Section 3.1: training inputs after flattening

[konistis]

Hi folks,
I would like some insight regarding section 3.1 (computer vision prior to the use of CNN).
When we setup the model training per batch,
the input tensors are [32, 1, 28, 28] and predicted tensors are [32, 784].
After flattening, the input tensor will become [32, 784],
I was used to the notion of having '1-d' vectors as inputs for training but
I see that the input looks more like a matrix now. Can you shed some light how this works?

Thank you for your time,
Kostas

btw, the class is AWESOME!