Multi-Scale Receptive Fields Convolutional Network for Action Recognition
Round 1
Reviewer 1 Report
It would be better if author can present the ROC, AUC curves to show the effectiveness of the paper.
Please add some current citations from 2021-23 in the literature. In addition, provide a justification for why your method is good in compared to others.
Overall paper is good
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
This study focuses on the development and testing of a deep residual neural network model with support for a multiscale block of receptive fields for action recognition on video data streams. The manuscript proposes the results of a more efficient formation of the feature space characterizing the action area of an object with different scale of representation in the frame of a video sequence. The stated goal and objectives of the study are relevant and in demand in a number of practical applications, such as control of production and agricultural processes, monitoring of people's behavior in public places. The main idea of the authors was to modify the residual blocks by integrating into them additional convolution layers with different scales of receptive fields and performing global average pooling on the output to form the final vector of features for recognition. The manuscript is well structured and laid out. However, a number of questions and suggestions arose in the course of reading the presented study materials:
- Perhaps the abstract would be good to specify the advantages of the proposed model over existing methods, including numerical expression.
- Row 274-275 - the authors present the limitations on the batch based on the characteristics of the GPU, I would like to know the configuration of the bench on which the training and testing was performed
- One of the tasks solved by the authors, including the reduction of computational complexity and performance improvement with the proposed solution. Perhaps it is worth presenting representative materials directly confirming this with respect to known approaches.
- It might be worth expanding the Conclusions section or adding a Discussion section to present the main limitations of this study.
- It would be great if the topic of optimizing the network while preserving the final accuracy was covered in a little more detail.
- Is there an option to add a link in the text of the manuscript to a public repository with the research material for better reproducibility and increased interest of readers?
In general, action recognition is a well-developed area of interest to both academia and business. The ideas and solutions proposed by the authors are well structured and presented, and the results presented are in line with the goals and objectives of the study. The manuscript is quite interesting, despite the relatively weak element of scientific novelty. It is suggested that it be accepted for publication after minor revisions.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Reviewer 3 Report
The authors have well organised the research article. However, there is some room for improvement:
1. Spell-check the entire article, as there are some spelling mistakes. Also, grammar-check the entire article.
2. Most references are dated before 2018, and only three are after 2018. Include recent research works in your article.
3. In the conclusion section, it is mentioned MTPDNet, but I don't find any reference about it in the article.
4. Conclusion section should be rewritten with full research findings.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 3 Report
The authors have made all necessary corrections. This article may be accepted for publication
