Round 1

Reviewer 1 Report

Dear authors:

Your are proposing a solution for a problem which will be more and more common in the near future. I have some concerns and recommendations about your work. 

1. Please add details about the implementation in hardware and software of your methods. 
2. summarize the structure  of your dataset in a single table, it would be useful to have de exact number of images, and their original size before the preprocessing for use with the DL models.
3. Add the computational cost (execution time) of the tested methods, to asses the efficency of the proposal, and it would be helpful to have a table which summarizes all the results. 
4. Check your document, in lines 164-169 it seems that some content is missing or the image was inserted in the wrong place. 
5. You have some issues with the references of your document, the message "Error! Reference source not found" appears more than once.
6. Check your images, specially figure 3 it seems that are some text missing in the blocks.
7. This is an optional recommendation. Check the origin of your references, there are several papers from conferences or non indexed sources.

Hope you find this recommendations useful. 

Author Response

The authors would like to express their appreciation for the reviewers’ time, reviewing our paper. We went over the paper again to address the valuable comments and feedback raised by reviewer 1 and edited the manuscript accordingly.

Reviewer 1

You are proposing a solution to a problem which will be more and more common in the near future. I have some concerns and recommendations about your work.

Comment 1: Please add details about the implementation in hardware and software of your methods.

Response 1: Authors would like to thank this comment. We want to withdraw the attention that the used models are well known pre-trained models which are cited in our paper.

Comment 2: summarize the structure  of your dataset in a single table, it would be useful to have de exact number of images and their original size before the preprocessing for use with the DL models

Response 2: Authors would like to thank this suggestion. The dataset was obtained in Blacksburg, VA using a smartphone app by Jahangiri and Rakha [12]. Some information about the used datasets has been included in the paper as below.

Table 1: Input of the used dataset structure

Comment 3: Add the computational cost (execution time) of the tested methods, to assess the efficiency of the proposal, and it would be helpful to have a table which summarizes all the results.

Response 3: Authors would like to thank these suggestions. We want to withdraw the attention that The pretrained models are fine-tuned offline, so we are not caring for training time, but we are caring for test time which is very short. The computational time is the forward path time, which is very short.

Comment 4: Check your document, in lines 164-169 it seems that some content is missing or the image was inserted in the wrong place.

Response 4: Authors would like to thank this comment. We have fixed the issue and re-entered the image in the right place.

Comment 5: You have some issues with the references of your document, the message "Error! Reference source not found" appears more than once.

Response 5: Thank you for bringing this up! We went over the whole manuscript and fixed all these issues.

Comment 6: Check your images, especially figure 3 it seems that are some text missing in the blocks.

Response 6: Authors would like to thank this comment. We have fixed the text in Fig. 3.

Comment 7: This is an optional recommendation. Check the origin of your references, there are several papers from conferences or non-indexed sources.

Response 7: Thanks for the recommendation. We believe some of the conference papers are very relevant to the presented work and thus it’s important to be cited in this research effort. As for the non-indexed sources, we have fixed most of them except for those which are technical reports or arXiv preprints that are not indexed, but we think it adds a great value to this manuscript.

Hope you find these recommendations useful.

Thank you very much for your valuable feedback!

Reviewer 2 Report

This paper proposed to use deep transfer learning for road users classification on smartphone sensors. Overall, the topic is interesting, but the structure of this paper is not well organized, and the presentation is not clear to the reviewer. More specifically,

• The authors aim to utilize deep learning-based transfer learning methods to tackle the classification task. Therefore, the reviewer strongly suggests adding more discussion and analysis in the related work, particularly various deep network architectures, by citing more advanced and latest works, e.g., “Feature Extraction for Hyperspectral Imagery: The Evolution from Shallow to Deep (Overview and Toolbox), IEEE Geoscience and Remote Sensing Magazine, 2020, DOI: 10.1109/MGRS.2020.2979764.”
• On page 4, 164-167, there seems to be a lack of important content.
• On the whole manuscript, there are many errors in citing the references. Please correct them.
• The experiments are insufficient. More compared methods and experimental results are needed.

Author Response

The authors would like to express their appreciation for the reviewers’ time, reviewing our paper. We went over the paper again to address the valuable comments and feedback raised by reviewer 2 and edited the manuscript accordingly.

Reviewer 2

This paper proposed to use deep transfer learning for road users classification on smartphone sensors. Overall, the topic is interesting, but the structure of this paper is not well organized, and the presentation is not clear to the reviewer. More specifically,

Comment 1: The authors aim to utilize deep learning-based transfer learning methods to tackle the classification task. Therefore, the reviewer strongly suggests adding more discussion and analysis in the related work, particularly various deep network architectures, by citing more advanced and latest works, e.g., “Feature Extraction for Hyperspectral Imagery: The Evolution from Shallow to Deep (Overview and Toolbox), IEEE Geoscience and Remote Sensing Magazine, 2020, DOI: 10.1109/MGRS.2020.2979764.”

Response 1: Authors appreciate this suggestion. We have cited this aforementioned research work in the Related Work section (L90-92)

Comment 2: On page 4, 164-167, there seems to be a lack of important content.

Response 2: Thanks for this note. We have reviewed the paragraph and it seems it wasn’t clearly presented. We edited it and linked it with the next paragraph to make it clearer.

Comment 3: “On the whole manuscript, there are many errors in citing the references. Please correct them.

Response 3: Thank you for bringing this up! We went over the whole manuscript and fixed all these issues.

Comment 4: The experiments are insufficient. More compared methods and experimental results are needed.

Response 4: Thank you for the suggestions. We added more work, Figure 3 and Figure 4, where more experiments and compared methods have been done and We added more work.

Round 2

Reviewer 1 Report

Dear authors:

You have provided answers for all my previous concerns. I have no more recommendations about this paper.

Reviewer 2 Report

The authors have well addressed the reviewer's concerns. No more comments.