ReflectionPad2d¶

class dragon.vm.torch.nn.ReflectionPad2d(padding)[source]¶

Pad input according to the last 2-dimensions by reflecting boundary.

The padded dimension is computed as:

\[\text{Dim}_{out} = \text{Dim}_{in} + pad_l + pad_r \]

Examples:

m = torch.nn.ReflectionPad2d(1)
x = torch.randn(1, 4, 4)
y = m(x)  # (1, 4, 4) -> (1, 6, 6)

See also

torch.nn.functional.pad(…)