conv2d_transpose¶

dragon.vm.tensorflow.nn.conv2d_transpose(
  input,
  filters,
  output_shape,
  strides,
  padding='SAME',
  data_format='NHWC',
  dilations=None,
  name=None
)[source]¶

Apply the 2d deconvolution.

The spatial output dimension is computed as:

\[\begin{cases} \text{DK}_{size} = dilation * (\text{K}_{size} - 1) + 1 \\ \text{Dim}_{out} = (\text{Dim}_{in} - 1) * stride + \text{DK}_{size} - 2 * pad \end{cases} \]

Parameters:

• input (dragon.Tensor) – The input tensor.
• filters (dragon.Tensor) – The weight tensor.
• output_shape (Union[Sequence[int], dragon.Tensor], optional) – The determined shape of output.
• strides (Sequence[int]) – The stride(s) of sliding window.
• padding (Union[{'VALID', 'SAME'}, Sequence[int]]) – The padding algorithm or padding sizes.
• data_format ({'NCHW', 'NHWC'}, optional) – The optional data format.
• dilations (Sequence[int], optional) – The rate(s) of dilated kernel.
• name (str, optional) – A optional name for the operation.

Returns:

dragon.Tensor – The output tensor.