Progressive Pruning of Light Dehaze Networks for Static Scenes
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis paper is quite significant and can be interesting to a specific 'audience'.
However, the Introduction should show better and more comprehensively 1) the value of this research in the current panorama of studies and 2) the specific research goals (perhaps more schematically summarized) and the strategies implemented by the Authors to achieve them.
The paper needs a proper Literature Review, possibly in a specific, dedicated section, between the Introduction and the part on the methods. A Literature Review, in an academic paper, cannot be delegated to the Introduction or be 'scattered' here and there, all over the text of the article. Moreover, more works should be added, even more general and/or basic, in order to provide the readers with a more 'all-embracing' survey / outline and to make the paper more 'user-friendly', even to a non-specialized 'audience'.
The methods look clear, to me, and quite well-presented. However, both the Methods' section and the figures all over the paper need necessarily to be double-checked by at least another Reviewer with a more direct expertise than me in the topic of the paper, for final validation and fairness. This is something that cannot be 'skipped' and that is fundamental for a final clearance of the article in view of a possible publication.
The experiments are well-described and, for once, I have not to ask for expansions in this 'sector'. Good work.
However, the paper has not a proper Discussion section, because the Discussion is 'incorporated', with the experiments, into section 3. That is not up to academic standards. The Discussion is the 'meat' of a paper and should have its dedicated section, naturally expanded (the analysis in section 3 is not enough) with more comments, observations, and 'analytical moments'. The Authors could also add some personal remarks, which would not compromise the 'scientificity' of the paper, but would add something valuable about the process of implementation of their project. This is something that needs to be addressed, with the addition of a new section and with a clear separation between the description of the experiments and their analysis and related discussion.
The Conclusion is too short, and also this is not up to academic standards. Like in a 'mirror' with the Introduction, the Authors should still briefly stress on the relevance of their paper in its specific field and summarize, for the sake of clarity, their findings, adding also something more on the limitations of their research and experiments. A synthetic, but exhaustive Conclusion is necessary and represents the only appropriate way to wrap an interesting paper like this up.
All in all, the article has its merits and definitely can be considered for publication, but only after a thorough revision, especially at the level of format.
Thank you very much.
Comments on the Quality of English LanguageThe English language is quite good.
However, the written style is clearly non-native and should be improved, by making it 'flowing' better - also, some 'infelicities' surface here and there, especially in the use of prepositions.
An accurate revision, possibly with the help of a native-speaker, would be highly beneficial to the text of this work.
Thank you.
Regards.
Author Response
Dear Reviewer,
Thank you for your thorough review and valuable suggestions for improving our manuscript. We have carefully addressed all your comments and made the corresponding revisions. Our responses to each point and the changes made are detailed in the attached PDF file.
Please refer to the revised manuscript for the complete implementation of these changes. We believe these modifications have significantly improved the quality and clarity of our paper.
Thank you again for your constructive feedback.
Best regards,
Sungkwan Youm
Author Response File: 
Author Response.pdf
Reviewer 2 Report
Comments and Suggestions for AuthorsThis manuscript introduces an innovative, adaptive pruning approach tailored for static scenes, aiming to optimize the utilization of computational resources, which is particularly significant for environments with constrained resources. Research focuses on scenes with fixed cameras, a common configuration in numerous practical applications such as traffic monitoring and outdoor surveillance, thereby granting the research high practical value. By comparing the pruned model with the non-pruned counterpart, the manuscript demonstrates the ability of the pruned model to maintain dehazing quality while reducing computational resource consumption, with results that are convincingly effective.
However, there are still some questions for improvement :
1. Repetition of titles 2.1 and 2.2, both Lightweight Model for Static Scenes
2. The two paragraphs in lines 391~395 and 396~400 are repeated
3. The manuscript mentions that the O-HAZY dataset was used for the experiment, but only the source and characteristics of the dataset are described. It is recommended to describe in detail the composition of the dataset, its size, the diversity of scenarios, and how it relates to the research objectives.
4. Although the article mentions some related work, it lacks direct comparison with existing technologies. It is suggested to increase the comparison with the current state-of-the-art technology or method to highlight the advantages of the method and the potential room for improvement.
5. By comparing the evaluation indexes of each channel before and after pruning to determine whether the channel should be pruned, this pruning method is accidental and lacks interpretability, it is recommended to increase the test data and add the analysis of model pruning, for example, which channels in the model are the most critical for the task of de-fogging after pruning.
Additionally, the following papers may help give you some insight in Model Pruning. Please introduce and discuss about these papers.
Zhao, M., Li, M., Peng, S. L., & Li, J. (2022). A Novel Deep Learning Model Compression Algorithm. Electronics, 11(7), 1066.
Author Response
Dear Reviewer,
Thank you for your thorough review and valuable suggestions for improving our manuscript. We have carefully addressed all your comments and made the corresponding revisions. Our responses to each point and the changes made are detailed in the attached PDF file.
Please refer to the revised manuscript for the complete implementation of these changes. We believe these modifications have significantly improved the quality and clarity of our paper.
