complextorch.nn.modules.upsampling#

Two flavors of complex-valued upsampling / interpolation:

Upsample — split form. Interpolates the real and imaginary parts independently. Matches the behavior of torchcvnn.nn.Upsample and complexPyTorch.complex_upsample.
PolarUpsample — polar form. Interpolates the magnitude \(|z|\) and the phase \(\arg z\) independently, then recombines via \(|z| \cdot \exp(j\,\arg z)\). Phase-preserving along smooth phase regions; useful for coherent signal models (radar, SAR). The cost is a visible discontinuity wherever the phase wraps from \(-\pi\) to \(+\pi\) — neither form is universally correct.

Classes#

`PolarUpsample`	Complex-Valued Upsample (polar form)
`Upsample`	Complex-Valued Upsample (split form)

Module Contents#

class complextorch.nn.modules.upsampling.PolarUpsample(size: _SizeT = None, scale_factor: float | tuple[float, Ellipsis] | None = None, mode: str = 'nearest', align_corners: bool | None = None, recompute_scale_factor: bool | None = None)[source]#

Bases: torch.nn.Module

Complex-Valued Upsample (polar form)#

Interpolates magnitude and phase independently. For a complex input \(z = |z|\, e^{j\arg z}\):

\[|z|' = \text{interp}(|z|), \quad \arg z' = \text{interp}(\arg z), \quad z' = |z|' \cdot e^{j\,\arg z'}\]

Phase-preserving along smooth phase regions but introduces discontinuities at phase wraps (\(\pm\pi\)). Choose between Upsample (split) and PolarUpsample (polar) based on whether your data has smooth phase (favor polar) or smooth real/imag parts (favor split).

All keyword arguments mirror torch.nn.Upsample.

Initialize internal Module state, shared by both nn.Module and ScriptModule.

extra_repr() → str[source]#

Return the extra representation of the module.

To print customized extra information, you should re-implement this method in your own modules. Both single-line and multi-line strings are acceptable.

forward(input: torch.Tensor) → torch.Tensor[source]#

align_corners = None#

mode = 'nearest'#

recompute_scale_factor = None#

scale_factor = None#

size = None#

class complextorch.nn.modules.upsampling.Upsample(size: _SizeT = None, scale_factor: float | tuple[float, Ellipsis] | None = None, mode: str = 'nearest', align_corners: bool | None = None, recompute_scale_factor: bool | None = None)[source]#