2D Convolution Visualization - Interactive Visualization

Convolution slides a small weight matrix (kernel) across an input, computing a dot product at each position to produce a feature map. The same kernel is applied everywhere - parameter sharing - so the network learns a feature detector (like an edge) and applies it everywhere in the image simultaneously. Stride controls how far the kernel moves each step; padding controls whether output dimensions shrink.

• Watch the kernel window slide across the input grid and see each output cell computed as an element-wise dot product
• Try classic kernels: Sobel edge detection, Gaussian blur, sharpening - see their effect immediately on the feature map
• Understand stride: stride=2 halves spatial dimensions and reduces computation, used in pooling-free modern architectures
• Learn zero-padding vs valid convolution - padding maintains spatial size across layers, essential for deep networks
• Grasp receptive field: a 3x3 kernel in layer 5 "sees" a much larger region of the original input due to stacking

Part of the EngineersOfAI Interactive 3D - free interactive visualizations covering every major concept in machine learning and AI engineering. Hover any element for a plain-English explanation. No code required.