Commit 87ec296b authored by André Anjos's avatar André Anjos 💬
Browse files

Documentation overhaul

parent 8f3758ce
.. vim: set fileencoding=utf-8 :
.. Andre Anjos <>
.. Tue 15 Oct 14:59:05 2013
The C++ API of ```` allows users to leverage from automatic converters
for classes in :py:class:``. To use the C API, clients should first,
include the header file ``<>`` on their compilation units and
then, make sure to call once ``import_xbob_io_base()`` at their module
instantiation, as explained at the `Python manual
Here is a dummy C example showing how to include the header and where to call
the import function:
.. code-block:: c++
#include <>
PyMODINIT_FUNC initclient(void) {
PyObject* m Py_InitModule("client", ClientMethods);
if (!m) return;
/* imports dependencies */
if (import_xbob_blitz() < 0) {
PyErr_SetString(PyExc_ImportError, "cannot import extension");
return 0;
if (import_xbob_io_base() < 0) {
PyErr_SetString(PyExc_ImportError, "cannot import extension");
return 0;
.. note::
The include directory can be discovered using
Generic Functions
.. cpp:function:: int PyBobIo_AsTypenum(bob::core::array::ElementType et)
Converts the input Bob element type into a ``NPY_<TYPE>`` enumeration value.
Returns ``NPY_NOTYPE`` in case of problems, and sets a
.. cpp:function:: PyObject* PyBobIo_TypeInfoAsTuple (const bob::core::array::typeinfo& ti)
Converts the ``bob::core::array::typeinfo&`` object into a **new reference**
to a :py:class:`tuple` with 3 elements:
The data type as a :py:class:`numpy.dtype` object
The shape of the object, as a tuple of integers
The strides of the object, as a tuple of integers
Returns ``0`` in case of failure, or a **new reference** to the tuple
described above in case of success.
.. cpp:function:: int PyBobIo_FilenameConverter (PyObject* o, PyObject** b)
This function is meant to be used with :c:func:`PyArg_ParseTupleAndKeywords`
family of functions in the Python C-API. It converts an arbitrary input
object into a ``PyStringObject`` (in Python2.x) and into a ``PyBytesObject``
(in Python3.x). If the input object is of type ``PyUnicodeObject``, which is
the default in Python3.x, the unicode code is properly decoded using
:c:func:`PyUnicode_AsEncodedString` with ``encoding`` set to
``Py_FileSystemDefaultEncoding`` and ``errors`` set to ``"strict"``. On
versions of Python >= 3.2, this is just an alias for
:c:func:`PyUnicode_FSConverter`, which does a similar job.
Objects which are not ``PyUnicodeObject`` are coerced into a bytes/string
object using :c:func:`PyObject_Bytes` (on Python3.x) and
:c:func:`PyObject_Str` (on Python 2.x).
Returns 0 if an error is detected, 1 on success.
Bob File Support
.. cpp:type:: PyBobIoFileObject
The pythonic object representation for a ``bob::io::File`` object.
.. code-block:: cpp
typedef struct {
boost::shared_ptr<bob::io::File> f;
} PyBobIoFileObject;
.. cpp:member:: boost::shared_ptr<bob::io::File> f
A pointer to a file being read or written.
.. cpp:type:: PyBobIoFileIteratorObject
The pythonic object representation for an iterator over a ``bob::io::File``
.. code-block:: cpp
typedef struct {
PyBobIoFileObject* pyfile;
Py_ssize_t curpos;
} PyBobIoFileIteratorObject;
.. cpp:member:: PyBobIoFileObject* pyfile
A pointer to the pythonic representation of a file.
.. cpp:member:: Py_ssize_t curpos
The current position at the file being pointed to.
Bob HDF5 Support
.. cpp:type:: PyBobIoHDF5FileObject
The pythonic object representation for a ``bob::io::HDF5File`` object.
