LLGF-Net: Learning Local and Global Feature Fusion for 3D Point Cloud Semantic Segmentation
Round 1
Reviewer 1 Report
The topic of the work is interesting. The authors propose to introduce a new LLGF-Net and declare its authorship. Is the proposed network entirely the authors' exclusive tool, or is it a conglomerate of other authors? The study lacks a comprehensive comparative analysis of the obtained simulation results and comparisons with the achievements of other researchers. Adding graphical representations of research results will improve the readability of the work.
There is also a lack of more complete information about the operation of the network itself, hence it is quite difficult to clearly understand what the innovation of the proposed solution is all about.
Author Response
Please see the attachment
Author Response File: 
Author Response.pdf
Reviewer 2 Report
The paper presented a semantic segmentation for the point clouds based on aggregation of local and global features. The application for point cloud semantic segmentation can be vast. The authors need to improve their presentation style and provide more details about technical concepts of related works. Some specific points:
1. What is the authors intuition on extracting local and global features separately. They need to explain that why it will be a good idea to have two different steps for local and global features separately. How it will impact segmentation. The authors can describe this based on related literature as well.
2. In the local feature extraction part they mentioned "PT [19]". They never explained what is PT approach and how their method would be different compared to it.
3. In scientific articles we usually use strong words. Statements like: "Based on the SA module, we hope to use the self-attention mechanism to learn the features of point clouds at different levels, and then splicing and combining these features to obtain the more effective information of point clouds." is quite vague and shows that the authors did not have enough confidence in their proposed method.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Reviewer 3 Report
The authors have presented a new semantic segmentation network named LLGF-Net, which uses features from both local and global levels of point clouds. The paper is well written and the quantitative results of the proposed network verifies the model. The manuscript can be accepted after consider following modifications:
- Define the expansion for the abbreviated name of “LLGF-Net”. Also, the other abbreviations should be expanded at the first seen.
- For better clarification kindly explain more about the specifications of the applied MLP networks. What is the difference between the applied MLP networks?
- Is the proposed network applied only for indoor scenes? What a bout the performance of the proposed network with other senses (such as outdoor sense) or another dataset?
- The plots of accuracy of the proposed method versus time and epoch and comparing it with the other related methods may be helpful to show the contribution of the proposed method.
- The authors have used some metrics, such as mIoU, mAcc, etc. More information about these metrics should be added for the readers.
Author Response
Please see the attachment
Author Response File: Author Response.pdf
Round 2
Reviewer 1 Report
The additions introduced by the authors significantly increase the quality and usefulness of the paper. The paper may be published in its current form.
Reviewer 2 Report
The authors have added detailed explanation where it was needed and it improved the quality of the work. I do not have any further comments.
Reviewer 3 Report
The authors have addressed all of my comments and the paper can now be accepted in the present from.
