A Lightweight Dense Connected Approach with Attention on Single Image Super-Resolution
Round 1
Reviewer 1 Report
In this paper a Lightweight Dense Connected Approach with Attention for Single Image Super-Resolution is proposed.
The paper includes an apprpriate introduction and state of the art.
With respect to the proposed method it is mentioned that the upscaled features are added to the bicubic-interpolated LR image to gain the final output of the upscale layer. Please include additional comments in this section to explain how this addition involves image enhancement.
Results clearly justify the improvement in terms of Peak to Signal Noise Ratio (PSNR) and Structural Similarity Index (SSIM). Please add the following aspects:
- Express Figure 5 also in terms of SSIM.
- Include a comparison in terms of computational time depending on the used algorithm.
The level of the use of the English language is unsatisfactory to be included in this publication. Authors whose primary language is not English are advised to seek help in the preparation of the paper.
Finally, conclusions must be enhanced in order to highlight the main paper results.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The work presented in this paper entitled "A Lightweight Dense Connected Approach with Attention for Single Image Super-Resolution " is good. The manuscript cannot be accepted in its current form unless the authors address the concerns that are described below:
- I suggest the Introduction be developed further in terms of the particular domain/scope of the journal in which this particular application lies.
- The authors must link the abstract/Introduction. The motivation of the relevance to their work is not well justified.
- In Section 2 (related work), It is suggested to briefly discuss other existing single image super-resolution (SISR) and Lightweight Dense Connected Approach with Attention for Single Image Super-Resolution (LDCASR) and provide evidence that the chosen architecture is best.
- Literature review should be discussed further, what overall technical gaps are observed in the paper, which led to the proposed system design.
- As the existing pre-trained model is trained for images, why do authors use transfer learning?
- In Figure 5, the training using L1 loss is shows better performance than others, can the authors provide more explanation?
- How training and testing data split and also about testing? Any fold method is used
- Justify how the work is applicable in real-time applications?
- Authors need to provide information regarding the training and test set data in more details
- Where the testing loss and the testing accuracy is explained? The resolution of the image is good enough?
- Future research work must be sufficiently widely argued. Please open a real window for future work in the conclusion section.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
- The paper needs some minor editorial corrections, as it is missing some spaces after punctuation marks, and before some citation brackets.
- The Figure 1 caption seems to be truncated.
- In the Loss function formulas (8-10), an IHR symbol is used, but is not defined in the text.
- Line 217 - incorrect reference to the loss function equations (7-9 instead of 8-10).
- The result presentation is not very clear, e.g. why LDCASR x8 is bolded in the Urban100 column, while LapSRN x8 had higher (better) PSNR, same with Set5, but this time LapSNR x8 has a better SSIM factor. You are comparing algorithms by two factors at the same time, but it is unclear whether you are using some score based on these factors, or how do you make the ranking.
- It is worth to define a clear comparison/ranking rules and consider some statistical tests to compare the proposed method with other methods (e.g. Friedman, Nemenyi, or Bonferroni-Dunn test).
- Conclusions to the paper are very modest (3 phrases), and they sound more like an abstract than conclusions.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
he authors have addressed my concerns, and I recommend publication in the journal.​
