spyrit.core.prep.DirectPoisson

class spyrit.core.prep.DirectPoisson(alpha: float, meas_op)

Bases: Module

Preprocess the raw data acquired with a direct measurement operator assuming Poisson noise. It also compensates for the affine transformation applied to the images to get positive intensities.

It computes \(m = \frac{2}{\alpha}y - H1\) and the variance \(\sigma^2 = 4\frac{y}{\alpha^{2}}\), where \(y = Hx\) are obtained using a direct linear measurement operator (see spyrit.core.Linear), \(\alpha\) is the image intensity, and 1 is the all-ones vector.

Args:

alpha: maximun image intensity \(\alpha\) (in counts)

meas_op: measurement operator (see meas)

Example:

>>> H = torch.rand([400,32*32])
>>> meas_op =  Linear(H)
>>> prep_op = DirectPoisson(1.0, meas_op)

Methods

denormalize_expe(x, beta[, h, w])	Denormalize images from the range [-1;1] to the range [0; \(\beta\)]
forward(x)	Preprocess measurements to compensate for the affine image normalization
sigma(x)	Estimates the variance of raw measurements
unsplit(x[, mode])	Unsplits measurements by combining odd and even indices.