improve solver doc, add python bindings guide

CONTRIBUTING.md

@@ -124,3 +124,17 @@ with `pre-commit install` such that they are run before each commit.
 
 None of the commit hooks will change anything in your commit, they mearly check and error if
 something is wrong.
+
+## Building the Documentation
+The [elsa documentation](https://ciip.in.tum.de/elsadocs/) is automatically built and deployed through the CI for each commit to master.
+To build it locally the following packages are required: `sphinx doxygen` which should be available in
+most major linux distributions or via pip. Additionally, the following sphinx extensions need to be installed via pip:
+`sphinx-rtd-theme m2r2 breathe`.
+Then simply build the documentation using ninja
+```
+mkdir -p build
+cd build
+cmake .. -GNinja
+ninja docs
+```
+The docs should then be available at `build/docs/sphinx`.

docs/conf.py

@@ -27,9 +27,12 @@ author = 'Tobias Lasser'
 # Add any Sphinx extension module names here, as strings. They can be
 # extensions coming with Sphinx (named 'sphinx.ext.*') or your custom
 # ones.
-extensions = [ "breathe", "m2r2", "sphinxcontrib.katex"]
+extensions = [ "breathe", "m2r2", "sphinxcontrib.katex", "sphinx.ext.autosectionlabel"]
 source_suffix = ['.rst', '.md']
 
+# Make sure the target is unique
+autosectionlabel_prefix_document = True
+
 # Add any paths that contain templates here, relative to this directory.
 templates_path = ['_templates']
 

docs/guides/python_bindings.md

+Python Bindings
+-----------------------------
+
+Elsa also comes with python bindings for almost all aspects of the framework. 
+This short guide aims to give an introduction into some simple cases and explain how one can
+easily translate C++ code into python code for faster prototyping and experimenting.
+One major benefit that comes with the python bindings is that we are able to natively
+use numpy arrays with our elsa data containers making it easy to work with other tools such as
+matplotlib.
+
+### Setup the python bindings
+In case you installed elsa via a `make install` the bindings should work out of the box.
+
+If you do not want to install elsa to your system or are developing locally make sure to
+add the path to your build folder to the `PYTHONPATH` for python to be able to find and import
+the binding code.
+
+Once everything is set up simply 
+```python
+import pyelsa as elsa
+```
+
+### 2D example
+To give a short outline into the python usage of elsa we will recreate the 2D example of the [Quickstart](Quickstart) section
+in python.
+
+```python
+import pyelsa as elsa
+import numpy as np
+import matplotlib.pyplot as plt
+
+size = np.array([128, 128])
+phantom = elsa.PhantomGenerator.createModifiedSheppLogan(size)
+volume_descriptor = phantom.getDataDescriptor()
+
+# generate circular trajectory
+num_angles = 180
+arc = 360
+
+sino_descriptor = elsa.CircleTrajectoryGenerator.createTrajectory(
+    num_angles, phantom.getDataDescriptor(), arc, size[0] * 100, size[0])
+
+# setup operator for 2d X-ray transform
+projector = elsa.SiddonsMethod(volume_descriptor, sino_descriptor)
+
+# simulate the sinogram
+sinogram = projector.apply(phantom)
+
+# setup reconstruction problem
+problem = elsa.WLSProblem(projector, sinogram)
+
+# solve the problem
+solver = elsa.CG(problem)
+n_iterations = 20
+reconstruction = solver.solve(n_iterations)
+
+# plot the reconstruction
+plt.imshow(np.array(reconstruction), '2D Reconstruction')
+plt.show()
+```
+
+As you can see the code is essentially equivalent to the C++ code shown in the Quickstart guide.
+All the top level elsa modules normally available through 
+```cpp
+#include "elsa.h"
+```
+are available under the top level `pyelsa` module. 
+All C++ functions and classes essentially have the same signatures.
+One important aspect of the python bindings is, however, that in places where the C++ code would expect
+`Eigen` vectors or matrices we can natively use `numpy` arrays as well as convert elsa `DataContainer` back to numpy
+via
+
+```python
+data_container: elsa.DataContainer = ...
+back_to_numpy = np.array(data_container)
+```
+
+This is important if we e.g. want to show a reconstruction image using matplotlib as it does not support the elsa
+data containers.
+
+### 3D reconstruction viewing
+The tight integration with numpy and matplotlib also enables us to directly implement a 3D reconstruction viewer
+within our experimentation code.
+Let's take a simple 3D phantom reconstruction example using a CUDA projector to be more performant
+
+```python
+import pyelsa as elsa
+import numpy as np
+import matplotlib.pyplot as plt
+
+size = np.array([64, 64, 64])  # 3D now
+phantom = elsa.PhantomGenerator.createModifiedSheppLogan(size)
+volume_descriptor = phantom.getDataDescriptor()
+
+# generate circular trajectory
+num_angles = 180
+arc = 360
+
+sino_descriptor = elsa.CircleTrajectoryGenerator.createTrajectory(
+    num_angles, phantom.getDataDescriptor(), arc, size[0] * 100, size[0])
+
+# setup operator for 2d X-ray transform
+projector = elsa.JosephsMethodCUDA(volume_descriptor, sino_descriptor)
+
+# simulate the sinogram
+sinogram = projector.apply(phantom)
+
+# setup reconstruction problem
+problem = elsa.WLSProblem(projector, sinogram)
+
+# solve the problem
+solver = elsa.CG(problem)
+n_iterations = 20
+reconstruction = np.array(solver.solve(n_iterations))
+```
+
+We can now implement a simple matlab viewer plugin to scroll through our 3D reconstruction using the mouse wheel as shown in
+the [matplotlib documentation](https://matplotlib.org/stable/gallery/event_handling/image_slices_viewer.html)
+```python
+class IndexTracker:
+    def __init__(self, ax, X):
+        self.ax = ax
+        ax.set_title('use scroll wheel to navigate images')
+
+        self.X = X
+        rows, cols, self.slices = X.shape
+        self.ind = self.slices // 2
+
+        self.im = ax.imshow(self.X[:, :, self.ind])
+        self.update()
+
+    def on_scroll(self, event):
+        if event.button == 'up':
+            self.ind = (self.ind + 1) % self.slices
+        else:
+            self.ind = (self.ind - 1) % self.slices
+        self.update()
+
+    def update(self):
+        self.im.set_data(self.X[:, :, self.ind])
+        self.ax.set_ylabel('slice %s' % self.ind)
+        self.im.axes.figure.canvas.draw()
+```
+we then simply set up our viewer
+```python
+
+fig, ax = plt.subplots(1, 1)
+tracker = IndexTracker(ax, reconstruction)
+fig.canvas.mpl_connect('scroll_event', tracker.on_scroll)
+plt.show()
+```
+and scroll through our 3D reconstruction.
+

