Topological Data Analysis-friendly CAD/3D point cloud dataset [P]

Hi everyone,

I’m looking for a suitable 3D point cloud dataset — or a CAD/mesh dataset from which I can sample point clouds — for a small research/report project.

The goal is to compare Topological Data Analysis (TDA) as a preprocessing / feature extraction method against more standard 3D point cloud preprocessing methods, under different perturbations such as:

• Gaussian jitter / noise
• random point deletion / subsampling
• small deformations
• scaling / rotations
• outliers or other synthetic corruptions

The comparison would be based on the classification accuracy of a downstream model after preprocessing.

I do not necessarily need many classes. Even a binary classification dataset would be enough. What matters most is that the classes should differ in their topological structure, ideally in the number of holes / loops / cavities, so that TDA has a meaningful signal to detect.

For example, something like:

• sphere / ball-like objects vs torus / ring-like objects
• solid object vs object with a tunnel
• objects with different numbers of handles or holes

Ideally, each class should contain many samples (600+), or the dataset should contain enough CAD/mesh models so that I can sample many point clouds from them.

Does anyone know of a dataset that fits this description? I would also appreciate suggestions for CAD repositories, synthetic dataset generators, or benchmark datasets where such class pairs could be extracted.

Thanks!