Voxel Dataset

The Voxel dataset is a constructed dataset of 3D shapes designed to present a unique problem for ML and NAS tools. Instead of a photo of a 3D object, we exploit ML's ability to work across N number of 'colour' channels and use this dimension as a third dimension for images.

This dataset is one of the three hidden datasets used by the 2024 NAS Unseen-Data Challenge.

The images include 70,000 generated 3D Images of seven different shapes that we generated by creating a 20x20x20 grid of points in 3d space, and randomly generated different 3D shapes (see below) and recorded which of the points the shape collided with, generating the voxel like shapes in the dataset.

The data has a shape of (n, 20, 20, 20) where n is the number of samples in the corresponding set (50,000 for training, 10,000 for validation, and 10,000 for testing).

For each class (shape), we generated 10,000 samples evenly distributed between the three sets.

The three classes and corresponding numerical labels are as follows:
Sphere: 0,
Cube: 1,
Cone: 2,
Cylinder: 3,
Ellipsoid: 4,
Cuboid: 5,
Pyramid: 6

NumPy (.npy) files can be opened through the NumPy Python library, using the `numpy.load()` function by inputting the path to the file into the function as a parameter. The metadata file contains some basic information about the datasets, and can be opened in many text editors such as vim, nano, notepad++, notepad, etc