docs/index.rst

@@ -12,6 +12,7 @@
 
    guides/quickstart-cxx.md
    guides/admm-cxx.md
+   guides/python_bindings.md
 
 .. toctree::
    :caption: Modules
@@ -23,7 +24,7 @@
    modules/operators
    modules/functionals
    modules/problems
-   modules/solvers
+   modules/solvers/solvers
    modules/projectors
    modules/projectors_cuda
    modules/proximity_operators

docs/modules/solvers.rst → docs/modules/solvers/api.rst

-************
-elsa solvers
-************
+.. _elsa-solvers-api:
+
+***********
+elsa solvers API
+***********
 
 .. contents:: Table of Contents
 

docs/modules/solvers/choosing_a_solver.rst

+.. _elsa-solvers-choosing-a-solver:
+
+******************
+Choosing a Solver
+******************
+
+.. contents:: Table of Contents
+
+A first example
+===============
+
+This guide assumes you have read the :ref:`Quickstart Guide <guides/quickstart-cxx:Quickstart>` or are already familiar with elsa.
+
+A typical reconstruction pipeline in elsa looks like follows.
+We would have a volume descriptor for some given 2d or 3d volume defining the
+are to be reconstructed as well as a detector descriptor that describes the trajectory
+of our measurement positions. Finally we also have a sinogram that contains the actual
+measurement data.
+
+.. code-block:: cpp
+
+  VolumeDescriptor volumeDescriptor;
+  DataContainer sinogram;
+  PlanarDetectorDescriptor sinoDescriptor;
+  SiddonsMethod projector(volumeDescriptor, sinoDescriptor);
+
+To compute a reconstruction we would construct an optimisation problem, typically
+a weighted least squares problem ``WLSProblem`` which we would then solve using an iterative
+optimisation method (e.g. Gradient Descent) to get the actual reconstructed image.
+
+.. code-block:: cpp
+
+  WLSProblem wlsProblem(projector, sinogram);
+  GradientDescent solver(problem);
+  auto reconstruction = solver.solve(20);
+
+Since there are multiple options for optimisation methods available in elsa choosing the
+right solver for the task is quite important.
+
+Solver Options
+===============
+
+The following table should be regarded as an overview over what solvers exists in elsa and what their rough properties are.
+The stated convergence speeds should be taken with a grain of salt as they are evaluated on general problems
+(see the later sections) and can differ when specific solvers are applied to more specialized problems.
+
++----------------------------------------------------------+-------------------------------------------------------+-------------+
+| Solver                                                   | Requirements on the problem                           | Convergence |
++==========================================================+=======================================================+=============+
+| Gradient Descent (GD)                                    | None                                                  | Slow        |
++----------------------------------------------------------+-------------------------------------------------------+-------------+
+| Nesterov's Fast Gradient Method (FGM)                    | None                                                  | Medium      |
++----------------------------------------------------------+-------------------------------------------------------+-------------+
+| Optimized Gradient Method (OGM)                          | None                                                  | Medium      |
++----------------------------------------------------------+-------------------------------------------------------+-------------+
+| Separable Quadratic Surrogate Ordered Subsets (SQS OS)   | None                                                  | Fast        |
++----------------------------------------------------------+-------------------------------------------------------+-------------+
+| Conjugate Gradient Descent (CG)                          | Problem must be convertible to a quadric problem      | Fast        |
++----------------------------------------------------------+-------------------------------------------------------+-------------+
