HW1: Experimenting with Gaussian kernel affects further results

[aiphae]

In Part 3: Canny Edge Detector, it is necessary to implement gaussian_kernel, and then it says:

# Test with different kernel_size and sigma <--------------
kernel_size = 5
sigma = 1.4

# Load image
img = io.imread('iguana.png', as_gray=True)

# Define 5x5 Gaussian kernel with std = sigma
kernel = gaussian_kernel(kernel_size, sigma)

# Convolve image with kernel to achieve a smoothed effect
smoothed = conv(img, kernel) <-------------- Shouldn't be used later!

...

However, the smoothed image is then used later throughout the non-maximum supression part. And experimenting (as said!) with kernel size and sigma values causes false inconsistencies between own algorithm and the reference iguana_non_max_suppressed.npy.

Proposed change:
Define the Gaussian kernel and produce a smoothed image AGAIN before testing the non-maximum supression algorithm.