.. code-block:: cpp
typedef struct {
boost::shared_ptr<bob::io::HDF5File> f;
} PyBobIoHDF5FileObject;
.. cpp:member:: boost::shared_ptr<bob::io::HDF5File> f
A pointer to a Bob object being used to read/write data into an HDF5
.. cpp:function:: int PyBobIoHDF5File_Check(PyObject* o)
Checks if the input object ``o`` is a ``PyBobIoHDF5FileObject``. Returns
``1`` if it is, and ``0`` otherwise.
.. cpp:function:: int PyBobIoHDF5File_Converter(PyObject* o, PyBobIoHDF5FileObject** a)
This function is meant to be used with :c:func:`PyArg_ParseTupleAndKeywords`
family of functions in the Python C-API. It checks the input object to be of
type ``PyBobIoHDF5FileObject`` and sets a **new reference** to it (in
``*a``) if it is the case. Returns ``0`` in case of failure, ``1`` in case
of success.
.. include:: links.rst
......@@ -58,12 +58,12 @@ source_suffix = '.rst'
master_doc = 'index'
# General information about the project.
project = u''
project = u''
import time
copyright = u'%s, Idiap Research Institute' % time.strftime('%Y')
# Grab the setup entry
distribution = pkg_resources.require('')[0]
distribution = pkg_resources.require('')[0]
# The version info for the project you're documenting, acts as replacement for
# |version| and |release|, also used in various other places throughout the
......@@ -129,7 +129,7 @@ if sphinx.__version__ >= "1.0":
#html_title = None
# A shorter title for the navigation bar. Default is the same as html_title.
#html_short_title = 'xbob_io_base'
#html_short_title = 'xbob_io_image'
# The name of an image file (relative to this directory) to place at the top
# of the sidebar.
......@@ -180,14 +180,14 @@ html_favicon = 'img/favicon.ico'
# If true, an OpenSearch description file will be output, and all pages will
# contain a <link> tag referring to it. The value of this option must be the
# base URL from which the finished HTML is served.
# image URL from which the finished HTML is served.
#html_use_opensearch = ''
# This is the file name suffix for HTML files (e.g. ".xhtml").
#html_file_suffix = None
# Output file base name for HTML help builder.
htmlhelp_basename = 'xbob_io_base_doc'
htmlhelp_basename = 'xbob_io_image_doc'
# -- Options for LaTeX output --------------------------------------------------
......@@ -201,7 +201,7 @@ latex_font_size = '10pt'
# Grouping the document tree into LaTeX files. List of tuples
# (source start file, target name, title, author, documentclass [howto/manual]).
latex_documents = [
('index', 'xbob_io_base.tex', u'Core Bob I/O Routines',
('index', 'xbob_io_image.tex', u'Image I/O Support for Bob',
u'Biometrics Group, Idiap Research Institute', 'manual'),
......@@ -241,7 +241,7 @@ rst_epilog = """
# One entry per manual page. List of tuples
# (source start file, name, description, authors, manual section).
man_pages = [
('index', 'xbob_io_base', u'Core Bob I/O Routines Documentation', [u'Idiap Research Institute'], 1)
('index', 'xbob_io_image', u'Image I/O Support for Bob', [u'Idiap Research Institute'], 1)
# Default processing flags for sphinx
......@@ -2,25 +2,11 @@
.. Andre Anjos <>
.. Fri 16 May 11:48:13 2014 CEST
User Guide
This section gives an overview of the operations for storing and retrieving the
basic data structures in |project|, such as `NumPy`_ arrays. |project| uses
`HDF5`_ format for storing binary coded data. Using the |project| support for
`HDF5`_, it is very simple to import and export data.
`HDF5`_ uses a neat descriptive language for representing the data in the HDF5
files, called Data Description Language (`DDL`_).
To perform the functionalities given in this section, you should have `NumPy`_
and |project| loaded into the `Python`_ environment.
.. testsetup:: *
import numpy
import tempfile
import os
......@@ -28,324 +14,22 @@ and |project| loaded into the `Python`_ environment.
temp_dir = tempfile.mkdtemp(prefix='bob_doctest_')
HDF5 standard utilities
Before explaining the basics of reading and writing to `HDF5`_ files, it is
important to list some `HDF5`_ standard utilities for checking the content of
an `HDF5`_ file. These are supplied by the `HDF5`_ project.
