From 91724aaad37dbc27b44348e6bf2300533a0f0429 Mon Sep 17 00:00:00 2001
From: Olegs NIKISINS <onikisins@italix03.idiap.ch>
Date: Tue, 18 Jul 2017 14:38:48 +0200
Subject: [PATCH] Added reduced training data option to VideoSvmPadAlgorithm +
created baselines configs for Aggregated Db
---
.../face/algorithm/VideoSvmPadAlgorithm.py | 86 +++++++++++++-
bob/pad/face/config/database/aggregated_db.py | 2 -
.../config/frame_diff_svm_aggregated_db.py | 106 +++++++++++++++++
bob/pad/face/config/qm_svm_aggregated_db.py | 112 ++++++++++++++++++
doc/baselines.rst | 15 +++
doc/resources.rst | 16 +++
setup.py | 2 +
7 files changed, 334 insertions(+), 5 deletions(-)
create mode 100644 bob/pad/face/config/frame_diff_svm_aggregated_db.py
create mode 100644 bob/pad/face/config/qm_svm_aggregated_db.py
diff --git a/bob/pad/face/algorithm/VideoSvmPadAlgorithm.py b/bob/pad/face/algorithm/VideoSvmPadAlgorithm.py
index 9cbaeb7d..d84b0752 100644
--- a/bob/pad/face/algorithm/VideoSvmPadAlgorithm.py
+++ b/bob/pad/face/algorithm/VideoSvmPadAlgorithm.py
@@ -60,6 +60,19 @@ class VideoSvmPadAlgorithm(Algorithm):
``frame_level_scores_flag`` : :py:class:`bool`
Return scores for each frame individually if True. Otherwise, return a
single score per video. Default: False.
+
+ ``save_debug_data_flag`` : :py:class:`bool`
+ Save the data, which might be usefull for debugging if ``True``.
+ Default: ``True``.
+
+ ``reduced_train_data_flag`` : :py:class:`bool`
+ Reduce the amount of final training samples if set to ``True``.
+ Default: ``False``.
+
+ ``n_train_samples`` : :py:class:`int`
+ Number of uniformly selected feature vectors per class defining the
+ sizes of sub-sets used in the final traing of the SVM.
+ Default: 50000.
"""
def __init__(self,
@@ -68,7 +81,10 @@ class VideoSvmPadAlgorithm(Algorithm):
n_samples = 10000,
trainer_grid_search_params = { 'cost': [2**p for p in range(-5, 16, 2)], 'gamma': [2**p for p in range(-15, 4, 2)]},
mean_std_norm_flag = False,
- frame_level_scores_flag = False):
+ frame_level_scores_flag = False,
+ save_debug_data_flag = True,
+ reduced_train_data_flag = False,
+ n_train_samples = 50000):
Algorithm.__init__(self,
@@ -78,6 +94,9 @@ class VideoSvmPadAlgorithm(Algorithm):
trainer_grid_search_params = trainer_grid_search_params,
mean_std_norm_flag = mean_std_norm_flag,
frame_level_scores_flag = frame_level_scores_flag,
+ save_debug_data_flag = save_debug_data_flag,
+ reduced_train_data_flag = reduced_train_data_flag,
+ n_train_samples = n_train_samples,
performs_projection=True,
requires_projector_training=True)
@@ -87,6 +106,9 @@ class VideoSvmPadAlgorithm(Algorithm):
self.trainer_grid_search_params = trainer_grid_search_params
self.mean_std_norm_flag = mean_std_norm_flag
self.frame_level_scores_flag = frame_level_scores_flag
+ self.save_debug_data_flag = save_debug_data_flag
+ self.reduced_train_data_flag = reduced_train_data_flag
+ self.n_train_samples = n_train_samples
self.machine = None
@@ -217,6 +239,44 @@ class VideoSvmPadAlgorithm(Algorithm):
return features_subset
+ #==========================================================================
+ def select_quasi_uniform_data_subset(self, features, n_samples):
+ """
+ Select quasi uniformly N samples/feature vectors from the input array of samples.
+ The rows in the input array are samples. The columns are features.
+ Use this function if n_samples is close to the number of samples.
+
+ **Parameters:**
+
+ ``features`` : 2D :py:class:`numpy.ndarray`
+ Input array with feature vectors. The rows are samples, columns are features.
+
+ ``n_samples`` : :py:class:`int`
+ The number of samples to be selected uniformly from the input array of features.
+
+ **Returns:**
+
+ ``features_subset`` : 2D :py:class:`numpy.ndarray`
+ Selected subset of features.
