LION.operators.Operator

class LION.operators.Operator(device=None)

Bases: Module, ABC

The base class of LION’s operators: a linear map, as a torch module, with its adjoint and the shapes it maps between.

LION uses the general words, as an operator can be anything:

• domain: what the operator is applied to, usually the image x;

• range: what it gives, the measurements y.

In CT, the operator is the projection: its domain is the image (the geometry’s image_shape) and its range the sinogram. Some operators’ domains are not images, e.g. a wavelet synthesis’s is its coefficients.

A subclass defines forward, __call__ (return self.forward(x)), adjoint, domain_shape and range_shape; autograd() if it can be differentiated through; inverse if it has one; default_reconstruction_fn if it has a usual reconstruction. device: where it computes (None: unset; to_device moves it).

Initialize internal Module state, shared by both nn.Module and ScriptModule.

__init__(device=None)

Initialize internal Module state, shared by both nn.Module and ScriptModule.

Methods

__init__([device])	Initialize internal Module state, shared by both nn.Module and ScriptModule.
add_module(name, module)	Add a child module to the current module.
adjoint(y)	The adjoint applied to y (Aᵀ y): from the range (the measurements) to the domain (the images).
apply(fn)	Apply fn recursively to every submodule (as returned by .children()) as well as self.
autograd()	The operator and its adjoint as torch autograd functions of batches, (forward, adjoint): functions whose gradients torch knows (the gradient of A is A^T, and of A^T is A), for models and solvers whose gradients go through the operator (e.g. LPD, Sparse2Inverse).
bfloat16()	Casts all floating point parameters and buffers to bfloat16 datatype.
buffers([recurse])	Return an iterator over module buffers.
children()	Return an iterator over immediate children modules.
compile(*args, **kwargs)	Compile this Module's forward using torch.compile().
cpu()	Move all model parameters and buffers to the CPU.
cuda([device])	Move all model parameters and buffers to the GPU.
default_reconstruction(y)	The operator's usual reconstruction of measurements y (default_reconstruction_fn, e.g. FDK for CT).
default_reconstruction_fn()	The operator's usual reconstruction, as a function f(y, op) of measurements and operator that a record can name (e.g. fdk for CT).
double()	Casts all floating point parameters and buffers to double datatype.
eval()	Set the module in evaluation mode.
extra_repr()	Return the extra representation of the module.
float()	Casts all floating point parameters and buffers to float datatype.
forward(x)	The operator applied to x (A x), defined by a subclass.
get_buffer(target)	Return the buffer given by target if it exists, otherwise throw an error.
get_extra_state()	Return any extra state to include in the module's state_dict.
get_parameter(target)	Return the parameter given by target if it exists, otherwise throw an error.
get_submodule(target)	Return the submodule given by target if it exists, otherwise throw an error.
gram(x)	The Gram operator, AᵀA, applied to x.
half()	Casts all floating point parameters and buffers to half datatype.
inverse(y)	The inverse applied to y, for operators that have one (e.g. an orthogonal transform): a subclass defines it only then, and its default reconstruction is then the inverse.
ipu([device])	Move all model parameters and buffers to the IPU.
load_state_dict(state_dict[, strict, assign])	Copy parameters and buffers from state_dict into this module and its descendants.
modules([remove_duplicate])	Return an iterator over all modules in the network.
mtia([device])	Move all model parameters and buffers to the MTIA.
named_buffers([prefix, recurse, ...])	Return an iterator over module buffers, yielding both the name of the buffer as well as the buffer itself.
named_children()	Return an iterator over immediate children modules, yielding both the name of the module as well as the module itself.
named_modules([memo, prefix, remove_duplicate])	Return an iterator over all modules in the network, yielding both the name of the module as well as the module itself.
named_parameters([prefix, recurse, ...])	Return an iterator over module parameters, yielding both the name of the parameter as well as the parameter itself.
parameters([recurse])	Return an iterator over module parameters.
register_backward_hook(hook)	Register a backward hook on the module.
register_buffer(name, tensor[, persistent])	Add a buffer to the module.
register_forward_hook(hook, *[, prepend, ...])	Register a forward hook on the module.
register_forward_pre_hook(hook, *[, ...])	Register a forward pre-hook on the module.
register_full_backward_hook(hook[, prepend])	Register a backward hook on the module.
register_full_backward_pre_hook(hook[, prepend])	Register a backward pre-hook on the module.
register_load_state_dict_post_hook(hook)	Register a post-hook to be run after module's load_state_dict() is called.
register_load_state_dict_pre_hook(hook)	Register a pre-hook to be run before module's load_state_dict() is called.
register_module(name, module)	Alias for add_module().
register_parameter(name, param)	Add a parameter to the module.
register_state_dict_post_hook(hook)	Register a post-hook for the state_dict() method.
register_state_dict_pre_hook(hook)	Register a pre-hook for the state_dict() method.
requires_grad_([requires_grad])	Change if autograd should record operations on parameters in this module.
set_extra_state(state)	Set extra state contained in the loaded state_dict.
set_submodule(target, module[, strict])	Set the submodule given by target if it exists, otherwise throw an error.
share_memory()	See torch.Tensor.share_memory_().
state_dict(*args[, destination, prefix, ...])	Return a dictionary containing references to the whole state of the module.
to(*args, **kwargs)	Move and/or cast the parameters and buffers.
to_device(device)	Moves the operator and its sub-operators to device, updating their device attributes (which nn.Module.to doesn't).
to_empty(*, device[, recurse])	Move the parameters and buffers to the specified device without copying storage.
train([mode])	Set the module in training mode.
type(dst_type)	Casts all parameters and buffers to dst_type.
xpu([device])	Move all model parameters and buffers to the XPU.
zero_grad([set_to_none])	Reset gradients of all model parameters.