Dumps the content of the file using the DDL.
Lists the content of the file using DDL, but does not show the data.
Finds the differences between HDF5 files.
I/O operations using the class ``
Writing operations
Let's take a look at how to write simple scalar data such as integers or
.. doctest::
>>> an_integer = 5
>>> a_float = 3.1416
>>> f ='testfile1.hdf5', 'w')
>>> f.set('my_integer', an_integer)
>>> f.set('my_float', a_float)
>>> del f
If after this you use the **h5dump** utility on the file ``testfile1.hdf5``,
you will verify that the file now contains:
.. code-block:: none
HDF5 "testfile1.hdf5" {
GROUP "/" {
DATASET "my_float" {
DATASPACE SIMPLE { ( 1 ) / ( 1 ) }
(0): 3.1416
DATASET "my_integer" {
DATASPACE SIMPLE { ( 1 ) / ( 1 ) }
(0): 5
.. note::
In |project|, when you open a HDF5 file, you can choose one of the following
**'r'** Open the file in reading mode; writing operations will fail (this is the default).
**'a'** Open the file in reading and writing mode with appending.
**'w'** Open the file in reading and writing mode, but truncate it.
**'x'** Read/write/append with exclusive access.
The dump shows that there are two datasets inside a group named ``/`` in the
file. HDF5 groups are like file system directories. They create namespaces for
the data. In the root group (or directory), you will find the two variables,
named as you set them to be. The variable names are the complete path to the
location where they live. You could write a new variable in the same file but
in a different directory like this:
.. doctest::
>>> f ='testfile1.hdf5', 'a')
>>> f.create_group('/test')
>>> f.set('/test/my_float', numpy.float32(6.28))
>>> del f
Line 1 opens the file for reading and writing, but without truncating it. This
will allow you to access the file contents. Next, the directory ``/test`` is
created and a new variable is written inside the subdirectory. As you can
verify, **for simple scalars**, you can also force the storage type. Where
normally one would have a 64-bit real value, you can impose that this variable
is saved as a 32-bit real value. You can verify the dump correctness with
.. code-block:: none
GROUP "/" {
GROUP "test" {
DATASET "my_float" {
DATASPACE SIMPLE { ( 1 ) / ( 1 ) }
(0): 6.28
Notice the subdirectory ``test`` has been created and inside it a floating
point number has been stored. Such a float point number has a 32-bit precision
as it was defined.
.. note::
If you need to place lots of variables in a subfolder, it may be better to
setup the prefix folder before starting the writing operations on the
:py:class:`` object. You can do this using the method
:py:meth:``. Look up its help for more information and usage
Writing arrays is a little simpler as the :py:class:`numpy.ndarray` objects
encode all the type information we need to write and read them correctly. Here
is an example:
.. doctest::
>>> A = numpy.array(range(4), 'int8').reshape(2,2)
>>> f ='testfile1.hdf5', 'a')
>>> f.set('my_array', A)
>>> del f
The result of running ``h5dump`` on the file ``testfile3.hdf5`` should be:
.. code-block:: none
DATASET "my_array" {
DATASPACE SIMPLE { ( 2, 2 ) / ( 2, 2 ) }
(0,0): 0, 1,
(1,0): 2, 3
You don't need to limit yourself to single variables, you can also save lists
of scalars and arrays using the function :py:meth:``
instead of :py:meth:``.
Reading operations
Reading data from a file that you just wrote to is just as easy. For this task
you should use :py:meth:``. The read method will read
all the contents of the variable pointed to by the given path. This is the
normal way to read a variable you have written with
:py:meth:``. If you decided to create a list of scalar
or arrays, the way to read that up would be using
:py:meth:`` instead. Here is an example:
.. doctest::
>>> f ='testfile1.hdf5') #read only
>>>'my_integer') #reads integer
>>> print('my_array')) # reads the array
[[0 1]
[2 3]]
>>> del f
Now let's look at an example where we have used
:py:meth:`` instead of
:py:meth:`` to write data to a file. That is normally
the case when you write lists of variables to a dataset.
