B-Rep Distance Functions (BR-DF)
How to Represent a B-Rep Model by Volumetric Distance Functions?

The code has not been fully cleaned, and the current README is not yet readable. The final version will be released by early this year.

We present a novel geometric representation for CAD Boundary Representation (B-Rep) based on volumetric distance functions, dubbed B-Rep Distance Functions (BR-DF). BR-DF encodes the surface mesh geometry of a CAD model as signed distance function (SDF). B-Rep vertices, edges, faces and their topology information are encoded as per-face unsigned distance functions (UDFs). An extension of the classical Marching Cubes algorithm converts BR-DF directly into watertight CAD B-Rep model.

Requirements

• Linux
• Python 3.9
• CUDA 11.8
• PyTorch 2.2
• Diffusers 0.27

Data

Download ABC STEP files (100 folders).

Extract SDF voxel and UDF voxels from STEP files.

bash data_process_script.sh

Training

1. Download checkpoints "abc folder" from the above link.
2. Download "pkl.tar" from https://1sfu-my.sharepoint.com/:f:/g/personal/fuyangz_sfu_ca/EoBgkMc1LZZLkCFsQKFV2B0Bjnr5QLuop76jYwTpK3NyjA?e=cFBB6n. This is the pre computed latent of all abc data.
3. unzip pkl.tar and put at the location: brep_proj/data/latent_cache/pkl
4. run: python bbox_sdf_diffusion/run.py --config bbox_sdf_diffusion/train_large.yaml

Inside yaml, trainer_params->devices is the number of GPUs, trainer_params->num_nodes is the number of cluster nodes.

Testing

python bbox_sdf_diffusion/sample.py

Pretrained Checkpoint

download checkpoint folder from https://1sfu-my.sharepoint.com/:f:/g/personal/fuyangz_sfu_ca/EjjVLHgS1UVElW46mgsfFj8BAhRcTz_2wuxowGjhuBbR-w?e=FfF5WN

Then run,

python bbox_sdf_diffusion/sample.py