+ """
+
+ if features.shape[0] <= n_samples:
+
+ features_subset = features
+
+ else:
+
+ uniform_step = (1.0 * features.shape[0]) / n_samples
+
+ element_num_list = range(0,n_samples)
+
+ idx = [np.int(uniform_step*item) for item in element_num_list]
+
+ features_subset = features[idx, :]
+
+ return features_subset
+
+
#==========================================================================
def split_data_to_train_cv(self, features):
"""
@@ -447,7 +507,9 @@ class VideoSvmPadAlgorithm(Algorithm):
trainer_grid_search_params = { 'cost': [2**p for p in range(-5, 16, 2)], 'gamma': [2**p for p in range(-15, 4, 2)]},
mean_std_norm_flag = False,
projector_file = "",
- save_debug_data_flag = True):
+ save_debug_data_flag = True,
+ reduced_train_data_flag = False,
+ n_train_samples = 50000):
"""
First, this function tunes the hyper-parameters of the SVM classifier using
grid search on the sub-sets of training data. Train and cross-validation
@@ -492,6 +554,15 @@ class VideoSvmPadAlgorithm(Algorithm):
Save the data, which might be usefull for debugging if ``True``.
Default: ``True``.
+ ``reduced_train_data_flag`` : :py:class:`bool`
+ Reduce the amount of final training samples if set to ``True``.
+ Default: ``False``.
+
+ ``n_train_samples`` : :py:class:`int`
+ Number of uniformly selected feature vectors per class defining the
+ sizes of sub-sets used in the final traing of the SVM.
+ Default: 50000.
+
**Returns:**
``machine`` : object
@@ -574,6 +645,12 @@ class VideoSvmPadAlgorithm(Algorithm):
real = features_norm[0:real.shape[0], :] # The array is now normalized
attack = features_norm[real.shape[0]:, :] # The array is now normalized
+ if reduced_train_data_flag:
+
+ # uniformly select subsets of features:
+ real = self.select_quasi_uniform_data_subset(real, n_train_samples)
+ attack = self.select_quasi_uniform_data_subset(attack, n_train_samples)
+
data = [np.copy(real), np.copy(attack)] # data for final training
machine = trainer.train(data) # train the machine
@@ -612,7 +689,10 @@ class VideoSvmPadAlgorithm(Algorithm):
kernel_type = self.kernel_type,
trainer_grid_search_params = self.trainer_grid_search_params,
mean_std_norm_flag = self.mean_std_norm_flag,
- projector_file = projector_file)
+ projector_file = projector_file,
+ save_debug_data_flag = self.save_debug_data_flag,
+ reduced_train_data_flag = self.reduced_train_data_flag,
+ n_train_samples = self.n_train_samples)
f = bob.io.base.HDF5File(projector_file, 'w') # open hdf5 file to save to
diff --git a/bob/pad/face/config/database/aggregated_db.py b/bob/pad/face/config/database/aggregated_db.py
index 7e15a7dd..e12abfcd 100644
--- a/bob/pad/face/config/database/aggregated_db.py
+++ b/bob/pad/face/config/database/aggregated_db.py
@@ -1,10 +1,8 @@
#!/usr/bin/env python
"""Aggregated Db is a database for face PAD experiments.
-
This database aggregates the data from 3 publicly available data-sets:
`REPLAYATTACK`_, `REPLAY-MOBILE`_ and `MSU MFSD`_.
-
You can download the data for the above databases by following the corresponding
links.
diff --git a/bob/pad/face/config/frame_diff_svm_aggregated_db.py b/bob/pad/face/config/frame_diff_svm_aggregated_db.py
new file mode 100644
index 00000000..0379d881
--- /dev/null
+++ b/bob/pad/face/config/frame_diff_svm_aggregated_db.py
@@ -0,0 +1,106 @@
+#!/usr/bin/env python2
+# -*- coding: utf-8 -*-
+
+"""
+This file contains configurations to run Frame Differences and SVM based face PAD baseline.
+The settings of the preprocessor and extractor are tuned for the Replay-attack database.
+In the SVM algorithm the amount of training data is reduced speeding-up the training for
+large data sets, such as Aggregated PAD database.
+The IQM features used in this algorithm/resource are introduced in the following papers: [WHJ15]_ and [CBVM16]_.
+"""
+
+
+#=======================================================================================
+sub_directory = 'frame_diff_svm'
+"""
+Sub-directory where results will be placed.
+
+You may change this setting using the ``--sub-directory`` command-line option
+or the attribute ``sub_directory`` in a configuration file loaded **after**
+this resource.
