GRID: GRID Resample by Information Distribution

Round 1

Reviewer 1 Report

This is a nice paper which clearly explains a simple and effective technique to use grid sampling superpixels in a variety of segmentation techniques. The paper compares the efficiency, the accuracy and the number of segments generated from all of these methods which is interesting and helpful.

There is a sensible introduction with some useful relevant citations provided. The introduction is a little dense and would possibly benefit the text being broken up a little (possibly through sub-headings or a figure). Some of the tabular information would be better displayed as a plot for ease of comparison. The equations in section 2 and 3 would benefit from clearer explanation, symbols are mostly specified but the real meaning of these is not always clear. Several useful experiments are done in section 4, the results of these could be slightly better displayed for ease of reading but the raw content is there. There is a sensible conclusion where the conclusions drawn are based on the evidence given in the paper which is good.

There are a few typos/grammatical mistakes, to give one example (I won't list them all) - "Owning to the non-iterative implementation" (p2, line 87) "Owning" is not correct here. The paper would benefit from a copy-edit.

Acronyms for CW and WS should be defined in the paper (on first usage on page 5). From reading the papers cited these must be Compact Watershed for CW and Watershed Superpixel for WS, these should be clearly specified on first usage.

Tables 1-3 would (also) benefit from being in a graphical form for ease of comparison, possibly showing average improvement per algorithm of the Grid approach across the superpixel number range. This would enable readers to compare the results from method rather than looking at many individual numbers.

Author Response

Dear Editor and Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “GRID: GRID Resample by Information Distribution” (symmetry-888921). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to us. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in blue in the paper using the "Track Changes" function in Microsoft Word.

The main modifications can be summarized as follows:

(1) We have thoroughly rechecked the manuscript and also invited an English native researcher to proofread our work. Lots of grammatical mistakes, typos and inappropriate sentences have been corrected or rewritten in the revised manuscript.

(2) Some inappropriate expressions have been adjusted, which would be more objective and precise.

(3) We have added a comparison table of key notations in the manuscript, which enables the symbols and equations easier for readers to understand.

(4) We have converted Table 1-3 to Figure 4-6 for better quantitive contrasts.

(5) The correlation of our idea with the symmetrical concept is complemented in both the abstract and main text (subsection 3.2).

Again, thank you for your timely notification and your consideration of this paper. We hope that the revised manuscript is now suitable for publication. If you have any question about this paper, please contact us without hesitation.

Best regards!

Responds to the reviewer #Reviewer 1

Comments and Suggestions for Authors

This is a nice paper which clearly explains a simple and effective technique to use grid sampling superpixels in a variety of segmentation techniques. The paper compares the efficiency, the accuracy and the number of segments generated from all of these methods which is interesting and helpful.

Authors:

The authors thank the reviewer for his/her valuable comments. They are all constructive and very helpful for revising and improving our paper. The point-to-point modifications are summarized as follows:

Question 1#

There is a sensible introduction with some useful relevant citations provided. The introduction is a little dense and would possibly benefit the text being broken up a little (possibly through sub-headings or a figure). Some of the tabular information would be better displayed as a plot for ease of comparison. The equations in section 2 and 3 would benefit from clearer explanation, symbols are mostly specified but the real meaning of these is not always clear. Several useful experiments are done in section 4, the results of these could be slightly better displayed for ease of reading but the raw content is there. There is a sensible conclusion where the conclusions drawn are based on the evidence given in the paper which is good.

Authors:

Thanks for your kind reminder. We have re-organized Section 1 and resolved it into three major parts, Graph-Based Superpixel Segmentation, Gradient-Based Superpixel Segmentation and Background of Seed-Demand Superpixels. We believe that the three subsections could alleviate the dense content in the first section and clarify the context.

Besides, the notations frequently used in this paper are summarized by a new Table 1 at the beginning of Section 2. Some ambiguous symbols, such as Π() in Equation (7) and (12) are modified thoroughly.

Question 2#

There are a few typos/grammatical mistakes, to give one example (I won't list them all) - "Owning to the non-iterative implementation" (p2, line 87) "Owning" is not correct here. The paper would benefit from a copy-edit.