Thank you again for your constructive feedback.
Best regards,
Sungkwan Youm
Author Response File: Author Response.pdf
Reviewer 3 Report
Comments and Suggestions for AuthorsThis paper addressed an adaptive pruning method for Light DeHaze Networks, focusing on fixed camera environments. Measures including a fine-tuning strategy for specific scenes, channel importance analysis, and an adaptive pruning approach are presented. The English writing is good. However, it needs major revsion before acceptance.
1. The contributions of the paper should be made clear in introduction section.
2. page 4, the caption of Figure 1 should be revised.
3. the titles of 2.1 and 2.2 are the same, one of them should be corrected.
4. all equations in the paper should be numbered.
5. It is not clear how to conduct the fine-tuning. Neither, it is not clear how to perform 10% and 60% prune. More details are expected.
6. page 11, the second paragraph is repeated.
7. fonts in Figures 4~11 and 13 are too small.
8. why underscore layer7? As mentioned above, the details of measures including fine-tuning and pruning are not clear.
9. the explanation of figure 9, or conclusion from the figure, is not clear.
10. how to understand "adaptive" pruning? as can be seen from Algorithm, it is a kind of iterative algorithm, instead of adaptive one.
11. legends in Figure 9, 10 and 11 are inconsitent with those in text.
12. The details on how to build up the six models to be compared are not clear.
13. From Figure 11, it is not clear which fine-tuning is preferred.
14. in many places, Figure ?? and Fig. ?? are detected, please check and revise.
15. Figure 12, where is the original image? the caption needs to be revised?
16. the conclusions in 3.3 Performance in Fixed Camera Environments need supporting materials, especially for "Adaptation to Lighting Changes" and "Real-world Application". I guess some results are missing in the paper.
17. where is Figure 14?
Author Response
Dear Reviewer,
Thank you for your thorough review and valuable suggestions for improving our manuscript. We have carefully addressed all your comments and made the corresponding revisions. Our responses to each point and the changes made are detailed in the attached PDF file.
Please refer to the revised manuscript for the complete implementation of these changes. We believe these modifications have significantly improved the quality and clarity of our paper.
Thank you again for your constructive feedback.
Best regards,
Sungkwan Youm
Author Response File: Author Response.pdf
Reviewer 4 Report
Comments and Suggestions for AuthorsThe paper focuses on an important area of industry research pruning images for static photo/fixed cameras.
My suggestions:
The formatting of the paper is not correct – for example, figure 1 is probably text style not figure caption like in the template. What is more authors by mistake do not delete the instruction from the template in Figure 1 title: “Figures should be placed in the main text near to the first time they are cited. A caption on a single line should be centred.” – please delete it.
Figure 1 is not professional. I suggest using UML style for example line in the paper: “Solving Product Allocation Problem (PAP) by Using ANN and Clustering”.
The hyperlink to the figures is not needed. If authors want it please just change the colour of the hyperlinks – i.e. in line 138, 418, 472, 482, etc.
Figures 2, and 3 are of low quality, please change them and put in the resolution at least 300 dpi.
What means the three dots in Figure 2 under the last Conv box on the left? It needs to be clarified.
The formulas are not numbered, please add the numbering.
I suggest adding the explanation of variables/list of abbreviations.
Figures 4 - 9 should be bigger, or the values and axis description should be presented in larger font, now is hard to read.
Please enlarge the size of the font in Figures 10-11, and 13.
In line 504 there is a lack of figure number. Instead, it is just: “??”. The same is in lines 530, 540.
In line 581 Fig. 14 is mentioned but there is no figure 14 in the paper.
For Figure 5 – 7 I suggest adding the box and whiskey bars.
I suggest adding a few keywords after the abstract.
There is a lack of description of input data – mainly statistics for this data. I suggest making some charts with statistics and adding a description of input data at least like in the paper: “Big Data Analytics and Anomaly Prediction in the Cold Chain to Supply Chain Resilience”.
The state of the art is poor. I suggest extending the literature review based on the last 3 years' publication.
The structure of the paper is weak. There is no Result chapter, and results are mixed with the chapter “Experiments and Performance Analysis”. I suggest changing the title of the chapter to “Results” and adding subchapters “Experiments Description” and “Performance Analysis” separately.
Author Response
Dear Reviewer,
Thank you for your thorough review and valuable suggestions for improving our manuscript. We have carefully addressed all your comments and made the corresponding revisions. Our responses to each point and the changes made are detailed in the attached PDF file.
Please refer to the revised manuscript for the complete implementation of these changes. We believe these modifications have significantly improved the quality and clarity of our paper.
Thank you again for your constructive feedback.
Best regards,
Sungkwan Youm
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe manuscript was significantly improved.
The positive attitude shown by the Authors is appreciated.
The Discussion could be expanded still a little, but that is not a stringent requirement.
As it is, the paper can be considered for publication.
Thank you very much.
Reviewer 3 Report
Comments and Suggestions for AuthorsMy concerns have been addressed well. I have no further comments.
Reviewer 4 Report
Comments and Suggestions for AuthorsI have no more comments.
