LaneFormer: Real-Time Lane Exaction and Detection via Transformer

Round 1

Reviewer 1 Report

The novelty of the proposed research paper is highlighted by the proposed lane line detection that makes use of a correction module to adjust the dimensions of the extracted features to provide improved accuracy.

The proposed research is significant as lane line detection represents an important research field, and this has become the most common attention mechanism implemented on modern cars. Advanced lane line detection represents one of the most important elements that enables autonomous driving system and a high number of researchers from different companies and research institutes are developing advanced algorithms aimed to improve intelligent driving.

The paper is organized and well documented. The introduction provides an overview related to lane line detection as well as the state-of-the-art associated with lane line detection algorithms both traditional and algorithms based on neural networks. The article has a good amount of recent published references from 2020, 2021 and 2022 as well as older literature references. The methods section provides the proposed architecture starting with the input, feature fusion, feature correction and non-local transformer to enable the output. The backbone is based on the Residual Neural Network. The Feature correct, transformer decoder section, lane detection model is clearly described. The results section presents the proposed method applied on the TuSimple dataset. The performance comparison between the proposed method and other lane line detection methods are presented in Table 1. The comparison between the proposed transformer decoder module and other current existing method is presented in Table 2. The conclusions section is based on the research findings. This section is short but on point to highlighting the proposed lane line detection system.

The proposed work has a high interest for readers and scientists that developing algorithms for lane line detection and other attention mechanisms that enable driving aids and autonomous driving systems.

My only concern for the proposed article is the following:

Concern 1: The references are not numbered consecutively according to the first mention. The first reference is [40] within the abstract and the second references mention are numbered [7 13 15 19 26 28 32 36]

Author Response

Dear Reviewer:

I will adjust the numbers in the order of citation and the specific revisions will be reflected in the latest submitted manuscript.

Thank you very much for your valuable comments.

Reviewer 2 Report

The authors proposed an attractive network: LaneFormer; they use the encode-decode method to downsample the extracted features three times, upsample them three times, then fuse them in their respective channels to extract the slender lane line structure. At the same time, a correction module is designed to adjust the dimensions of the extracted features using MLP, judging whether the feature is completely extracted through the form of the loss function, and finally send the feature into the transformer network and find the lane line points through the attention mechanism, and finally design a road and camera model to fit the identified lane line feature points. Overall, the idea seems good, but the reviewer suggests addressing the below concerns before publication.

1.     The startup of the introduction section is fine. But later, the authors used quite long paragraphs. The authors should revise the long paragraphs by dividing them into small ones.

2.     The reviewer suggests that you either add a separate literature review section or simplify your introduction to catch the readers' attention. The readers' loss at the mid of the paragraph.

3.     Please clarify/revise your contributions at the end of the introduction. It will be better to add a paragraph at the end of the introduction regarding the structure of the paper.

4.  Proofread the paper couple of times. It has a lot of grammar issues.

5.     Please explain the symbols used in your Equations.

6.     What is the cost you paid for getting better results than the convolutional neural network?

7.     Is it possible to add this feature to intelligent transportation systems or vehicular ad hoc networks?

Author Response

Dear Reviewer:

Thank you very much for your comments on the revision of our manuscript. They are very valuable and helpful in improving our paper and also important to guide our research direction; we fully accept your comments and revise and reply to them.

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The authors have touched upon an interesting study that deals with lane line detection. I have some minor comments to improve the paper;

- The introduction is very lengthy. It would be good if some points could be discussed later and make the introduction concise

- Paper needs some moderate English editing. 

- Please reorganize Table 2. It is very difficult to understand since it is clumsy

Author Response

Dear Reviewer:

Thank you very much for your comments on the revision of our manuscript. They are very valuable and helpful in improving our paper and also important to guide our research direction; we fully accept your comments and revise and reply to them.

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

All my concerns are addressed carefully.