optim.pocs

Contents

optim.pocs#

bartorch.optim.pocs()#

Sweep a list of projections, optim.POCS(projections)(y, x0=x0).

Unlike the solvers beside it this one has no encoding argument: what the sets are, and what the data is, belong to the projections themselves.

Parameters:

  • y (tensor) – What the iterate looks like. Without x0 the sweep starts at zero of its shape, where pocs_recon2 starts.

  • projections (sequence) – The sets to project onto, as POCSBlock takes them: a callable on a tensor, or a bartorch.priors term.

  • x0 (tensor, default=None) – Where to start instead.

  • **settings – Settings of POCS, which is maxiter (50).

Returns:

The iterate the last sweep left.

Return type:

torch.Tensor

Contents