Efficient way of building an uint32 numpy matrix from uint8

Question

I am storing a matrix (a python list of lists, not a numpy matrix) containing numbers in the range of [0;255] - representable well as uint8s. However, I would like to build a matrix which packs these numbers into 32bit ones.

Is there an efficient/nice way of doing this in numpy? Currently I'm rebuilding the matrix row by row with numpy.frombuffer() but it feels like there should be a more convenient way of achieving this.

Current code:

def convert_8to32bit_matrix(mat):
    ret_mat = np.zeros(shape=(mat.shape[0], int(mat.shape[1]/4)))
    for i, row in enumerate(mat):
        ret_mat[i] = np.frombuffer(row, dtype=np.uint32)

    return ret_mat

Answer 1

Use numpy.ndarray.astype method:

mat8 = np.array([[1, 2], [3, 4]]).astype(np.uint8)
mat32 = mat8.astype(np.uint32)

print(mat32.dtype) #uint32