+| Iterative Shrinkage-Thresholding Algorithm (ISTA)        | Requires a regularization term                        | Medium      |
++----------------------------------------------------------+-------------------------------------------------------+-------------+
+| Fast Iterative Shrinkage-Thresholding Algorithm (FISTA)  | Requires a regularization term                        | Medium      |
++----------------------------------------------------------+-------------------------------------------------------+-------------+
+| Alternating Direction Method of Multipliers (ADMM)       | <WIP>                                                 | <WIP>       |
++----------------------------------------------------------+-------------------------------------------------------+-------------+
+
+For a more detailed description on the internals of each solvers also see the :ref:`API documentation <elsa-solvers-api>`.
+To learn more about how to use ordered subsets and specifically SQS OS read the :ref:`Ordered Subsets Introduction <elsa-solvers-ordered-subsets>`.
+
+Solver Convergence
+==================
+This section will take a closer look at the convergence speed of the different solvers for a simple phantom reconstruction
+problem. The ADMM solver will be excluded here as it requires some special tweaking and cannot be compared to the other solvers
+as easily.
+
+To compare the convergence speed of the elsa solvers we set up a simple 2D 128 x 128 phantom reconstruction problem.
+We run each solver for a number of iterations and plot the Mean Square Error of the reconstruction compared to the ground truth
+over the number of iteration. The code containing these experiments can be found
+`here <https://gitlab.lrz.de/IP/elsa/-/blob/master/examples/solver_experiments.py>`_.
+The SQS solver is plotted both in its ordered subset mode and in normal mode where it is roughly equivalent to the Nesterov's Fast Gradient Method.
+
+.. image:: convergence_all_solvers.png
+  :width: 900
+  :align: center
+  :alt: SolverConvergence
+
+As we can see the main winners of this experiment are Conjugate Gradient Descent (CG) and SQS in ordered subset mode (SQS OS).
+The only other solver coming somewhat close in terms of convergence speed by number of iterations is the Optimized Gradient Method (OGM).
+Since the computational complexity of the solvers differs quite a bit we also need to have a look at how fast they converge
+when measured in terms of run time.
+
+Time complexity
+==================
+
+To measure the actual time-based convergence speed of the different solvers we again run the same experiment using
+a sample 2D 128 x 128 phantom reconstruction. We run each solver for a number of iterations up to 50 and plot the
+Mean Square Error of the reconstruction over the actual execution time of the optimization process.
+
+    Note: the irregularities stem from running the timing experiment on a normal multi-threaded desktop pc
+
+.. image:: time_all_solvers.png
+  :width: 900
+  :align: center
+  :alt: SolverTime
+
+This experiment now paints a slightly different picture of the actual convergence speeds than the previous one.
+Again the SQS OS and CG have the fastest convergence speed but the main surprise comes when looking at the FGM and OGM
+solvers which have a very high convergence speed when measured in the number of required iterations. When only measuring
+the execution time both of those solver have a slow initial convergence speed but do catch up to the other medium speed
+solvers after some time. This is due to the fact that both of these solvers have a more expensive implementation than
+other first order methods (e.g. Gradient Descent) while having better convergence properties. OGM even is proven to reach
+the theoretical maximum convergence speed for first order solvers.
+
+As a rule of thumb we can conclude that using either CG or SQS OS for general reconstruction problems in elsa will
+yield the best results. In case of more specialized problems using a solver that is specifically designed for such
+problems might be more desirable though.
+

