When we created the segmentation dataset, some models failed to generate high quality watertight meshes with close to 2500 edges. This prevented them being used for the MeshCNN benchmark. Consequently they were removed from the dataset.
What?
With the growing interest in B-Rep machine learning, and the availability of an open source pipeline for B-Rep data processing, it is considered useful for the community to provide the models which could not be meshed for the benchmark in the BRepNet paper. Hence the "Extended STEP Segmentation Dataset" provides these additional models along with the corresponding segmentation information. The files are provided in a format which is a "drop-in" replacement for the s2.0.0 dataset, so all scripts work on the extended dataset out of the box.
This PR adds a download link for the extended step dataset and includes some additional documentation. This includes a comparison of the distribution of faces and modeling operations when using the standard s2.0.0 dataset and the extended dataset.
karldd
commented
3 years ago
Why?
When we created the segmentation dataset, some models failed to generate high quality watertight meshes with close to 2500 edges. This prevented them being used for the MeshCNN benchmark. Consequently they were removed from the dataset.
What?
With the growing interest in B-Rep machine learning, and the availability of an open source pipeline for B-Rep data processing, it is considered useful for the community to provide the models which could not be meshed for the benchmark in the BRepNet paper. Hence the "Extended STEP Segmentation Dataset" provides these additional models along with the corresponding segmentation information. The files are provided in a format which is a "drop-in" replacement for the s2.0.0 dataset, so all scripts work on the extended dataset out of the box.
This PR adds a download link for the extended step dataset and includes some additional documentation. This includes a comparison of the distribution of faces and modeling operations when using the standard s2.0.0 dataset and the extended dataset.