spyrit.core.noise.PoissonApproxGaussSameNoise.forward

PoissonApproxGaussSameNoise.forward(x: tensor) → tensor

Simulates measurements corrupted by Poisson noise

Args:

x: Batch of images. The input is directly passed to the measurement operator, so its shape depends on the type of the measurement operator.

Shape:

x: \((*, h, w)\) if self.meas_op is a static measurement operator, \((*, t, c, h, w)\) if it is a dynamic measurement operator.

Output: \((*, M)\) (static measurements) or (*, c, M) (dynamic measurements)

Example 1: Two noisy measurement vectors from a Linear measurement operator

>>> H = torch.rand([400,32*32])
>>> meas_op = Linear(H)
>>> noise_op = PoissonApproxGaussSameNoise(meas_op, 10.0)
>>> x = torch.FloatTensor(10, 32*32).uniform_(-1, 1)
>>> y = noise_op(x)
>>> print(y.shape)
>>> print(f"Measurements in ({torch.min(y):.2f} , {torch.max(y):.2f})")
>>> y = noise_op(x)
>>> print(f"Measurements in ({torch.min(y):.2f} , {torch.max(y):.2f})")
torch.Size([10, 400])
Measurements in (2255.57 , 2911.18)
Measurements in (2226.49 , 2934.42)

Example 2: Two noisy measurement vectors from a HadamSplit operator

>>> Perm = torch.rand([32*32,32*32])
>>> meas_op = HadamSplit(H, Perm, 32, 32)
>>> noise_op = PoissonApproxGaussSameNoise(meas_op, 200.0)
>>> x = torch.FloatTensor(10, 32*32).uniform_(-1, 1)
>>> y = noise_op(x)
>>> print(y.shape)
>>> print(f"Measurements in ({torch.min(y):.2f} , {torch.max(y):.2f})")
>>> y = noise_op(x)
>>> print(f"Measurements in ({torch.min(y):.2f} , {torch.max(y):.2f})")
torch.Size([10, 800])
Measurements in (0.00 , 55951.41)
Measurements in (0.00 , 56216.86)