In order to demonstrate 2D kernel-based filtering without relying on library code too much, convolutions.py gives some examples to play around with.
image = cv2.imread('clock.jpg', cv2.IMREAD_GRAYSCALE).astype(float) / 255.0
kernel = np.array([[1, 0, -1],
[1, 0, -1],
[1, 0, -1]])
filtered = cv2.filter2D(src=image, kernel=kernel, ddepth=-1)
cv2.imshow('horizontal edges', filtered)In addition, convolution_manual.py implements a manual 2D convolution to explain the concept.