I'm using nvidia3090 with cuda11.3. The OOM error happens with building 50 or 300. I don't konw whether it relies on the software or hardware environment. May I konw the hardware requirments? Thanks!

Dear authors,

Thanks for your work. Have you provided the data for BuildingNet-Mesh? Could you please introduce how to extract the mesh file from the given datasets? I would like to access the 3D mesh model for each building.

Best,

Hi, I'd like to try and make my own dataset of different buildings with parts other than the ones included with BuildingNet. Generating the 3D models (and point clouds) themselves shouldn't be a problem - it's everything else I'm not sure about.

For the minkowski pretrained features, should I simply download their code (https://github.com/NVIDIA/MinkowskiEngine) and run the models through it?

The required json files for each model used to train BuildingNet are fairly extensive (adjacency, containment, support, and similarity) and they don't seem especially trivial to recreate. The code for their creation doesn't appear to be in this repository, and I'm guessing it is part of the labeling application shown in the paper. Can the code for the labeling application be downloaded? I'm hoping in the end to output the label data automatically together with the 3D models when they are generated (that is, a synthetic training data generation pipeline) and being able to see some of that code would be a huge help.

Hi, authors,

For the usage of visualization, could you provide the color map for those 31 part labels (as mentioned in Table 7 in the original paper)?

Thanks~

Hi, authors,

When I load the obj file RESIDENTIALvilla_mesh6487.obj, using trimesh package, the loaded object doesn't contain the general property vertices and faces.

import trimesh
mesh = trimesh.load("RESIDENTIALvilla_mesh6487.obj")

Although the loaded object can be visualize by mesh.show(), but it's not convenient to access its vertices and face list. Is there any way to directly load the property vertices and faces?

Hi, authors,

According to the original paper, the representation for each node (subgroup) is 41-dim, which corresponds to four kinds of spatial features, such as: i) initial node representation, ii) 3D barycenter position of subgroup, iii) mesh surface area, iv) coordinates of the opposite corners of OBB. Could you elaborate on the specific feature dimensions related to the above four kinds of spatial features? e.g. how many dimensions does the initial node representation contain?

Thanks~

Dear authors, Thanks for your work. When I use the point cloud of the building_net dataset, I find a lot of problems in the point cloud. I want to look for your help. I will give a few examples to illustrate as follow ： （1）in the COMMERCIALchurch_mesh3114 and COMMERCIALcity_hall_mesh0514, the undetermined point clouds clearly should belong to other categories, however, they are labeled as 0. Every point cloud file has the same problem.
（2）in the COMMERCIALhotel_building_mesh0130, There are points in the column category that should belong to the railing and window. There are other similar misclassifications. These errors are found through visualization. (3) Other point clouds have the same problem. I think it will take a lot of effort to correct these mistakes， thank you for your dedication. will you release a new version? I want to use the dataset to support your work, But the error rate of the current version of the dataset is difficult to use. I hope my suggestions are helpful to you.