What happens to the tensor when I change its type from Float to Long
This is supposed to be a segmentation mask of an image, and it gets completely wrecked when I change it to LongTensor.
Since you are working with floating point values, you would have to be careful about transforming these values to long, as you might lose all information.
E.g. assuming your tensor contains values in [0, 1] converting them to long could round all values to 0:
x = np.random.uniform(0, 1, (100, 100, 3))
plt.imshow(x) # shows valid image
plt.imshow(x.astype(np.int64)) # blank image
print(np.unique(x.astype(np.int64)))
> array([0], dtype=int64)