Structuring of the data about eigenvector-grouping HOT 3 CLOSED

Input point clouds should be of the format: $N \times 4$, where $N$ - number of points, and for each point there is a tuple $(x, y, z, L)$ where $(x, y, z)$ are 3D coordinates of a given point and $L$ is a GT label. If you have data in a different format you can just plug your own transforms since in the training_step in SegmentatorModule point cloud is already split into coordinates $(x, y, z)$ and labels $(L,)$.

from eigenvector-grouping.

Thanks for your clear explanation!

I assume that this is a format for the point clouds of one patient. How do I structure the point clouds of multiple patients?

from eigenvector-grouping.

Yes, that's for one patient, I stored point clouds for each patient in separate files. I'm using Monai dataset, which expects data to be provided as a list of dictionaries, so with separate files for each patient, my input list to the dataset object looks like this:

[
    {"input": "path/to/sample_1.npy"},
    {"input": "path/to/sample_2.npy"},
    ...
    {"input": "path/to/sample_n.npy"}
]

If you store all patients in one file you can just load it a priori, built a list of format provided above and skip LoadImaged transform.

from eigenvector-grouping.