Merge pull request #4 from Gradiant/development

Development

bob/paper/icb2019/gradgpad/scripts/generate_grad_gpad_protocols_text_lists.py

@@ -0,0 +1,57 @@
+#!/usr/bin/env python
+# Gradiant's Biometrics Team <[email protected]>
+# Copyright (C) 2017 Gradiant, Vigo, Spain
+
+import argparse
+import os
+
+from bob.gradiant.face.databases import AggregateDatabase, get_database_from_key, export_database_paths_from_file
+
+
+def has_args(args):
+    is_active = False
+    for arg in vars(args):
+        is_active = is_active or getattr(args, arg)
+    return is_active
+
+
+def write_dict_to_file(filename, d):
+    with open(filename, 'w') as f:
+        for k in d.keys():
+            f.write('{}\n'.format(k))
+
+
+def main():
+    parser = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
+    parser.add_argument('--json-file',
+                        type=str,
+                        dest='json_file',
+                        help='json file where the ROOT_PATH of each dataset is defined', required=True)
+
+    args = parser.parse_args()
+
+    if not has_args(args):
+        parser.print_help()
+    else:
+
+        base_output_path = 'gradgpad_protocols'
+        os.makedirs(base_output_path, exist_ok=True)
+
+        export_database_paths_from_file(args.json_file)  # temporary
+        aggregate_database = get_database_from_key('aggregate-database')  # returns an object of AggregateDatabase class
+
+        available_protocols = list(AggregateDatabase.get_available_protocols().keys())
+
+        for protocol in available_protocols:
+
+            protocol_output_path = '{}/{}'.format(base_output_path, protocol)
+            os.makedirs(protocol_output_path, exist_ok=True)
+
+            content = aggregate_database.get_ground_truth_list(protocol)
+            for subset, accesses in content.items():
+                filename = '{}/protocol_{}_{}.txt'.format(protocol_output_path, protocol, subset.lower())
+                write_dict_to_file(filename, accesses)
+
+
+if __name__ == '__main__':
+    main()

buildout.cfg

@@ -23,7 +23,7 @@ verbose = true
 [sources]
 bob.gradiant.pipelines = git https://github.com/Gradiant/bob.gradiant.pipelines.git rev=v1.1.0
 bob.gradiant.core = git https://github.com/Gradiant/bob.gradiant.core.git rev=v1.2.0
-bob.gradiant.face.databases = git https://github.com/Gradiant/bob.gradiant.face.databases.git rev=v2.0.0
+bob.gradiant.face.databases = git https://github.com/Gradiant/bob.gradiant.face.databases.git rev=v2.0.1
 bob.gradiant.pad.evaluator = git https://github.com/Gradiant/bob.gradiant.pad.evaluator.git rev=v1.2.0
 
 

doc/conf.py

@@ -35,7 +35,7 @@
 else:
     extensions.append('sphinx.ext.pngmath')
 
-os.makedirs('examples')
+os.makedirs('examples', exist_ok=True)
 shutil.copy('../rr.sh', 'examples/rr.sh')
 shutil.copy('../ci.sh', 'examples/ci.sh')
 shutil.copy('../experiments/helpers/database_paths_vis1.json', 'examples/database_paths.json')

doc/getting_started.rst

@@ -53,7 +53,8 @@ Doc
 
     bin/sphinx-build -b html doc/ doc/html/
 
-
+Continuous Integration
+----------------------
 
 In the following script (:code:`ci.sh`) you can check the different stages that are evaluated over the continuous integration pipeline:
 

doc/grad_gpad.rst

@@ -28,8 +28,29 @@ GRAD-GPAD presents two main stages:
     ii) evaluation, where a filtering step is applied, using the already extracted features and the common categorization of the datasets, to train and test over the selected features.
 
 
-How to run my own experiment with our provided framework?
----------------------------------------------------------
+In the time of the publication, the UVAD dataset haven't been parsed yet. Fortunately, now is available in the GRAD-GPAD framework.
+From the following figures we may observe how is distributed the proposed Aggregate Dataset for the Grandtest protocol.
+
+.. image:: img/grad-gpad-grandtest.png
+   :scale: 40 %
+   :alt: proposed_evaluation_framework
+   :align: center
+
+
+.. image:: img/grad-gpad-grandtest-specific.png
+   :scale: 40 %
+   :alt: proposed_evaluation_framework
+   :align: center
+
+
+Note:
+
+    The baseline for evaluation in the GRAD-GPAD without UVAD are the results reported in this paper,
+    whereas the baseline to evaluate with the new version of the GRAD-GPAD (with UVAD) are the results presented in the paper `Deep Anomaly Detection for Generalized Face Anti-Spoofing <https://arxiv.org/pdf/1904.08241.pdf>`_.
+
+
+How to run my own experiment with GRAD-GPAD python framework?
+-------------------------------------------------------------
 
 Once we have available the publicly datasets, we have to create a json with the :code:`ROOT_PATH` of each dataset. Take the following file as an example:
 
@@ -174,3 +195,14 @@ Finally, to run the experiment, just type:
 .. code-block:: sh
 
     bin/algorithmic_constrained_evaluation.py -r experiment/mypad/configuration_mypad.py
+
+
+Can the GRAD-GPAD framework be used in another platform/language/environment?
+-----------------------------------------------------------------------------
+
+The answer is yes, you can. We also provide several text list that define every of the protocols used.
+This list are generated automatically from the framework. If you find any problem, please open an issue to try to solve it.
+
+Find the protocols lists `here <https://github.com/Gradiant/bob.paper.icb2019.gradgpad/tree/master/gradgpad_protocols>`_.
+
+Note that you should modify the :code:`DATABASE-NAME-PLACEHOLDER` in these files for each dataset.