+"""
+
+
+#=======================================================================================
+# define preprocessor:
+
+from ..preprocessor import FrameDifference
+
+NUMBER_OF_FRAMES = None # process all frames
+CHECK_FACE_SIZE_FLAG = True # Check size of the face
+MIN_FACE_SIZE = 50 # Minimal size of the face to consider
+
+preprocessor = FrameDifference(number_of_frames = NUMBER_OF_FRAMES,
+ check_face_size_flag = CHECK_FACE_SIZE_FLAG,
+ min_face_size = MIN_FACE_SIZE)
+"""
+In the preprocessing stage the frame differences are computed for both facial and non-facial/background
+regions. In this case all frames of the input video are considered, which is defined by
+``number_of_frames = None``. The frames containing faces of the size below ``min_face_size = 50`` threshold
+are discarded. Both RGB and gray-scale videos are acceptable by the preprocessor.
+The preprocessing idea is introduced in [AM11]_.
+"""
+
+
+#=======================================================================================
+# define extractor:
+
+from ..extractor import FrameDiffFeatures
+
+WINDOW_SIZE=20
+OVERLAP=0
+
+extractor = FrameDiffFeatures(window_size=WINDOW_SIZE,
+ overlap=OVERLAP)
+"""
+In the feature extraction stage 5 features are extracted for all non-overlapping windows in
+the Frame Difference input signals. Five features are computed for each of windows in the
+facial face regions, the same is done for non-facial regions. The non-overlapping option
+is controlled by ``overlap = 0``. The length of the window is defined by ``window_size``
+argument.
+The features are introduced in the following paper: [AM11]_.
+"""
+
+
+#=======================================================================================
+# define algorithm:
+
+from ..algorithm import VideoSvmPadAlgorithm
+
+MACHINE_TYPE = 'C_SVC'
+KERNEL_TYPE = 'RBF'
+N_SAMPLES = 10000
+TRAINER_GRID_SEARCH_PARAMS = {'cost': [2**P for P in range(-3, 14, 2)], 'gamma': [2**P for P in range(-15, 0, 2)]}
+MEAN_STD_NORM_FLAG = True # enable mean-std normalization
+FRAME_LEVEL_SCORES_FLAG = True # one score per frame(!) in this case
+SAVE_DEBUG_DATA_FLAG = True # save the data, which might be useful for debugging
+REDUCED_TRAIN_DATA_FLAG = True # reduce the amount of training data in the final training stage
+N_TRAIN_SAMPLES = 50000 # number of training samples per class in the final SVM training stage
+
+algorithm = VideoSvmPadAlgorithm(machine_type = MACHINE_TYPE,
+ kernel_type = KERNEL_TYPE,
+ n_samples = N_SAMPLES,
+ trainer_grid_search_params = TRAINER_GRID_SEARCH_PARAMS,
+ mean_std_norm_flag = MEAN_STD_NORM_FLAG,
+ frame_level_scores_flag = FRAME_LEVEL_SCORES_FLAG,
+ save_debug_data_flag = SAVE_DEBUG_DATA_FLAG,
+ reduced_train_data_flag = REDUCED_TRAIN_DATA_FLAG,
+ n_train_samples = N_TRAIN_SAMPLES)
+"""
+The SVM algorithm with RBF kernel is used to classify the data into *real* and *attack* classes.
+One score is produced for each frame of the input video, ``frame_level_scores_flag = True``.
+The grid search of SVM parameters is used to select the successful settings.
+The grid search is done on the subset of training data.
+The size of this subset is defined by ``n_samples`` parameter.
+The final training of the SVM is done on the subset of training data ``reduced_train_data_flag = True``.
+The size of the subset for the final training stage is defined by the ``n_train_samples`` argument.
+The data is also mean-std normalized, ``mean_std_norm_flag = True``.
+"""
+
+
+
+
+
+
+
+
diff --git a/bob/pad/face/config/qm_svm_aggregated_db.py b/bob/pad/face/config/qm_svm_aggregated_db.py
new file mode 100644
index 00000000..4bd54236
--- /dev/null
+++ b/bob/pad/face/config/qm_svm_aggregated_db.py
@@ -0,0 +1,112 @@
+#!/usr/bin/env python2
+# -*- coding: utf-8 -*-
+
+"""
+This file contains configurations to run Image Quality Measures (IQM) and SVM based face PAD baseline.
+The settings of the preprocessor and extractor are tuned for the Replay-attack database.
+In the SVM algorithm the amount of training data is reduced speeding-up the training for
+large data sets, such as Aggregated PAD database.
+The IQM features used in this algorithm/resource are introduced in the following papers: [WHJ15]_ and [CBVM16]_.
+"""
+
+
+#=======================================================================================
+sub_directory = 'qm_svm_aggregated_db'
+"""
+Sub-directory where results will be placed.
+
+You may change this setting using the ``--sub-directory`` command-line option
+or the attribute ``sub_directory`` in a configuration file loaded **after**
+this resource.
