CNN Receptive Field Growth Across Layers

A convolutional neural network never looks at the whole image at once inside a single layer — each output unit only "sees" a small, growing patch of the input, called its receptive field. This simulator stacks real convolutional feature maps in 3D, picks one output neuron at the top of the stack, and traces the exact recurrence that determines how far back down through the layers — and how wide across the original input — that neuron's influence reaches. Adjust the number of layers, kernel size, stride and dilation and watch the highlighted receptive-field patch expand or contract on every layer live, with the effective receptive-field size, output feature-map resolution and cumulative stride ("jump") read out numerically.