CNN Feature Map Explorer | mysimulator.uk

Convolution

Kernel size 3×3

Stride 1

Filters (layer depth) 3

Slide speed 1.0×

Input map

–

Output map

–

Kernel position

–

Active filter

–

How it works

A convolutional layer slides a small learnable kernel across the input feature map. At every position the kernel's weights are multiplied element-wise with the patch of pixels underneath it and summed into a single number — that becomes one pixel of the output feature map. Sliding the kernel over the whole input builds the entire next layer, and stacking several kernels (filters) in parallel lets the layer detect several different patterns — edges, corners, textures — at once, each producing its own output map.

out[i,j] = Σ_u Σ_v  in[i·s+u, j·s+v] · K[u,v]
outSize = floor((inSize − kernelSize) / stride) + 1

Kernel size — the width/height of the sliding filter window; a bigger kernel sees more context per step but produces a smaller output map.
Stride — how many pixels the kernel jumps between steps; a stride of 2 skips positions and roughly halves the output map's resolution.
Filters (layer depth) — how many independent kernels run in parallel over the same input, each becoming its own slab in the output activation volume.
Output map — the resulting grid of activations, shown growing pixel-by-pixel as the kernel finishes each position; its size is derived directly from kernel size and stride via the formula above.

Real-world relevance: this exact sliding dot-product is the core operation inside every CNN image classifier — stacking many such convolutional layers is how networks like ResNet and VGG build from raw pixels up to high-level concepts like "eye" or "wheel".

Input feature map Kernel window Output activation

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.