MVG-Net: LiDAR Point Cloud Semantic Segmentation Network Integrating Multi-View Images

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

For the multimodal deep learning segmentation task,  the issue of "cross-modal consistency" is very important, It is an interesting work, however, there are some issues and work in the article that need to be improved:  

 

1、The quality of the images are  necessary to improve so that the article is easily understandable， such as Fig. 1, Fig. 2 and Fig. 4,   

 

2、For multimodal data fusion, the alignment accuracy of the images and point clouds is critical for model performance, but the method of assessing alignment accuracy doesn't mentioned in the paper.   

 

3、In the analysis of the experimental results, it is recommended that the best performing metrics are indicated in bold .

Author Response

We feel honored for careful reading and constructive suggestions from the editors and reviewers on our manuscript. We have carefully considered all comments and suggestions from the reviewers and revised our manuscript accordingly. Please see the attachment for detailed information.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

There some comments that I hope will help you improving this manuscript.

(1)    Whether the “point cloud” is complex or singular, both are mixed in the manuscript.

(2)    In this paper, the multimodal data of point cloud and image is fused. (a) How to solve the matching between image and point cloud? Does the matching accuracy affect the classification results? (b) What is the essential difference between the proposed method and the classification of point clouds with color and spectral information? It is hoped that the authors will explain and compare and analyze them in the manuscript.

(3)    The author reviews the current situation in detail, but the existing problems and the problems solved by the author still need to be strengthened.

(4)    In the Line 275, section “Dataset”. Is the model applicable when the author used original images as to train the model, but the other types images in the test data? In addition, orthophoto images were used in the Line 489. Are the images in the same form used in the model?

(5)    The results presented in the manuscript are all using dataset from very small area. It is recommended to add a complete dataset cover large area.

Author Response

We feel honored for careful reading and constructive suggestions from the editors and reviewers on our manuscript. We have carefully considered all comments and suggestions from the reviewers and revised our manuscript accordingly. Please see the attachment for detailed information.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The quality of the manuscript has improved to some extent. The author doesn't seem to be concerned about the issue of small dataset from very small area. But I am still concerned about the issue of insufficient data volume and lack of spectral suitability.

Author Response

We feel honored for careful reading and constructive suggestions from the editors and reviewers on our manuscript. We have carefully considered all comments and suggestions from the reviewers and revised our manuscript accordingly. Please see the attachment for detailed information.

Author Response File: Author Response.pdf