Authors:

Thanks for your careful inspections. We have totally rechecked the paper, lots of grammatical mistakes, typos and inappropriate words have been found, and they have been corrected in the revised manuscript.

Question 3#

Acronyms for CW and WS should be defined in the paper (on first usage on page 5). From reading the papers cited these must be Compact Watershed for CW and Watershed Superpixel for WS, these should be clearly specified on first usage.

Authors:

Thanks for your kind reminder. We have corrected the sentence that first introduces CW and WS in Section 1 as follows:

(L99-100) To influence the compactness of watershed algorithms that over-segment an image without iteration, Watershed Superpixels (WS) [20] and Compact Watershed (CW) [21] introduce spatial constraint to a SLIC‐like grid respectively.

Question 4#

Tables 1-3 would (also) benefit from being in a graphical form for ease of comparison, possibly showing average improvement per algorithm of the Grid approach across the superpixel number range. This would enable readers to compare the results from method rather than looking at many individual numbers.

Authors:

(L365, L395, L402) Thanks for your suggestion. We have converted Table 1-3 to Figure 4-6 for better quantitive contrasts.

We appreciate for Editors/Reviewers’ warm work earnestly, and hope that the correction will meet with approval. Once again, thank you very much for your comments and suggestions.

Reviewer 2 Report

This paper proposes an initialization strategy to improve the performance of the grid sampling based superpixel segmentation algorithms. The main idea is to recursively generate the initial seeds that are aware of the image content,  half of the expected number of seeds being firstly located by the conventional initialization method. Then, new seeds are created that lower the amount of color information in smaller ranges. This recursive procedure is more sensitive to the complexity of regional information.

The efficiency of the method is verified by embedding it into four superpixel segmentation algorithms and replacing the conventional grid sampling. Experimental results are reported that show that the method acquire the controllability of outputting the exact number of superpixels expected by the user with comparable quantitative metrics on the BSDS500 dataset.

Author Response

Dear Editor and Reviewers:

Thank you for your letter and for the reviewers’ comments concerning our manuscript entitled “GRID: GRID Resample by Information Distribution” (symmetry-888921). Those comments are all valuable and very helpful for revising and improving our paper, as well as the important guiding significance to us. We have studied comments carefully and have made correction which we hope meet with approval. Revised portion are marked in blue in the paper using the "Track Changes" function in Microsoft Word.

The main modifications can be summarized as follows:

(1) We have thoroughly rechecked the manuscript and also invited an English native researcher to proofread our work. Lots of grammatical mistakes, typos and inappropriate sentences have been corrected or rewritten in the revised manuscript.

(2) Some inappropriate expressions have been adjusted, which would be more objective and precise.

(3) We have added a comparison table of key notations in the manuscript, which enables the symbols and equations easier for readers to understand.

(4) We have converted Table 1-3 to Figure 4-6 for better quantitive contrasts.

(5) The correlation of our idea with the symmetrical concept is complemented in both the abstract and main text (subsection 3.2).

Again, thank you for your timely notification and your consideration of this paper. We hope that the revised manuscript is now suitable for publication. If you have any question about this paper, please contact us without hesitation.

Best regards!

Responds to Reviewer #2

Comments and Suggestions for Authors:

This paper proposes an initialization strategy to improve the performance of the grid sampling based superpixel segmentation algorithms. The main idea is to recursively generate the initial seeds that are aware of the image content, half of the expected number of seeds being firstly located by the conventional initialization method. Then, new seeds are created that lower the amount of color information in smaller ranges. This recursive procedure is more sensitive to the complexity of regional information.

The efficiency of the method is verified by embedding it into four superpixel segmentation algorithms and replacing the conventional grid sampling. Experimental results are reported that show that the method acquire the controllability of outputting the exact number of superpixels expected by the user with comparable quantitative metrics on the BSDS500 dataset.

Authors:

We appreciate for Editors/Reviewers’ warm work earnestly, and hope that the correction will meet with approval. Once again, thank you very much for your comments and suggestions.