewer 2 Report

Thank you for submitting your paper. The objective of the presented study is to extract color properties for detecting objects. In short averaged color values are used with R-CNN. Some issues must be resolved before considering the publication of the manuscript. Details follow:

0. The novelty of the study should be more specifically expressed at the end of the introduction.

1. In the abstract, please be more specific while comparing the aim with other studies, such as "predicting the color of image”. Is it color segmentation or color prediction?

2. As a suggestion, it will be better to modify sentences in Lines 113-122, replacing them with a single paragraph including numbers and parenthesis

3. In Figure 1, please explain Filter & store, and pre-processing steps, please include filtering types such as gaussian filtering, etc…

4. The letters in Figure 1 are not prepared to be reader-friendly.

5. Please indicate each pixel range, 8 bits 10 bits, etc. Please define band numbers three, and four… in Table 1. It will be better to express the categorization using normalized data.

6. In the pseudo color it requires two loops for reaching a pixel which is placed in a matrice that has rows and columns. Please correct the use of f that is used for window size which has two parameters, width, and height.

7. In figure 2, there is no background for the extracted image but in figure 3 there is a background. If there is a masking operation please use identical steps for comparison.

8. Please use a more specific term for “predicted”. Are the authors indicating “processed image” or predicted objects etc...

9. The letters are too small in Figure 4.

10. Figure 5 and 6 are not clear. Please use larger images and please use contouring operation or a better boxing color. Please explain the result by extending the Figure caption.

11. Figures 7-14 should be redrawn averaged results and normalized outputs should be given.

12. Please include CPU and GPU properties while comparing the results with other state-of-the-art studies.

 

Author Response

 "Please see the attachment."

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The paper is improved. However, the answers of the authors are confusing, it is not easy to understand because it is not well organized and explained.

Please give responses point by point. Please give detailed answers.

Please address the updated line numbers, and figures.

Please indicate the improved works by addressing line numbers instead of using "Change addressed ".

Additionally, In the "4.3.2. Pseudo Code of AVW" pseudo code, if you add an increment of j in the for loop, there are two times incrementing occurs. Could you explain the aim of this step?   

Author Response

"Please see the attachment."

Author Response File: Author Response.pdf

Round 3

Reviewer 2 Report

I have seen the manuscript is improved.  Here you can a few question that must be resolved.

Reviewer Question 1:

Additionally, In the "4.3.2. Pseudo Code of AVW" pseudo code, if you add an increment of j in the for loop, there are two times incrementing occurs. Could you explain the aim of this step? 

Author answer 1:

Here is a gist of how the loop works: 1. If we have a list of pixels: L = [a,b,c,d,e,f,g,....]. For AVW, we take average of every f pixels, where f is the window size, that is pre-computed. 2. So, if f is 3, average value of a,b,c is considered first (Here, j in the for-loop is 0,1,2). After taking the average, since j=3 < total length of the list, the loop continues. 3. In the next window, we take the average of d,e,f (Here, j in the for-loop is 3,4,5). And the process continues till j values reaches the length of the list L. 

The reviewer is questioning why the author is increasing the increment variable for two (2) times at the each cycle. Are the author is used this for jumping some of the pixels etc...

In Figure 1 the letter sizes are not matching. It is not possible to read legends. Please use figure caption to explain these data. 

The changes in lines 477- 478. are not addressing the answers for the question "12. Please include CPU and GPU properties while comparing the results with other state-of-the art studies." 

Author Response

"Please see the attachment."

Author Response File: Author Response.pdf

Round 4

Reviewer 2 Report

The authors have clearly improved the manuscript and made it clearer. There is no further question on my side.