Conv3D

3-dimensional (volumetric) convolution.

A convolution is the operation of sliding a moving window, also called a kernel, across the input data and for each kernel position we calculate a weighted average and add a bias. The weights and bias are learnable parameters. This node applies a 3D moving window.

The only supported input data type is Float32, and the input must have at least 4 dimensions. The output will have the same data type as the input.

By clicking the node the following parameters can be configured on the right panel:

• Filters: Number of convolutions to perform. This will be se size of the first dimension in the output.
• Kernel size: The size of the moving window.
• Stride: Number of places to move the moving window after each computation.
• Dilation: Gap between each position in the kernel.
• Use bias: If true, a learnable parameter of shape (<Filters>) is added to the output after the convolution is done.
• Padding: One of None, Same and Custom. If None, then no padding will be added to the input. Is Same, padding will be added to the input so that the output shape is the same as the input shape. If Custom, you can manually specify the amount of padding to use.
• Padding mode: Defines what values to fill the padding with. One of Zeros, Reflect, Replicate or Circular. If Zeros, then values of 0 will be added in the padding. If Reflect, then the padding will contain the input data, but reflected around the edge. If Replicate, then the padding will contain the value that is closest to the edge of the input. If Circular, then the padding will contain the input data but flipped around to the other side.
• Number of connections: Increases the number of connections on this node. The first input will lead to the first output, the second input to the second output and so on. This is useful because even though the connections are separate they share the same weight and bias.
• Weight initializer: Initial value for the weight.
• Bias initializer: Initial value for the bias.