CDTracker: Coarse-to-Fine Feature Matching and Point Densification for 3D Single-Object Tracking

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Authors must answer some points, before this paper is considered for publication. The overall quality of the research is good. 

1.      Please provide a brief description on the datasets KITTI [10] and Waymo [11] used in your experiment.

2.      I am afraid that the number of sequences used for training (0-16), (17-18) for validation and (19-20) for testing is too low. Can the number of samples used for training, testing/validation be increased? If the number of samples is limited, I would suggest you to implement and apply cross-validation technique as a more reliable technique for performance estimation.

3.      Why you set the feature dimension to 32? Did you try will other values of C and what’s their impact?

4.      Can you propose an idea of how to solve your methodology drawback of having to small number of cloud points?

Author Response

Dear reviewer

Thank you for your time and suggestions, which are beneficial for improving our paper. We have uploaded the response to your suggestion and revise our manuscript.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Content

----------

This paper introduces the CATracker, a novel 3D Single Object Tracking (SOT) network, comprising four integral modules: feature extraction, a coarse-to-fine processing stage, feature matching, and relatively dense sampling, culminating in object localization. Empirical evaluations utilizing the KITTI dataset have demonstrated notable enhancements in the Cyclist category, with a Success rate increase from 64.6% to 71.2% and a Precision rate improvement from 91.7% to 93.2%.

 

Comments

--------------

This paper conforms to the standard and the description is complete in terms of the innovative angle of topic selection, method selection, research quality, writing quality, etc. In my opinion, it is excellent without flaw.

 

Evaluation

--------------

Given the above, I'm in a position to accept.

Comments on the Quality of English Language

Minor editing of English language required

Author Response

Dear reviewer

Thank you for your patience and suggestions, which can be beneficial for improving our manuscript. We have uploaded the revision of our paper.

Best regards

Chenghan 

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

The paper proposes a 3D single object tracking method that integrates specific modules, such as cosine similarity [3,4] and attention mechanism [5,8], from different previous approaches in the literature in a coarse-to-fine feature matching model based on a hybrid similarity learning mechanism. This integration is novel and addresses gaps in the existing literature, as discussed in the second paragraph of the Introduction. 

 

The manuscript clearly outlines the work's goals and contributions and provides sufficient discussion based on the state-of-the-art literature. The results demonstrate slightly better performance compared with correlated methods. For a better presentation, it would be interesting to analyze why the proposed method is better for some objects and worthless for others. 

 

In the second paragraph of the Introduction, the authors argue about some gaps in the existing literature. To corroborate the text, some references for those conclusions must be indicated.

 

Aligning Figure 2 more closely with its reference in the text would improve comprehension, ensuring the authors' work is easily understood.

 

For better understanding and replicability, provide detailed explanations and references for the following operations: MaxPoll, MLP, SelfAttention, and CrossAtention. In the current terms, it is not possible to replicate them. 

 

In section 4, give some description of the used dataset. In subsection 4.4., discuss how the other models perform under these conditions. 

Comments on the Quality of English Language

In general, the paper is well-written and structured. I suggest a final review for a better version of the manuscript. 

Author Response

Dear reviewer

Thank you for your patience and suggestions, which can be beneficial for improving our manuscript. We have uploaded the revision of our paper.

Best regards

Chenghan 

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

Aiming at feature matching and the sparsity of point clouds in 3D single object tracking (3D SOT). This manuscript proposed CDTracker to solve these problems. The intentions of the manuscript are good, but the following questions need to be answered. Finally, a major revision is given.

 

1. In the manuscript, it is necessary to increase the effectiveness analysis of the proposed method and compare it with other methods. 1) The fine feature matching seems to only utilize the attention mechanism in Section 3.2.2. 2)Please explain the difference between section 3.3 and reference 9.

 

2. The introduction of related work is too brief, and literature search needs to be increased.

 

3. The comparison method is not novel enough, and there are many new methods in the references. It is suggested to add newer methods for comparison.

 

4. Because the comparison method is not novel enough, the effectiveness of the method needs to be confirmed, and the improvement of the proposed method is not obvious.

 

5. Pay attention to the grammar and formatting in the manuscript.

The highest result of easy in Pedestrian category in Table 2 is incorrectly highlighted.

Comments on the Quality of English Language

Pay attention to the grammar and formatting in the manuscript.

The highest result of easy in Pedestrian category in Table 2 is incorrectly highlighted.

Author Response

Dear reviewer

Thank you for your patience and suggestions, which can be beneficial for improving our manuscript. We have uploaded the revision of our paper.

Best regards

Chenghan 

Author Response File: Author Response.docx

Round 2

Reviewer 4 Report

Comments and Suggestions for Authors

Aiming at feature matching and the sparsity of point clouds in 3D single object tracking (3D SOT). This manuscript proposed CDTracker to solve these problems. After the revision, some questions have been answered clearly and the English expression has been improved. However, there are still some small questions left as follows:

1）In Section 4.2, please analyze the decline in the performance of the categories. For example, the performance of van decreases compared to the baseline in Table 1, and the performance of Vehicle decreases compared to the baseline in Table 2.

Comments on the Quality of English Language

The English expression has been improved. It is recommended to check grammar and spelling.

Author Response

Thank you for your suggestions! We appreciate your advice and your comments can be beneficial for our paper. We have improved our paper according to your comments. Our response is in the attachment.

Author Response File: Author Response.pdf