Convolution Operations

A convolution operation slides a filter (kernel) over an input feature map, computing dot products to produce output feature maps.

Key Parameters

• Kernel size: Spatial extent of the filter (3×3 most common)
• Stride: Step size when sliding (1 = every pixel, 2 = skip)
• Padding: Zeros added around input to control output size
• Channels: Number of filters = number of output channels

Output Size Formula

$$H_{out} = \left\lfloor\frac{H_{in} + 2P - K}{S}\right\rfloor + 1$$