Attributes

T_destination
call_super_init
domain_shape	The shape of the operator's domain: what it is applied to, usually the image (in CT, the geometry's image_shape).
dump_patches
range_shape	The shape of the operator's range: the measurements it gives (in CT, the sinogram).
training

forward(x)

The operator applied to x (A x), defined by a subclass. Call the operator, op(x), rather than this method (as for torch modules).

adjoint(y)

The adjoint applied to y (Aᵀ y): from the range (the measurements) to the domain (the images).

autograd()

The operator and its adjoint as torch autograd functions of batches, (forward, adjoint): functions whose gradients torch knows (the gradient of A is A^T, and of A^T is A), for models and solvers whose gradients go through the operator (e.g. LPD, Sparse2Inverse).

An operator that can be differentiated defines it; the base class raises, so one that can’t isn’t used there by mistake.

• An operator torch can’t differentiate (e.g. CT’s, a tomosipo/astra call) wraps itself, as CTProjectionOp does with tomosipo’s to_autograd.

• An operator made only of torch operations (no numpy, no external library) is differentiated by torch as it is. If you know yours is, say so:

  def autograd(self):
      return self, self.adjoint
  

default_reconstruction_fn()

The operator’s usual reconstruction, as a function f(y, op) of measurements and operator that a record can name (e.g. fdk for CT). It is what "default" stands for in a model’s or reconstructor’s initial_reconstruction and a solver’s recon_fn, and they record it in its place; it is also the baseline methods are compared with (LION.evaluation.baseline). An operator with an inverse (inverse()) has that, inverse_reconstruction; others that have one define it.

default_reconstruction(y)

The operator’s usual reconstruction of measurements y (default_reconstruction_fn, e.g. FDK for CT).

gram(x)

The Gram operator, AᵀA, applied to x.

inverse(y)

The inverse applied to y, for operators that have one (e.g. an orthogonal transform): a subclass defines it only then, and its default reconstruction is then the inverse.

property domain_shape

The shape of the operator’s domain: what it is applied to, usually the image (in CT, the geometry’s image_shape).

property range_shape

The shape of the operator’s range: the measurements it gives (in CT, the sinogram).

to_device(device)

Moves the operator and its sub-operators to device, updating their device attributes (which nn.Module.to doesn’t). Returns the operator.