.. doctest::
>>> f ='testfile2.hdf5', 'w')
>>> f.append('arrayset', numpy.array(range(10), 'float64'))
>>> f.append('arrayset', 2*numpy.array(range(10), 'float64'))
>>> f.append('arrayset', 3*numpy.array(range(10), 'float64'))
>>> print(f.lread('arrayset', 0))
[ 0. 1. 2. 3. 4. 5. 6. 7. 8. 9.]
>>> print(f.lread('arrayset', 2))
[ 0. 3. 6. 9. 12. 15. 18. 21. 24. 27.]
>>> del f
This is what the ``h5dump`` of the file would look like:
.. code-block:: none
HDF5 "testfile4.hdf5" {
GROUP "/" {
DATASET "arrayset" {
DATASPACE SIMPLE { ( 3, 10 ) / ( H5S_UNLIMITED, 10 ) }
(0,0): 0, 1, 2, 3, 4, 5, 6, 7, 8, 9,
(1,0): 0, 2, 4, 6, 8, 10, 12, 14, 16, 18,
(2,0): 0, 3, 6, 9, 12, 15, 18, 21, 24, 27
Notice that the expansion limits for the first dimension have been correctly
set by |project| so you can insert an *unlimited* number of 1D float vectors.
Of course, you can also read the whole contents of the arrayset in a single
.. doctest::
>>> f ='testfile2.hdf5')
>>> print('arrayset'))
[[ 0. 1. 2. 3. 4. 5. 6. 7. 8. 9.]
[ 0. 2. 4. 6. 8. 10. 12. 14. 16. 18.]
[ 0. 3. 6. 9. 12. 15. 18. 21. 24. 27.]]
As you can see, the only difference between
:py:meth:`` and
:py:meth:`` is on how |project| considers the
available data (as a single array with N dimensions or as a set of arrays with
N-1 dimensions). In the first example, you would have also been able to read
the variable `my_array` as an arrayset using
:py:meth:`` instead of
:py:meth:``. In this case, each position readout
would return a 1D uint8 array instead of a 2D array.
Array interfaces
What we have shown so far is the generic API to read and write data using HDF5.
You will use it when you want to import or export data from |project| into
other software frameworks, debug your data or just implement your own classes
that can serialize and de-serialize from HDF5 file containers. In |project|,
most of the time you will be working with :py:class:`numpy.ndarrays`\s. In
special situations though, you may be asked to handle
:py:class:``\s. :py:class:`` objects create a
transparent connection between C++ (`Blitz++`_) / Python (`NumPy`_) arrays and
file access. You specify the filename from which you want to input data and
the :py:class:`` object decides what is the best codec to be
used (from the extension) and how to read the data back into your array.
To create an :py:class:`` from a file path, just do the
.. doctest::
>>> a ='testfile2.hdf5', 'r')
>>> a.filename
:py:class:``\s simulate containers for
:py:class:`numpy.ndarray`\s, transparently accessing the file data when
requested. Note, however, that when you instantiate an
:py:class:`` it does **not** load the file contents into
memory. It waits until you emit another explicit instruction to do so. We do
this with the :py:meth:`` method:
.. doctest::
>>> array =
>>> array
array([[ 0., 1., 2., 3., 4., 5., 6., 7., 8., 9.],
[ 0., 2., 4., 6., 8., 10., 12., 14., 16., 18.],
[ 0., 3., 6., 9., 12., 15., 18., 21., 24., 27.]])
Every time you say :py:meth:``, the file contents will be
read from the file and into a new array.
