Generate point cloud representation of the .vtk 3d object. This representation is compatible with PyTorch Geometric library.
First implement read_data_objects function in utils.py. You need to represent each 3d object as a dict with following attributes:
{
'path': (string): absolute path to this .vtk object
'y': (int): label of the object
... and optional attributes
}Note that you can pass more attributes to the dict. Those will be automatically passed to the PyTorch Geometric abstraction as well.
To generate the dataset use generate.py script
python generate.py \
--path /path/to/dataset \
--num_points 1024Args:
patharguments means where your dataset will be stored. Dataset generated topathjust once. All instances of Dataset object (e.g. in your train code) with existing path will just read the data instead of processing it.num_pointsNumber of points uniformly sampled from the shape surface