What is the memory layout of the underlying data buffer used by the Tensorflow C API?

Question

I am using TF_TensorData to read the output from my graph, which is a tensor of dimensions {1, 520, 420, 1}. My objective is to pull out a 520 x 420 image of floats. I was wondering what is the layout of the underlying data buffer retrieved with TF_TensorData. On the input side, am I correct in assuming the layout of the data buffer passed to TF_NewTensor would be the same?

In the general case, what is the layout of a tensor of arbitrary dimensions?

Answer 1

For your two tensors, yes the layout is the same.

As per the documentation in tensorflow/c/c_api.h , data is laid out in row-major order.

Answer 2

{1, 520, 420, 1} is the same shape as {520 , 420}

On the input side, am I correct in assuming the layout of the data buffer passed to TF_NewTensor would be the same?

must be the same, just import some data and printout.