tnmf.backends.PyTorch_FFT

A module that provides a PyTorch based backend for computing the gradients of the factorization model. Shift-invariance is implemented via efficient convolution operations in Fourier space.

Module Contents

Classes

PyTorch_FFT_Backend

Reconstruction is performed via an efficient convolution in Fourier space.

class tnmf.backends.PyTorch_FFT.PyTorch_FFT_Backend(**kwargs)

Bases: tnmf.backends._PyTorchBackend.PyTorchBackend

Reconstruction is performed via an efficient convolution in Fourier space.

reconstruct(self, W: torch.Tensor, H: torch.Tensor) → torch.Tensor

Compute sum_m H[n, m, _, … ] * W[_ , m, c, …] –> R[n, c, …]