Conv2d

2D Convolutional Layer

Applies a 2D convolution over an input signal composed of several input planes. Fundamental building block for computer vision models (CNNs).

Parameters:

• in_channels: Number of channels in the input image
• out_channels: Number of channels produced by the convolution
• kernel_size: Size of the convolving kernel (int or (height, width))
• stride: Stride of the convolution (default: 1)
• padding: Zero-padding added to both sides (default: 0)
• dilation: Spacing between kernel elements (default: 1)
• groups: Number of blocked connections (default: 1)
• bias: If true, adds learnable bias (default: true)

Shape Contract:

• Input: [batch, in_channels, height, width]
• Output: [batch, out_channels, out_height, out_width]

Notes:

• Output spatial size: (input_size + 2*padding - kernel_size) / stride + 1
• Common patterns: 3x3 kernel, stride=1, padding=1 (preserves size)
• Depthwise: groups=in_channels, separable convolutions

Signature

neuron Conv2d(in_channels, out_channels, kernel_size, stride=Int(1), padding=Int(0), dilation=Int(1), groups=Int(1), bias=Bool(true))

Ports

Inputs:

• default: [batch, in_channels, height, width]

Outputs:

• default: [batch, out_channels, *, *]

Implementation

Source { source: "core", path: "convolutions/Conv2d" }