azula.guidance.tmpd

Tweedie Moment Projected Diffusion (TMPD) internals.

References

Tweedie Moment Projected Diffusions For Inverse Problems (Boys et al., 2023)
https://arxiv.org/abs/2310.06721

Classes

TMPDenoiser

Creates a TMPD denoiser module.

Descriptions

class azula.guidance.tmpd.TMPDenoiser(denoiser, y, A, var_y)

Creates a TMPD denoiser module.

Parameters:

  • denoiser (Denoiser) – A denoiser \(q_\phi(X \mid X_t)\).

  • y (Tensor) – An observation \(y \sim \mathcal{N}(A x, \Sigma_y)\).

  • A (Callable[[Tensor], Tensor]) – The forward operator \(x \mapsto A x\).

  • var_y (float | Tensor) – The noise variance \(\Sigma_y\).