Commit 3e54e924 authored by Amir MOHAMMADI's avatar Amir MOHAMMADI

improve docs

parent a0d5cc56
Pipeline #21319 passed with stages
in 20 minutes and 32 seconds
......@@ -130,7 +130,7 @@ class SwanVideoFile(VideoBioFile, SwanFile):
Parameters
----------
padfile : :any:`SwanVideoPadFile`
padfile : :any:`SwanVideoFile`
The high-level pad file
Yields
......@@ -149,7 +149,7 @@ class SwanVideoFile(VideoBioFile, SwanFile):
Parameters
----------
padfile : :any:`SwanVideoPadFile`
padfile : :any:`SwanVideoFile`
The high-level pad file
Returns
......
......@@ -16,11 +16,12 @@ on your computer:
$ bob config set bob.db.swan.directory /path/to/swan/database
$ bob config set bob.db.swan.annotation_dir /path/to/swan/annotations
Documentation
-------------
Package Documentation
---------------------
.. toctree::
:maxdepth: 2
protocols
py_api
.. automodule:: bob.db.swan
.. automodule:: bob.db.swan.common
.. automodule:: bob.db.swan.query_bio
.. automodule:: bob.db.swan.query_pad
Dataset Protocols
=================
Biometric Recognition
---------------------
The biometric recognition protocol for both face and voice is
``grandtest0-voice-bio``. In this protocol, session 2 data is used for
enrollment and all other sessions are used for probing. The IDIAP and MPH-FRA
are used for development and NTNU and MPH-IND are used for evaluation. Voice
videos are used for both face and voice. That is why the protocol has the word
``voice`` in its name.
Vulnerability Analysis
----------------------
The vulnerability analysis protocols consist of evaluating a biometric system
twice: once using only bona-fide samples consisting of genuines and
zero-effort-impostors and using only bona-fide genuines (no
zero-effort-impostors) and presentation attacks. The protocols are called
*licit* and *spoof*, respectively. The ``grandtest1-voice-licit`` protocol is
used for both face and voice modalities. The licit protocol is very similar to
the biometric recognition protocol except that only session 1 is used for
probing. Only session 1 is used in probing because the presentation attacks are
created from session 1 and this allows a comparison with minimal differences
between bona-fide and presentation attacks.
``grandtest1-voice-spoof`` and ``grandtest1-face-spoof`` are used as spoof
protocols of voice and face modalities, respectively. All presentation attacks
are used in spoof protocols as probes.
Presentation Attack Detection
-----------------------------
PAD protocols are created according to the ``SWAN-PAD-protocols`` document.
Bona-fide session 2 data is split into 3 sets of training, development, and
evaluation. The bona-fide data from sessions 3,4,5,6 are used for evaluation as
well. PA samples are randomly split into 3 sets of training, development, and
evaluation. All the random splits are done 10 times to created 10 different
protocols. The PAD protocols contain only one type of attacks. For convenience,
PA_F and PA_V protocols are created for face and voice, respectively which
contain all the attacks.
Python API
----------
.. automodule:: bob.db.swan
.. automodule:: bob.db.swan.common
.. automodule:: bob.db.swan.query_bio
.. automodule:: bob.db.swan.query_pad
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment