bizerfr/bpnet

BPNet

BPNet: Bézier Primitive Segmentation on 3D Point Clouds (IJCAI-23)

Data Preparation

Please download the pre-processed ABC dataset.

unzip the dataset and put it in the data folder, or modify the data root path in the options.py

Training

configure your training settings in options.py, and then:

python train.py

Testing

configure your testing settings in options.py, and then:

python test.py

bpnet's Issues

Code release

Hi,

I am reading your publication; It is quite an excellent work!
Could you please release the source code?

Thank you!

Inference and labels

Dear Authors,

thank you very much for providing this amazing model! can you please give me an example, how can I run the pre-trained on model on mesh file? Regarding labeling, does the inference mesh become labels?

Best,
Ahmad

Reconstructing Bezier Surfaces

Hello,

I am currently working on a project involving point cloud data and Bezier surfaces reconstruction. After exploring your BPNet, I have a specific query that I'm hoping to get some insights on.

Is it possible to use the output parameters obtained from BPNet to reconstruct Bezier surfaces from point cloud data? If yes, could you provide some guidance or reference materials on how this can be achieved effectively using your system?

Looking forward to your insights.

Thank you for your time and assistance.

Pre-processing Training Data

Hi,

I would like to retrain the model with the same training dataset but with other labels. In the paper, you mentioned that you did the pre-processing/labeling using CGAL and OpenCascade data set. I am wondering, how did you generate the training dataset? I am actually interested in 10-18 Labels so I would like to retrain your model again. Thx!

Best,
Ahmad

non normal pretrained model

It seems like your pretrained model trained with data having normals. So while I am inferencing the with not normal input, there will be a mismatch between my input and your pretrained model ! So is there any chance to fix it without training the new model or whether you have the non-normal pretrained model ?