Urban Building Mesh Polygonization Based on 1-Ring Patch and Topology Optimization

Round 1

Reviewer 1 Report

The manuscript titled ‘Urban Building Mesh Polygonization Based on 1-Ring Patch and Topology Optimization’ presents a method for deriving polygonal watertight models for buildings from triangular meshes with noise. No (learned) previous knowledge (e.g., parameter distributions) or model knowledge (roof/building primitives) are used. Even if the results of the method in Figure 8 seem to be very promising compared to other methods, the way the method is presented in the manuscript is far from being understandable and the presentation does not justify the results presented in Figure 8. The authors mention three scientific contributions: a) a new modeling primitive (1-Ring Patch), b) optimization, and c) using structural building information for a more detailed and accurate polygonal model. In my point of view, all three contributions are c questionable and not justified by the manuscript. The paper has to be rewritten completely.

My judgment is justified in detail in the following:

• The algorithm for generating the 1-ring-patches (section 2.1, 1. – 4.) is unclear. Terms are used inconsistently (e.g., center point vs. central point). What is the result of the algorithm?  I do not see why there are triangles not contained in a patch. The message of Figure 3 is unclear. The added value of 1-ring-patches (which are mentioned in the title of the manuscript) in general is unclear. Is it just an aggregation of triangles to polygons?
• Calculating of the 3-ring planarity (starting from line 193): The sentence ‘The 2-ring neighborhood points of a vertex correspond to the union of the 1-ring neighborhood points of that vertex’ is unclear. Why do you need complex index structures in order to avoid repetitions in a 3-neighborhood? I think a k-Ring-Neighborhood is more flexible and adaptive than a 3-Ring-Neighboorhood.
• The message of Figure 1 is unclear. I do not see 1-ring-patches in b), planar primitives in c), topological optimizations in d) and faces in e).
• Region growing (section 2.2): a region is enlarged based on a threshold for distance and deviation of normal direction. This is obvious and a standard method.
• The method for classification of triangles as ground, façade and roof is obvious. There are many other approaches which are more sophisticated. Why is the classification necessary?
• You user SABMP and mention that you make some improvements. What are these improvements? In general, the relation of your approach to SABMP is unclear.
• Optimization of candidate faces (starting from line 411): your algorithm decides whether a candidate face is preserved (by 3 criteria). If a face is not preserved, how is the hole filled in order to get water tightness? How are the values in line 431 derived? What is N in (4) to (7)?
• The conclusions are far too short and do not precisely give the contributions compared to other approaches.
• The paper is poorly formatted; blanks to separate references/captions are missing.

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

The manuscript proposes a polygonal model generation and representation strategy based on the polygonization algorithm. It is pivoted to 3D urban meshes, derived from SfM and MVS techniques. Each processing phase involving 1-ring patch reconstruction of the triangulated network, plane extraction, topology correction and construction as well as simplification, is optimized leading to a polygonal mesh of high fidelity, detail and accuracy. The introduction of a new modeling primitive and the proposed solution towards topological problems caused by plane segmentation, further enhance the contribution novelty, as being part of a well-defined and valid methodology. Despite that the complete recovery of the building structure is not currently achieved due to wrong adjacency relations, topological definition issues and complex geometry of source data, it is more responsive and performs better than other established methods in terms of plane extraction, topology optimization, planarity calculation and mesh generation.

The content is interesting and well-organized and the methodology is sound and explained in detail, allowing replication for other 3D meshing cases. Personally, I appreciate that the authors analyze the limitations, the issues encountered during development and the problems derived under certain circumstances and cases. The writing style is articulate with a good technical dictionary. The overall evaluation is positive. Nevertheless, the following remarks can be referred to it and a minor revision would be welcome.

General Comments:

1. The introduction is very detailed presenting the general area of interest, stating the current knowledge regarding mesh simplification methods, polyhedral structures, optimization strategies based on valid topological relations and plane extraction and outlining the strengths and limitations of previous studies. It also, establishes the originality of the presented research. However, I would suggest to enhance the literature with a reference to meshing techniques exploited by open-source, free and/or commercial photogrammetric and 3D modelling software, like MeshLab which was actually used, Geomagic Studio (Wrap), VisualSFM, AliceVision, AgisoftMetashape etc, after SfM and MVS.
2. The authors should illustrate their workflow or specific processing phases with a flow-chart.
3. Due to the technical scope of the article, the following should be mentioned: i) the set of technological tools used for development/evaluation like programming software and language, as well as ii) the compatible graphics APIs, hardware platform and the minimum CPU/GPU specifications for attaining the optimal performance results.
4. Consider to mention some (or all) of these aspects: i) potential supported file formats (.obj, .ply, .gltf, .dae, .fbx etc), ii) precision of storing the final mesh data and iii) size (MB) of the final 3D models based on the naively stored position and normals (index/vertex buffers for normals)

