Artifact2ArtifactSplittingMaskGenerator#

class deepinv.physics.generator.Artifact2ArtifactSplittingMaskGenerator(img_size, split_size=2, device='cpu', rng=None)[source]#

Bases: Phase2PhaseSplittingMaskGenerator

Artifact2Artifact splitting mask generator for dynamic data.

To be exclusively used with deepinv.loss.mri.Artifact2ArtifactLoss. Randomly selects a chunk from dynamic data (i.e. data of shape (B, C, T, H, W)) in the T dimension and puts zeros in the rest of the mask.

Artifact2Artifact was introduced by Liu et al.[1].

If input_mask not passed, a blank input mask is used instead.

Parameters:

  • img_size (tuple[int]) – size of the tensor to be masked without batch dimension of shape (C, T, H, W). Note this can be overriden on-the-fly by passing in img_size or input_mask arguments to step.

  • split_size (int, tuple[int]) – time-length of chunk. Must divide img_size[1] exactly. If tuple, one is randomly selected each time.

  • device (str, torch.device) – device where the tensor is stored (default: ‘cpu’).

  • rng (torch.Generator) – torch random number generator.

References:

batch_step(input_mask=None, img_size=None, persist_prev=False)[source]#

Create one batch of splitting mask.

Parameters:

  • input_mask (torch.Tensor, None) – optional mask to be split. If None, all pixels are considered. If not None, only pixels where mask==1 are considered. Batch dimension should not be included in shape.

  • img_size (tuple) – if not None, generate masks of this 2D image shape and override img_size attribute, must be of form (H, W).

  • persist_prev (bool) – if True, the selected chunk will be different from the previous time it was called. This is used so input chunk is compared to a different output chunk. Default to False.

Returns:

mask without batch dimension of shape specified either by img_size, input_mask, or class attribute img_size.

Return type:

dict