3DPlanNet: Generating 3D Models from 2D Floor Plan Images Using Ensemble Methods

Round 1

Reviewer 1 Report

The purpose of the authors sounds very interesting, and authors well presented their work; I would suggest adding some clarifications, particularly:

• Could the authors add some more references? It would help the reader in contextualising the topic.
• Could the authors add in the Introduction something more about the possible practical usage of their work (even considering that further improvements are needed)?
• Line 42: which is the exact amount of training data? “30 or less” does not sound much precise in my opinion.
• Could the author add some information about the type of 2D plan images involved?
• Could the authors add some further information about the involved datasets (lines 209-216)?
• Could the authors clarify the plan scaling section (particularly concerning the comparison with actual area value)?
• Could the authors add some further details in the object generation section?

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper presents a machine learning framework for generating three-dimensional models from two-dimensional floor plans. The methods essentially uses rule-based algorithms, together with data-driven algorithms, to generate the 3D maps. The authors claim the algorithm is 97% accurate, though no evidence is presented to back up such claims. I recommend the paper be accepted with minor modifications.

Comments

1. As stated before, the authors claim that the algorithm is 97% efficient, and although the authors make a comprehensive explanation of the algorithm itself with sound justification, no evidence is presented in the paper that the proposed algorithm is in fact 97% efficient. The authors need to include a clear and comprehensive example where such efficiency is clearly presented to be able to demonstrate this 97% efficiency, otherwise the reader is to take the authors claim without evidence.
2. Many of the plots shown in the paper are very small and hard to clearly distinguish the relevant elements (walls, doors, windows, etc). I recommend the authors provide bigger plots of the floor-plans
3. The authors make the claim that their algorithm uses only "30 data", which is extremely vague. What do the authors mean by "data"? Is the complete floor plans with appropriate 3D renders? What are the elements of these "30 data"? This wording is confusing and not clear at all.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Respected Author's,

Before proceeding further, I have a few comments which are listed below,

1.  In the abstract, Line number 2 & 3,  please recorrect the sentence.
2. In figure 2, please change the resolution of the image.
3. I have small confusion, what are the advantages rule base method with other existing methods.
4. author, please add more related works, to enhance the novelty of the work.
5. kindly add the novelty of your paper.
6. There is no comparison with existing algorithms, kindly compare the accuracy with other resent algorithms.
7. there is no recent citation is added in the work, kindly add it and compare the same.
8. kindly provide the accuracy with two decimal points with a percentage like 95.21%.
9. Need of ensemble method and what level ensemble method works.
10. what about computational complexity for creating 3d images.
11. Figure 11 resolution was poor.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

Respected authors,

1.Kindly add the comparison table for different methods.

2. Kindly draw the object recognition garph in terms of each class accuracy

Author Response

Please see the attachment.

Author Response File: Author Response.pdf