How does barracuda tensor convert an array of float into Tensor of shape (1, SIZE, SIZE, 3)

Question

Im using Barracuda in Unity to do real-time inference on a neural network model by drawing webcam images. Currently I'm struggling to do image processing as efficiently as with Python. Output is a mess. I believe most of my tensor are upside down or not the exact values that I want to feed into model...

Does anyone know if, in the Unity barracuda framework:

Given:

float[] TransformedPixels = {1, 2, 3, 4, 5, 6, 7,8 ,9, 10, 11, 12};

return new Tensor(1, IMAGE_SIZE=2, IMAGE_SIZE=2, 3, transformedPixels);

What kind of Tensor will I receive as the numpy equivalent in Python?

1.
[[[1, 2, 3],
[4, 5, 6]],
[[7, 8, 9],
[10, 11, 12]]]

or 

2.
[[[1,2],
[3,4]],

[[5,6],
[7,8]]],

[[9,10],
[11,12]]]

Type 1 is what i wish I would get because using np.reshape(TransformedPixels, (1,64,64,3)) would get me the same thing and it's more intuitive for us ML scientists. Anyway does Unity Barracuda has a way to print/debug logout a tensor with all the brackets so that we can check what values are we feeding into the model exactly?