docs/modules/solvers/ordered_subsets.rst

+.. _elsa-solvers-ordered-subsets:
+
+******************
+Ordered Subsets
+******************
+
+.. contents:: Table of Contents
+
+Introduction
+===============
+
+The elsa problem and solvers API supports solvers using ordered subsets to achieve faster initial convergence speeds.
+
+    Note: the only solver which currently implements ordered subsets is the Separable Quadratic Surrogate (SQS) method.
+
+First order solvers using ordered subset function similarly to stochastic gradient descent. Instead of operating on the
+full input data the solver computes the gradients needed for its update step only on a subset (or batch in SGD terminology) of the input data.
+This process of reducing the amount of data used in the gradient computation can greatly increase initial convergence speeds.
+One essential aspect of using OS methods for tomographic reconstruction problems is the process of sampling the initial
+input data into subsets as one requirement of OS methods is the approximate equivalence of the gradient of the full problem with the
+gradient of the ordered subsets. Since tomographic measurement data has some inherent structure stemming from the geometry of the
+detector and the detector poses during measurement the subset sampling process should yield subsets whose measurement geometry
+approximates that of the full measurement data.
+
+To use ordered subsets in elsa we would start with our normal reconstruction input of a 2D or 3D volume descriptor,
+a data container with the measurement data and a detector descriptor containing the measurement geometry and trajectory.
+
+.. code-block:: cpp
+
+    VolumeDescriptor volumeDescriptor;
+    DataContainer sinogram;
+    PlanarDetectorDescriptor sinoDescriptor;
+
+Instead of now directly instantiating a ``WLSProblem`` with the descriptors and a projector we first need to divide
+our input data into subsets. This is done by instantiating a ``SubsetSampler`` and pass it the number of subsets we want.
+
+    Note: we also need to pass the type of detector descriptor as a template parameter due to some implementation internals.
+
+.. code-block:: cpp
+
+    index_t nSubsets{4};
+    SubsetSampler<PlanarDetectorDescriptor, real_t> subsetSampler(
+        volumeDescriptor, sinoDescriptor, nSubsets);
+
+After instantiating our subset sampler we can then create a ``WLSSubsetProblem`` which we can pass to a solver which
+supports ordered subsets. The solvers will check whether they received a ``SubsetProblem`` and automatically
+run in ordered subsets mode. The API of a ``SubsetProblem`` also supports its usage as a normal ``Problem`` but will
+not yield any benefits from its subset processing.
+
+.. code-block:: cpp
+
+    WLSSubsetProblem<real_t> problem(
+        *subsetSampler.getProjector<SiddonsMethod<real_t>>(),
+        subsetSampler.getPartitionedData(sinogram),
+        subsetSampler.getSubsetProjectors<SiddonsMethod<real_t>>());
+    SQS solver{problem};
+    auto reconstruction = solver.solve(20);
+
+
+Sampling Strategies
+===================
+
+The ``SubsetSampler`` API provides several strategies to generate subsets from input data to accommodate for
+different measurement trajectories and detector geometries in order to keep the subset gradients approximately equal.
+The desired sampling strategy can simply be passed as an argument to the ``SubsetSampler`` constructor.
+
+Currently there are two strategies implemented available as enums in ``SubsetSampler::SamplingStrategy``:
+
++---------------------------+--------------------------------------------------------------------+
+| Strategy                  | Description                                                        |
++===========================+====================================================================+
+| ``ROUND_ROBIN``           | | a simple round robin based splitting (the default method), may   |
+|                           | | not work well for more complex 3D trajectories                   |
++---------------------------+--------------------------------------------------------------------+
+| ``ROTATIONAL_CLUSTERING`` | | tries to assign data points to subsets based on their rotational |
+|                           | | distance such that each subset approximates the geometry of the  |
+|                           | | full measurement data                                            |
++---------------------------+--------------------------------------------------------------------+
+
+
+Convergence
+===========
+
+A comparison between the convergence of SQS OS and most other solvers in elsa can be found in the :ref:`Solvers Overview <elsa-solvers-choosing-a-solver>`.
+
+One very tunable parameter when using ordered subset solvers is the number of subsets. Generally the more subsets used
+during the optimization the faster the initial convergence speed will be. The limit to how many subsets can and should
+be used for a given problem is given by the number of measurement data points.
+
+To get a rough understanding of how the SQS OS method performs when tweaking the number of subsets we evaluated a small
+3D 32 x 32 x 32 phantom reconstruction problem. The code for these experiments can be found
+`here <https://gitlab.lrz.de/IP/elsa/-/blob/master/examples/solver_experiments.py>`_.
+
+.. image:: convergence_sqs.png
+  :width: 900
+  :align: center
+  :alt: SQSConvergence
+
+As we can see increasing the number of subsets generally improves the convergence speed in terms of the number of iterations
+needed. The same holds true for the actual execution time needed to achieve a certain accuracy as the next plots shows.
+
+    Note: the irregularities stem from running the timing experiment on a normal multi-threaded desktop pc
+
+.. image:: time_sqs.png
+  :width: 900
+  :align: center
+  :alt: SQSTime
+
+Increasing the number of subsets used can greatly improve the actual convergence speed of an ordered subsets based solver.
+In practice some experimentation might be needed to find the maximum number of subsets usable for the given problem and
+input data set, as a too small number of data points per subset might lead to unstable convergence.
\ No newline at end of file

