I have a dataset [ge.in.tum.de] that contains animated volumetric smoke density. The dataset is stored as NPZ files (numpy array save file). How can I load the density of a saved smoke simulation as an animated volume to render it? At this point I would also be glad to at lest load one frame.

For context about the float data in the volumes. This is what simulation number 70 snapshot 150 looks like in matplotlib with a viridis color map, but the data in each voxel are simple float64 values. The other snapshops and simulations look similar.



This is how you load a single npz file in python, if you want to try it with the dataset. It just generates a generic numpy float array. So you could just make a tiny numpy array filled with random data yourself for the purpose of testing like this np.random.rand(3,3,3). 

import numpy as np
def npz_to_nparray(filepath):
    """
    loads npz file from location
    :param filepath: string to file location
    :return: numpy array 
    """
    data = np.load(filepath)
    data = data[data.files[0]]  # normal strange operations to get the the actual array inside
    data = np.transpose(data, (3, 1, 0, 2))  # turning the inner array layer outside and rightside up
    return data[0]

I have tried:

-I can load the npz files in python and in Houdini's python shell as a numpy array. I dont know how to put the data into a Houdini volume or which file format Houdini needs to import it.

-I tried to translate the npz files to vdb with the pyopenvdb libary for python. But after multiple days of trying I am unable to import pyopenvdb. I build pyopenvdb library on windows with vcpkg and cmake, but I cannot import the build package. I setup an ubuntu VM and tried importing pyopenvdb there, but I can not import it.

tl;dr: How to import animated smoke voxel data that is currently stored as a 4D python array?