convTranspose(weight:strides:padding:outputPaddingValues:dilations:groupCount:)
Adds a transposed convolution operation to the current graph.
Declaration
func convTranspose(weight: some BNNSGraph.Builder.OperationParameter<T>, strides: [Int], padding: BNNSGraph.Builder.ConvolutionPadding, outputPaddingValues: [Int]? = nil, dilations: [Int]? = nil, groupCount: Int = 1) -> BNNSGraph.Builder.Tensor<T>Parameters
- weight:
The convolution weights.
- strides:
An array that contains the kernel stride for each spatial dimension.
- padding:
An enumeration that specifies that the operation computes the padding from the input and output shapes. If you specify
custom, pass twice the number of spatial dimensions. - dilations:
An array that contains the kernel dilation for each spatial dimension.
- groupCount:
The number of convolution groups. Pass
1to specify non-grouped convolution.
Discussion
This operation assumes the source tensor is in (N, C, spatial dimensions...) format, where the number of spatial dimensions is either 1 or 2.