docs/modules/solvers/solvers.rst

+*******
+elsa solvers
+*******
+
+.. toctree::
+  :hidden:
+
+  choosing_a_solver.rst
+  ordered_subsets.rst
+  api.rst
+
+This is the documentation of elsa's solvers module. It provides implementations of different iterative solvers for
+general or more specific optimisation problems.
+
+Get started by reading a quick overview on :ref:`which solver to choose <elsa-solvers-choosing-a-solver>` or jump
+directly to the :ref:`API documentation <elsa-solvers-api>`.
+
+You can read more about the following topics:
+
+#. :ref:`Choose a solver <elsa-solvers-choosing-a-solver>` gives a basic overview over the different solvers and their strengths.
+#. :ref:`Ordered Subsets <elsa-solvers-ordered-subsets>` gives an overview over ordered subsets and supported solvers.
+#. :ref:`Solvers API <elsa-solvers-api>` provides a full documentation on the elsa solvers API.

elsa/generators/CMakeLists.txt

@@ -27,6 +27,7 @@ if(ELSA_BUILD_PYTHON_BINDINGS)
                 ${ELSA_MODULE_TARGET_NAME} bind_${ELSA_MODULE_NAME}.cpp
                 ${PROJECT_SOURCE_DIR}/tools/bindings_generation/hints/${ELSA_MODULE_NAME}_hints.cpp ${MODULE_SOURCES}
                 CircleTrajectoryGenerator.h
+                SphereTrajectoryGenerator.h
         )
 endif()
 

elsa/problems/WLSSubsetProblem.h

@@ -27,7 +27,6 @@ namespace elsa
          * @param[in] A the full system matrix of the whole WSL problem
          * @param[in] b data vector
          * @param[in] subsetAs the system matrices corresponding to each subset
-         * @param[in] subsetBs a data vector with a Block Descriptor containing the data vector vor
          * each subset
          */
         WLSSubsetProblem(const LinearOperator<data_t>& A, const DataContainer<data_t>& b,

elsa/projectors/SubsetSampler.h

@@ -34,7 +34,6 @@ namespace elsa
          *
          * @param[in] volumeDescriptor of the problem
          * @param[in] detectorDescriptor describes the geometry and trajectory of the measurements
-         * @param[in] sinogram is the actual measurement data
          * @param[in] nSubsets is number of subsets that should be generated
          * @param[in] samplingStrategy the strategy with which to sample the subsets
          */
@@ -54,7 +53,9 @@ namespace elsa
         DataContainer<data_t> getPartitionedData(const DataContainer<data_t>& sinogram);
 
