GeneratorMixture#

class deepinv.physics.generator.GeneratorMixture(generators, probs, use_batch_sampling=True, device='cpu', rng=None, verbose=False)[source]#

Bases: PhysicsGenerator

Base class for mixing multiple physics generators.

The mixture randomly selects a subset of batch elements to be generated by each generator according to the probabilities given in the constructor.

Parameters:

  • generators (list[PhysicsGenerator]) – the generators instantiated from deepinv.physics.generator.PhysicsGenerator.

  • probs (list[float]) – the probability of each generator to be used at each step

  • device (str) – device on which the generator is located, defaults to “cpu”

  • rng (torch.Generator) – a pseudorandom random number generator for the parameter generation. If None, a generator will be created on the specified device with a random seed.

  • verbose (bool) – whether to print warnings about the batch-compatibility of the generators, defaults to False.

Param:

bool use_batch_sampling: whether to sample a different generator for each element in the batch. This is only possible if all generators in the mixture produce parameters with the same keys and shapes. If not, a single generator will be sampled per batch. Defaults to True.

Examples:

Mixing two types of blur

>>> from deepinv.physics.generator import MotionBlurGenerator, DiffractionBlurGenerator
>>> from deepinv.physics.generator import GeneratorMixture
>>> _ = torch.manual_seed(0)
>>> g1 = MotionBlurGenerator(psf_size = (3, 3), num_channels = 1)
>>> g2 = DiffractionBlurGenerator(psf_size = (3, 3), num_channels = 1)
>>> generator = GeneratorMixture([g1, g2], [0.5, 0.5])
>>> params_dict = generator.step(batch_size=1)
>>> print(params_dict.keys())
dict_keys(['filter'])
step(batch_size=1, seed=None, **kwargs)[source]#

Returns a new set of physics’ parameters, according to the probabilities given in the constructor.

Parameters:

  • batch_size (int) – the number of samples to generate.

  • seed (int) – the seed for the random number generator.

Returns:

A dictionary with the new parameters, ie {param_name: param_value}.

Examples using GeneratorMixture:#

Tour of blur operators

Tour of blur operators