Area Contrast Distribution Loss for Underwater Image Enhancement

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This post is something interesting from the point of writing, but needs a few notes of improvement:

1. The notation in equation (3) is out of sync with the explanation in line 188 and so on. P?, \phi ?, \Theta ? …, wouldn't it be better if some of the abbreviations in equation (6) are explained first to improve readability.

2. Re-typing line 232, equation (6), and some possible misunderstandings are needed.

3. This paper has produced the formulation from (1) to (10), but in writing it requires an explanation that links between several equations. After all, this explanation is proof of the implementation of these equations in computation as the results are presented in Table 1, Table 2 and Table 3.

4. This paper develops an ACDL method to improve underwater images, but requires additional explanation or description if it cannot be described mathematically about ACDL as a novel method.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The author proposed a light network for underwater image enhancement. According to the authors,  a new lightweight neural network, Shallow-RepNet, is proposed with fewer parameters than existing models, potentially enabling faster and more efficient underwater image enhancement. However, the presentation of the paper and results are not quite well. The authors should carefully validate the proposed model and pay more attention to the academic writing. The authors should address the following comments/suggestions:

1. The abstract contains typographical errors (e.g., "slove" should be "solve") that should be corrected to improve clarity and professionalism. Besides, the authors should provide more specific information about the performance of the Shallow-RepNet compared to existing models, including quantitative metrics and/or qualitative examples. The abstract lacks information on the dataset(s) used for training and evaluation, as well as any potential limitations of the proposed methods.

2. The authors only added one figure that shows the visual results of the previous methods, however, I could not find the result of the proposed method.

3. There are many typos in the manuscript such as: Line 15: "The underwater robotics is" should be "Underwater robotics are" or "The field of underwater robotics is."

 

Line 19-20: "The poor underwater environment and poor lighting conditions of low-power and  high-speed cameras will hinder their development." This sentence is unclear and should be rephrased to better explain how poor lighting conditions and low-power, high-speed cameras negatively impact AUV development.

Line 20-21: "So this is an example of how this critically affects the performance of AUV." This sentence is vague and should be revised for clarity. Instead, provide specific examples or describe the impact in more detail.

Line 22: "especially on a a real underwater divice" should be "especially on a real underwater device."

Line 25: The sentence "At the same time, we also need to design a suitable lightweight model for a variety of environments." is incomplete and should be restructured to clarify the goal of designing a suitable lightweight model, e.g., "At the same time, there is a need to design suitable lightweight models that can adapt to a variety of environments."

Line 27: "In the past year" is too specific and may not be accurate. It is better to use a more general phrase, such as "In recent years."

Line 32: "Johnson et al. (2016) [12]" should be "Johnson et al. [12]" since the year is not necessary in this context.

4.  Make sure that the text is consistent in terms of formatting, especially with respect to citations. Consider providing a more comprehensive background on the challenges and current state of underwater image enhancement before delving into specific methods. Clearly articulate the goals and objectives of the paper to provide context for the reader.

5.  In table 1, the data shows that the prior models are better than the proposed model, the authors should provide more details about the results presented in the table.

6. The captions of table 1, table 2, table 3 have different formats. 

7. The authors mentioned the WaterNet but I couldn't find the citation for the mentioned model. 

8. The authors must provide visual results that claim that proposed model is better than prior models. However, the authors only provided one figure and there is no obvious difference among the results of proposed and prior models. 

 

 

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

In this paper, the authors propose a neural network based underwater image enhancement algorithm to solve the problem of colour distortion and low contrast. The authors propose to compute a loss function to describe the image degradation locally rather than globally for the whole image. The paper presents a description of the loss computation approach and a modified structure of the neural network proposed by the authors. At the end of the paper, the results of testing the developed algorithm are presented. The relevance of the paper is obvious. However, the text of the article makes it difficult to understand the novelty and practical relevance of the work.

This is worth mentioning as a comment on the paper:

1. The first thing that catches the eye is the huge number of typos, especially in the first part of the article. In the first sentence in the abstract: "effectively slove the problem". And then:

"Therefore, we propose an Area Contrastation Distribution Loss (ACDL).we train a flow model" (again in the abstract)

«on a a real underwater divice, require»

« To illustratethe the proposed approach»

« the quality of underwater images. we propose»

«Model of Trian» на рисунке 3.

etc….

2. I am not a native English speaker, but I think the translation of the article is poorly done. For example, the phrase "the true ground-truth" is, in my opinion, a tautology. And the term "ground-truth" occurs both with and without a hyphen, including in Figure 2. The meaning of the sentence "The poor underwater environment and poor lighting conditions of low-power and high-speed cameras will hinder their development" is not clear. Etc….

3. The article contains a fairly detailed review of the literature, but a paragraph should be added at the end with brief conclusions on this, outlining the novelty of the proposed solution.

4. Section 4.4 is poorly structured. Table 2 compares different image enhancement algorithms. In it, the authors call their algorithm "Our Model (Test)" and it has the least number of parameters and the best image processing speed, but not the best performance metrics (Table 1). Furthermore, in Table 3, there are several variants of the algorithm proposed by the authors under the names "Shallow-Repnet+BN" and "Shallow-Repnet + BN + Lacd", where these variants show the best performance. It is worth specifying which variant the authors mean by "Our Model (Test)". And the number of parameters and processing speed of "Shallow-Repnet+BN" and "Shallow-Repnet + BN + Lacd".

5. In section 4.4, it is worth mentioning the characteristics of the computer on which the performance of the algorithms was tested.

6. The authors position their network as lightweight and intended for on-line image processing. However, the Shallow-UWnet neural network on which the authors have based their network has a processing speed of 0.02 seconds per image, which is sufficient to process 50 images per second. However, its performance in Table 1 is superior to the neural network proposed by the authors.

7. The discussion section at the end of the paper is completely missing. Section 5 needs to be extended.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

During the first round of review, the authors did not modify the manuscript as per the suggested comments. In the point-to-point response letter, the authors provided a different response in reference to point 3, which I asked to outline the novelty of the proposed solution. However, the response was entirely different. Additionally, I commented on the visual results in the manuscript, and the authors' response was that Figure 3 shows the images obtained after processing using their proposed algorithm, whereas Figure 3 is the block diagram. I suggested to the authors to modify the captions of the tables presented in the manuscript. However, the captions of Table 2 and 3 are the same despite having different data presented based on algorithm measurement. Table 2 shows the ablation study in terms of SSIM and PSNR, while Table 3 shows the running time comparison and the number of parameters used in the prior and proposed models. In the previous version of the manuscript, the authors added visual results of the prior and proposed models in Figure 5. I suggested to the authors to clearly state that the proposed model is better in terms of visual clarity, color correction, etc. However, I did not find which one is the proposed model, as the authors mentioned ShallowRepnet, but only mentioned Hazed, WaterNet, FUniE-Gan, Deep SESR, Shallow-UWnet, Shallow-Repvgg, and clear. The last column in Figure 5 is labeled as clear, and it is unclear what it means. Finally, I suggested to the authors to provide more visual results showing how the proposed model is better than prior models. However, there were no new visual results in the modified manuscript. Additionally, the authors added two different terms for the proposed model, Shallow-Repvgg and ShallowRepnet, without clear explanation in the manuscript.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The article can be published in its current form.

Author Response

Understood. Thank you for your feedback!