ConvTranspose

Description

Transposed convolution

Input

Three inputs:

x: tensor of type float16.

filter: tensor of type float16.

bias: (optional) 1D tensor of type float16.

Output

One output:

y: tensor of the identical data type and shape as input x.

Attribute

auto_pad: string. Defaults to NOTSET, indicating that padding is explicitly used.

dilations: ints. The default value is all-1s. It indicates the hole value of each axis of the filter.

group: int. Number of input channel groups. Defaults to 1.

kernel_shape: ints. Defaults to the shape of the input filter.

output_padding: ints. Defaults to an all-0 array.

output_shape: ints, output shape, which is automatically calculated based on the pad.

pads: ints. The default value is an all-0 matrix. The pad value is specified for each axis.

strides: ints, stride of each axis. Defaults to an all-1 matrix.

ONNX Opset Support

Opset v8/v9/v10/v11/v12/v13/v14/v15/v16/v17/v18