         /**
-         * @brief return the full projector
+         * @brief return
+         *
+         * @tparam projector_t the type of projector to instantiate
          */
         template <typename Projector_t>
         std::unique_ptr<LinearOperator<data_t>> getProjector()
@@ -64,6 +65,8 @@ namespace elsa
 
         /**
          * @brief return a list of projectors that correspond to each subset
+         *
+         * @tparam projector_t the type of projector to instantiate
          */
         template <typename Projector_t>
         std::vector<std::unique_ptr<LinearOperator<data_t>>> getSubsetProjectors()

elsa/solvers/GradientDescent.h

@@ -36,7 +36,6 @@ namespace elsa
          * function.
          *
          * @param[in] problem the problem that is supposed to be solved
-         * @param[in] stepSize the fixed step size to be used while solving
          */
         GradientDescent(const Problem<data_t>& problem);
 

elsa/solvers/SQS.h

@@ -33,7 +33,7 @@ namespace elsa
          * in ordered subset mode.
          *
          * @param[in] problem the problem that is supposed to be solved
-         * @param[in] momentumAcceleration whether or not to enable momentum acceleration
+         * @param[in] momentumAcceleration whether to enable Nesterov's momentum acceleration
          * @param[in] epsilon affects the stopping condition
          */
         SQS(const Problem<data_t>& problem, bool momentumAcceleration = true,

examples/solver_experiments.py

+from dataclasses import dataclass, field
+from typing import List
+
+import time
+
+import numpy as np
+import matplotlib.pyplot as plt
+
+import pyelsa as elsa
+
+
+class IndexTracker:
+    def __init__(self, ax, X):
+        self.ax = ax
+        ax.set_title('use scroll wheel to navigate images')
+
+        self.X = X
+        rows, cols, self.slices = X.shape
+        self.ind = self.slices // 2
+
+        self.im = ax.imshow(self.X[:, :, self.ind])
+        self.update()
+
+    def on_scroll(self, event):
+        if event.button == 'up':
+            self.ind = (self.ind + 1) % self.slices
+        else:
+            self.ind = (self.ind - 1) % self.slices
+        self.update()
+
+    def update(self):
+        self.im.set_data(self.X[:, :, self.ind])
+        self.ax.set_ylabel('slice %s' % self.ind)
+        self.im.axes.figure.canvas.draw()
+
+
+@dataclass
+class SolverTest:
+    solver_class: elsa.Solver
+    solver_name: str
+    uses_subsets: bool = False
+    n_subsets: int = 5
+    needs_lasso: bool = False
+    sampling_strategy: str = 'ROUND_ROBIN'  # or 'ROTATIONAL_CLUSTERING'
+    extra_args: dict = field(default_factory=dict)
+
+
+def mse(optimized: np.ndarray, original: np.ndarray) -> float:
+    size = original.size
+    diff = (original - optimized) ** 2
+    return np.sum(diff) / size
+
+
+def instantiate_solvers(solvers: List[SolverTest], do_3d=False):
+    # generate the phantom
+    size = np.array([128, 128]) if not do_3d else np.array([32, 32, 32])
+    phantom = elsa.PhantomGenerator.createModifiedSheppLogan(size)
+    volume_descriptor = phantom.getDataDescriptor()
+
+    num_poses = 180
+
+    if do_3d:
+        # generate spherical trajectory
+        n_circles = 5
+        sino_descriptor = elsa.SphereTrajectoryGenerator.createTrajectory(
+            num_poses, phantom.getDataDescriptor(), n_circles, elsa.SourceToCenterOfRotation(size[0] * 100),
+            elsa.CenterOfRotationToDetector(size[0]))
+    else:
+        # generate circular trajectory
+        arc = 360
+        sino_descriptor = elsa.CircleTrajectoryGenerator.createTrajectory(
+            num_poses, phantom.getDataDescriptor(), arc, elsa.SourceToCenterOfRotation(size[0] * 100),
+            elsa.CenterOfRotationToDetector(size[0]))
+
+    # setup operator for 2d X-ray transform
+    projector = elsa.SiddonsMethod(volume_descriptor, sino_descriptor)
+