CNN Feature Map Explorer

Convolutional neural networks build their understanding of an image one small sliding window at a time. This simulator visualizes that process directly: a 2D input feature map sits on the left, and a kernel window sweeps across it row by row, computing a dot product between its weights and the patch of pixels underneath at every stop. Each dot product becomes one activation in the output feature map on the right, which fills in progressively as the kernel finishes its sweep. Adjust kernel size, stride and the number of parallel filters to see how each reshapes the output map's resolution and how many independent pattern detectors run side by side in a real convolutional layer.