OD-XAI: Explainable AI-Based Semantic Object Detection for Autonomous Vehicles
, Sudeep Tanwar 1,*, Wei-Chiang Hong 2,* and Ravi Sharma 3
Round 1
Reviewer 1 Report
The paper is sound and clear to express the importance and principle of the explainable artificial intelligence in autonomous vehichels. I think the manuscript for this version is acceptable.
Author Response
Comment 1: The paper is sound and clear to express the importance and principle of the explainable artificial intelligence in autonomous vehicles. I think the manuscript for this version is acceptable.
Response: The authors are thankful to the reviewer for an appreciating response that motivate us to work with full enthusiasm.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The paper discussed the applications of explainable AI in semantic object detection for autonomous vehicles. However, there are some errors in the presentation. The technical contributions of this paper are very limited. Thus, the reviewer believes this paper might be a better fit for a conference publication in this area or a technical report.
The authors can consider the following questions and suggestions to improve the quality of the paper:
-What is OD-XAI? What is Industry 5.0? In section 1.1 research contributions, why did the paper mention Industry 4.0? Why is industry 5.0 related to autonomous vehicles?
-Figure 2 is not understandable. We are in 2022. How do you have the sale number in 2023-2025?
-Section 2.4. Integration of XAI with Semantic Segmentation in AV is very limited. The paper needs to provide more discussion, information, events to show the importance of XAI integration.
-The authors need to define clearly the type of the paper: a systematic review paper or a regular technical paper. The structure of the paper needs to be revised carefully.
-Section 6.2, the paper needs to compare those XAI techniques, advantages, disadvantages and opportunities in autonomous driving applications.
-It would be better if the paper can provide some discussion of the human-machine interaction problem, usability testing of the XAI techniques.
Author Response
Please refer the attached file.
Author Response File: Author Response.pdf
Reviewer 3 Report
This is a well-written review-based work where authors applied various XAI methods and explained segmentation models.
This paper can be accepted for publication.
Author Response
Comment 1: This is a well-written review-based work where authors applied various XAI methods and explained segmentation models. This paper can be accepted for publication.
Response: The authors are thankful to the reviewer for an appreciating response that motivate us to work with full enthusiasm.
Author Response File: Author Response.pdf
Reviewer 4 Report
This is a very interesting paper on the topic of how to explain the inner workings of a deep learning method. The use case of autonomous vehicles is used in the paper. Three deep learning methods, ResNet-18, ResNet-50, and SegNet were tested. It was tested how well they perform in semantic segmentation and classification. Later, three approaches have been used in adding explanations to the working of the said deep learning methods.
Topic of the paper is very interesting and the read was good. However, the authors have to improve the paper considerably. The most pressing issue with the paper is that the contribution is not clear, it is not communicated well throughout the paper, and it is not clearly demonstrated. Research contribution is listed in Section 1.1, but is this really contribution? And reason why is the contribution vague is the fact that the literature review is lacking. To establish the research contribution in the proper way, it is necessary to describe other people work and to clearly demonstrate how your research differs and where it adds value and improvement. At the moment, the paper seems as a list of other people’s deep learning models and other people’s AI explanations approaches applied on other people’s data. What is the uniqueness in that? If no one has ever done it, you need to demonstrate and prove that through literature overview. From the end user perspective, it would be much more interesting to see did the use of the XAI improved user’s trust or similar research.
Also, I must admit that I could agree with some of the statements from the introduction regarding autonomous vehicles. What is this statement based on: The majority of these accidents occur due to a lack of attention by the driver and their highly unpredictable mindset? If you are using statistics about car accidents as a motivation for your work, what about autonomous vehicles statistics in car accidence? What proof there is for all the safety improvements you are stating?
The introduction should be rewritten completely. It should present sufficient information, so that important readers and evaluators do not need to read the rest of text, being assured in the contribution made, and validity of the text to follow. It should establish the paper’s motivation, methods, contribution, results, and position in scientific literature.
Add a literature overview section and establish the place of your research in the overall body of domain knowledge.
In Section 2 you introduce several terms. It is not clear what is their importance for the rest of the paper? What is the importance of degrees of automation for the rest of your research? Accidentally, you establish degrees and then you use levels… The only important parts of this part of the paper are subsection 2.3 and 2.4, the rest could be removed from the paper. Also, Fig. 4 and Fig. 5 how no information value and should be removed.
If the deep learning methods are the main part of your paper, then you should explain each of the models in sufficient details.
The conclusion needs to be more than just a sum of everything done in the paper. It should present some of the most important conclusion and discussions. What are the future researches you plan to conduct?
Some minor considerations:
Why is some text blue?
On line 145, there is a sentence: “There are essentially two stakeholders when it comes to AVs.” And afterwards, you enumerate several.
In section 4.2 in pre-processing, resolution of images is change because it is more suited to you needs. What does that mean? You should explain it.
Section 4.3.1 starts with they. Who or what is they?
What is the meaning of Section 6.3? It is completely out of the topic and style of your paper. It is hard to understand it. You should either rewrite it or completely remove it.
Author Response
Please refer the attached file.
Author Response File: Author Response.pdf
Reviewer 5 Report
- This paper trained and assessed three current DL semantic segmentation models for road detection in AVs.
- Scientific contribution and novelty of the paper should be explained. Semantic segmentation is well known subject in the literature. Subject of the paper is semantic segmentation models for road detection.
- XAI should be explianed with details.
- Technical content and design details for the proposed approach should be given with explanations.
Author Response
Please refer the attached file.
Author Response File: Author Response.pdf
Round 2
Reviewer 2 Report
I read the cover letter and the revised paper. The authors have corrected some errors and provided more information which help to understand the paper. However, the correction is not sufficient to improve the paper quality and contribution significantly to be classified as a journal paper. It requires a lot of effort and time to revise the paper carefully.
- If the authors select to develop the paper as an implementation technical paper. Please focus on the technical parts with a more detailed discussion of the applied technology and algorithms. The structure of the paper needs to be revised carefully!
- The discussion of Industry 5.0 in the introduction section is not understandable. Industry 5.0 is a new concept and it is not clear what digital transformation in Industry 5.0 looks like. In line 31, the paper mentioned that “…At present, we are in the age of Industry 5.0..”. Please provide reasons, references explaining why the authors believe that! Again, the paper should focus on Explainable AI, not Industry 5.0. It would be better to remove Industry 5.0 from the paper content.
- The paper may need a proofread for its writing.
-The additional discussion of section 2.4 Integration of XAI with Semantic Segmentation in AV and 2.4.1 Human-Computer Interaction (XAI and AVs) does not have citations. The discussion needs to be justified.
Author Response
Please see attached file.
Author Response File: Author Response.pdf
Reviewer 4 Report
I think that paper is now improved enough that it could be published.
Round 3
Reviewer 2 Report
The revision didn't improve the quality of the paper. The technical contributions and novelty of this paper are very limited. Thus, the reviewer believes this paper might be a better fit for a conference publication in this area or a technical report.
Author Response
Please refer the attached file.
Author Response File: Author Response.pdf
