CNN Convolution — Sliding Kernel & Feature Map
Feature map[i,j] = ReLU(Σ kernel · patch(i,j))
Filter
Preset kernel:
Horizontal edge
Vertical edge
Sobel X
Gaussian blur
Sharpening
Emboss
Stride:
1
Padding:
0
Animate Scan
Stop
Random Image
Checkerboard
Convolution:
slide 3×3 kernel across input, compute dot product at each position.
ReLU:
max(0, x) — negative activations become zero.
Output size = ⌊(N - K + 2P)/S⌋ + 1