Hi,

I have tried to draw the architecture of the model by hand according to the article, and then compared the original author's implementation in TensorFlow and PointNetClassifier in this repo. I might be wrong, but it seems that the PointNetClassifier is equivalent to Input transform. Below is the copy of TensorFlow implementation:
`def input_transf

In las file format, vlrs info is crucial to correctly describe point's spatial location.
However, if you try to use pyntcloud.las_header.vlrs, you wouldn't get anything, the value is alway none.

Through your source code, I can see you try to use
data["las_header"] = las.header
But it can not pass vlrs or evlrs info into new object, because in laspy, vlrs info is dynamically obtained afterw

I tired to execute the this funtion def create_labels(places, size, corners, resolution = 0.5, x=(0, 90), y=(-50, 50), z=(-4.5, 5.5), scale=4, min_value=np.array([0., -50., -4.5])): but the returned variables sphere_center, train_corners have the size of (0,3) and (0,8,3) but they are empty.
Any idea about that?
Thank you