A Lightweight Model Enhancing Facial Expression Recognition with Spatial Bias and Cosine-Harmony Loss
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors1- In the abstract, you mentioned that you achieved high accuracy without any evaluation measure, also in this case of facial emotion detection, the dataset name should be mentioned.
2- The word Novel is considered only for Patent research work, or state-of-the-art work, and both approaches are not achieved in your paper.
3- Increase the number of related work regarding the FER.
4- Highlight the modification you have achieved on Loss Function? as there are two losses for training and validation.
5- Bias can be there in CNN and some kinds of GAN, what kinds of Bias do you use, and what steps do you think its developed than others?
6- Explain the reason for using 𝑐𝑜𝑠𝑖𝑛𝑒_𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 than other distance measures, especially the data here in your case is not text ?
7- AffectNet is considered a good dataset, and in your case, you are not comparing your findings to other researchers, compare your work with other related research works on this dataset based on class distribution.
8- Add the strategy you have followed when selecting the dataset, size, number of samples in each class, and the results apps others, your experimental results only highlight your work.
9- I recommend checking and validating your findings with the following link regarding FER, https://paperswithcode.com/sota/facial-expression-recognition-on-affectnet.
10- data availability, you have to mention the source of data as I worked on them earlier, and it requires consent from the original authors, you have to get permission from them to use both datasets.
11- it's better to add some experimental results of real-time detection of facial expression for some images once you get approval from dataset authors.
Comments on the Quality of English LanguageMinor editing of English language required.
Author Response
Thank you very much for your positive and constructive comments and suggestions on our manuscript. please refer to the document for detailed reply
Author Response File: 
Author Response.docx
Reviewer 2 Report
Comments and Suggestions for AuthorsMy few comments about this manuscript are listed below.
1. Contributions are fine. However, authors are advised to make in bullets and remove numbering
2. Section 2.1 headings, please make initials capital.
3. Figure 1 is nice and depicts nicely the proposed system flow. However, in depth details are missing. Authors are strongly advised to include the critical details of their proposed method here.
4. Please put the pseudo code of your proposed method.
5. On page 5, equation numberings are missing.
6. On page 6, figure 3 is poor and red line of text is visible in image. please correct this and make all boxes transparent in your paper.
7. Text in Fig. 4 is not readable in print version. Please improve the text resolution.
8. Headings 3.3.1 and 3.3.2 are same. Why?
9. After Fig. 7, discussion should be included about proposed method.
10. Limitations of the proposed algorithm are missing. Please include and critically discuss.
11. In section 4, add few sentences about possible future work.
12. Sometimes there is discontinuity in the text. For instance on page 4, bold headings are just single line. Authors should elaborate more details in these headings, such as
Residual Bottleneck: Capture complex features and facilitate information flow.
Include more details, so that beginners and practitioners get more details.
13. include the computational complexity of your method. training and test time etc
14. Overall, manuscript is interesting and needs some detailed discussions and critical findings.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
Comments and Suggestions for AuthorsAuthors have accomodated my suggestions.
