How to rewrite the cv2.resize() function?

Question

Python

Task: Reduce image size by 2 times
There is a function: resized = cv2.resize(img, dim, interpolation=cv2.INTER_CUBIC)
which uses bicubic interpolation.
They asked me to rewrite the assignment by hand.
That is, it is necessary to thin out the image by calculating the value of the thinned image as the arithmetic mean of four adjacent elements of the original image.
The cv2 library function works in the same way, but I have no idea how to rewrite it myself..

Answer 1

img = np.float32(img)
img = img[0::2, :] + img[1::2, :]
img = img[:, 0::2] + img[:, 1::2]
img = np.uint8(img / 4)

img = np.uint8(
    (np.float32(img[0::2, 0::2]) + img[0::2, 1::2] + img[1::2, 0::2] + img[1::2, 1::2]) / 4
)

kernel = np.ones((2, 2), dtype=np.float32) / 4
img = cv2.filter2D(img, cv2.CV_8U, kernel, anchor=(0, 0))[::2, ::2]

img = np.lib.stride_tricks.as_strided(
    img, (*np.array(img.shape[:2]) // 2, 4, 3), (*np.array(img.strides[:2]) * 2, *img.strides[1:])
).mean(axis=-2).astype(np.uint8)