Saving arrays to the :py:class:`` is as easy, just call the
:py:meth:`` method:
.. doctest::
>>> f ='copy1.hdf5', 'w')
>>> f.write(array)
Numpy ndarray shortcuts
To just load an :py:class:`numpy.ndarray` in memory, you can use a short cut
that lives at :py:func:``. With it, you don't have to go
through the :py:class:`` container:
.. doctest::
>>> t ='testfile2.hdf5')
>>> t
array([[ 0., 1., 2., 3., 4., 5., 6., 7., 8., 9.],
[ 0., 2., 4., 6., 8., 10., 12., 14., 16., 18.],
[ 0., 3., 6., 9., 12., 15., 18., 21., 24., 27.]])
You can also directly save :py:class:`numpy.ndarray`\s without going through
the :py:class:`` container:
.. doctest::
>>>, 'copy2.hdf5')
User Guide
.. note::
By importing this package, you can use |project| native array reading and
writing routines to load and save files using various image formats.
Under the hood, we still use the :py:class:`` API to execute
the read and write operations. Have a look at the manual section for
:py:mod:`` for more details and other shortcuts available.
.. code-block:: python
Reading and writing images
>> import
>> import #under the hood: loads Bob plugin for image files
>> x ='myfile.jpg')
|project| provides support to load and save data from many different file types
including Matlab ``.mat`` files, various image file types and video data. File
types and specific serialization and de-serialization is switched automatically
using filename extensions. Knowing this, saving an array in a different format
is just a matter of choosing the right extension. This is illustrated in the
following example, where an image generated randomly using the method `NumPy`
In the following example, an image generated randomly using the method `NumPy`
:py:meth:`numpy.random.random_integers`, is saved in JPEG format. The image
must be of type uint8 or uint16.
must be of type ``uint8`` or ``uint16``:
.. doctest::
......@@ -353,129 +37,11 @@ must be of type uint8 or uint16.
>>>'uint8'), 'testimage.jpg') # saving the image in jpeg format
>>> my_image_copy ='testimage.jpg')
.. tip::
To find out about which formats and extensions are supported in a given
installation of |project|, just call ```` on your prompt. It
will print a list of compiled-in software and supported extensions.
The loaded image files can be 3D arrays (for RGB format) or 2D arrays (for
greyscale) of type ``uint8`` or ``uint16``.
Dealing with videos
|project| has support for dealing with videos in an equivalent way to dealing
with images:
.. doctest::
>>> my_video = numpy.random.random_integers(0,255,(30,3,256,256))
>>>'uint8'), 'testvideo.avi') # saving the video avi format with a default codec
>>> my_video_copy ='testvideo.avi')
Video reading and writing is performed using an `FFmpeg`_ (or `libav`_ if
`FFmpeg`_ is not available) bridge. |project|'s :py:meth:``
method will allow you to choose the output format with the same extension
mechanism as mentioned earlier. `FFmpeg`_ will then choose a default codec for
the format and perform encoding. The output file can be as easily loaded using
For finer control over the loading, saving, format and codecs used for a
specific encoding or decoding operation, you must directly use either
:py:class:`` or :py:class:``
classes. For example, it is possible to use
:py:class:`` to read videos frame by frame and avoid
overloading your machine's memory. In the following example you can see how to
create a video, save it using the class :py:class:``
and load it again using the class :py:class:``. The
created video will have 30 frames generated randomly.
.. note::
Due to `FFmpeg`_ constrains, the width and height of the video need to be
multiples of two.
.. doctest::
>>> width = 50; height = 50;
>>> framerate = 24
>>> outv ='testvideo.avi', height, width, framerate, codec='mpeg1video') # output video
>>> for i in range(0, 30):
... newframe = (numpy.random.random_integers(0,255,(3,height,width)))
... outv.append(newframe.astype('uint8'))
>>> outv.close()
>>> input ='testvideo.avi')
>>> input.number_of_frames
>>> inv = input.load()
>>> inv.shape
(30, 3, 50, 50)
>>> type(inv)
<... 'numpy.ndarray'>
Videos in |project| are represented as sequences of colored images, i.e. 4D
arrays of type ``uint8``. All the extensions and formats for videos supported
in version of |project| installed on your machine can be listed using the
|project|'s utility ````.
.. testcleanup:: *
import shutil