Specific Comments:

• line 220. Consider splitting the long sentence into two sentences.
• line 307. Please rephrase: ‘Since the regional growth is realized according to…”
• line 402: Reduce the figure caption and/or include these details on the main body.
• line 547. Consider omitting ‘boring’
• line 595: Please specify the source of data/technology (image-based, laser scanning, LiDAR etc.) for the generation of your 3D models

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Dear authors,

you have written very interesting article according to my opinion, and to my field of interest. English language needs some improvements. You cited the literature correctly, but before reference in the text you have never added space in front of the brackets – please add.

The good points of the article are:

1. interesting topic regarding smart cities and twin cities 3D modeling problems.
2. comprehensive analysis

According to my opinion, the weak points of your article are:

1. design of the article
2. at times the paper is written like an report not like an article
3. method – ours?!
4. it would be nice to see real image of an object

The improvement should be done in the following:

Ad. 1)

I suggest that section 3 Results & analysis is divided into two sections:

- section 3 – Results

- section 4 – Analysis of achieved results or Discussion

According to suggestion you can rearrange the current section 3.

To my opinion the conclusion is “weak” and needs further improvements. In conclusion you only highlighted future work. It would be nice if you highlight benefits of your work.

Ad. 2)

At times the paper is written like a report not like an article – example: line 149-156. Please rearrange that.

Ad. 3)

You named your method – OURS. I suggest that you provide meaningful name for your method.

Ad. 4)

It would be nice to see real image of one object who is used as an example. Please see comments bellow for lines 491-492

Suggestion by “lines”:

- line 164 – please correct the section title

- line 205 – add space after figure 3

- line 483 – there is a solo letter s after ground?

- line 484 – acronym RMS used for the first time – please provide the full term (root mean square error?)

- line 491-492 – you stated: As can be seen from Figure 13, our model is similar to the real building. On figure 13 there real building is not shown. To my opinion it should be, and that is something that is missing in the whole article for all examples. It would ensure better understanding of your model in comparison with others, and it would ensure easier reading. I suggest that you provide at least for this model the real image of the building.

- line 493 – legend on figure 12 – are meters correct units?

- line 493 – what about the comparison with VSA?

- line 502 – table 2 – units for time are missing?

- line 525 - our method – please write capital O.

Best regards.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The manuscript has been revised, but my concerns stated in my first review have not been addressed in a sufficient way. The main unsolved concerns are the scientific contribution beyond SABMP and the scientific contribution of the 1-Ring-Patch aspect.

The scientific contribution beyond SABMP is still unclear. There is a figure 2 in the response to the reviewers’ comments, but is has not been included in the manuscript.  Why? The added value of your approach should be shown not only to the reviewer, but to the reader. You have to state explicitly the aspects in the manuscript which have not been considered in SABMP in detail. You have deleted the paragraph on ‘Optimization of candidate faces’ and refer to SABMP instead, reducing the added value compared to SABMP. In fig. 2, ‘Generating candidate faces’ is stated as your improvement, but also included in the SABMP part. Explain.

The contribution ‘1-ring-patch’ concept which is crucial for your approach (it appears on the title and is stated as improvement in fig 2 in the response to the reviewers’ comments) is still unclear. You aggregate triangles which are the ‘umbrella’ of one vertex. There is no specific structure (planarity) of these patches. In the explanation (lines 183-188) this concept is motivated by noise and a rugged surface. If you aggregate triangles incident to a vertex to an aggregated face, the degree of noise or ruggedness does not change (or do you change the geometry of the aggregated face which would result in gaps in the boundary if the face?).  Express the added value of the 1-ring-patch concept more precisely.

Other comments:

• Figure 1 is still unclear. There are some arrows missing. The text refers to a), b), c) d), but the parts of fig. 1 have no numbering. There are 12 parts in fig. 1, the text refers to four.
• The algorithmic structure of algorithm in 2.1 (generation of 1-ring-patch) is still not clear. I suppose you maintain a list of new center points which is initiated in step 1 (adding one vertex), processed in step 2/3 (processing of a single vertex and removing it from the list in step 2) and in step 3, new  center points are added to the list. This structure should be made explicit. 

Author Response

Please see the attachment

Author Response File: Author Response.docx