Really well-structured walkthrough — the progression from continuous to discrete convolution before jumping into image kernels is exactly the right order to build intuition. The explanation of why odd-sized kernels are necessary (clean center pixel, symmetric padding) is something most tutorials skip entirely. One thing I found particularly valuable is the note on Laplacian of Gaussian (LoG) — the fact that the second derivative amplifies noise is a detail that only clicks after you've seen the Laplacian fire on texture regions you didn't expect. The NumPy implementation makes it very concrete too. Looking forward to the next part of the Computer Vision series — curious if you'll cover how learned kernels in CNNs relate back to these hand-crafted ones like Sobel and Laplacian.