+"""
+
+
+#=======================================================================================
+# define preprocessor:
+
+from ..preprocessor import VideoFaceCrop
+
+CROPPED_IMAGE_SIZE = (64, 64) # The size of the resulting face
+CROPPED_POSITIONS = {'topleft' : (0,0) , 'bottomright' : CROPPED_IMAGE_SIZE}
+FIXED_POSITIONS = None
+MASK_SIGMA = None # The sigma for random values areas outside image
+MASK_NEIGHBORS = 5 # The number of neighbors to consider while extrapolating
+MASK_SEED = None # The seed for generating random values during extrapolation
+CHECK_FACE_SIZE_FLAG = True # Check the size of the face
+MIN_FACE_SIZE = 50
+USE_LOCAL_CROPPER_FLAG = True # Use the local face cropping class (identical to Ivana's paper)
+RGB_OUTPUT_FLAG = True # Return RGB cropped face using local cropper
+
+preprocessor = VideoFaceCrop(cropped_image_size = CROPPED_IMAGE_SIZE,
+ cropped_positions = CROPPED_POSITIONS,
+ fixed_positions = FIXED_POSITIONS,
+ mask_sigma = MASK_SIGMA,
+ mask_neighbors = MASK_NEIGHBORS,
+ mask_seed = None,
+ check_face_size_flag = CHECK_FACE_SIZE_FLAG,
+ min_face_size = MIN_FACE_SIZE,
+ use_local_cropper_flag = USE_LOCAL_CROPPER_FLAG,
+ rgb_output_flag = RGB_OUTPUT_FLAG)
+"""
+In the preprocessing stage the face is cropped in each frame of the input video given facial annotations.
+The size of the face is normalized to ``cropped_image_size`` dimensions. The faces of the size
+below ``min_face_size`` threshold are discarded. The preprocessor is similar to the one introduced in
+[CAM12]_, which is defined by ``use_local_cropper_flag = True``. The preprocessed frame is the RGB
+facial image, which is defined by ``RGB_OUTPUT_FLAG = True``.
+"""
+
+
+#=======================================================================================
+# define extractor:
+
+from ..extractor import VideoQualityMeasure
+
+GALBALLY=True
+MSU=True
+DTYPE=None
+
+extractor = VideoQualityMeasure(galbally=GALBALLY,
+ msu=MSU,
+ dtype=DTYPE)
+"""
+In the feature extraction stage the Image Quality Measures are extracted from each frame of the preprocessed RGB video.
+The features to be computed are introduced in the following papers: [WHJ15]_ and [CBVM16]_.
+"""
+
+
+#=======================================================================================
+# define algorithm:
+
+from ..algorithm import VideoSvmPadAlgorithm
+
+MACHINE_TYPE = 'C_SVC'
+KERNEL_TYPE = 'RBF'
+N_SAMPLES = 10000
+TRAINER_GRID_SEARCH_PARAMS = {'cost': [2**P for P in range(-3, 14, 2)], 'gamma': [2**P for P in range(-15, 0, 2)]}
+MEAN_STD_NORM_FLAG = True # enable mean-std normalization
+FRAME_LEVEL_SCORES_FLAG = True # one score per frame(!) in this case
+SAVE_DEBUG_DATA_FLAG = True # save the data, which might be useful for debugging
+REDUCED_TRAIN_DATA_FLAG = True # reduce the amount of training data in the final training stage
+N_TRAIN_SAMPLES = 50000 # number of training samples per class in the final SVM training stage
+
+algorithm = VideoSvmPadAlgorithm(machine_type = MACHINE_TYPE,
+ kernel_type = KERNEL_TYPE,
+ n_samples = N_SAMPLES,
+ trainer_grid_search_params = TRAINER_GRID_SEARCH_PARAMS,
+ mean_std_norm_flag = MEAN_STD_NORM_FLAG,
+ frame_level_scores_flag = FRAME_LEVEL_SCORES_FLAG,
+ save_debug_data_flag = SAVE_DEBUG_DATA_FLAG,
+ reduced_train_data_flag = REDUCED_TRAIN_DATA_FLAG,
+ n_train_samples = N_TRAIN_SAMPLES)
+"""
+The SVM algorithm with RBF kernel is used to classify the data into *real* and *attack* classes.
+One score is produced for each frame of the input video, ``frame_level_scores_flag = True``.
+The grid search of SVM parameters is used to select the successful settings.
+The grid search is done on the subset of training data.
+The size of this subset is defined by ``n_samples`` parameter.
+The final training of the SVM is done on the subset of training data ``reduced_train_data_flag = True``.
+The size of the subset for the final training stage is defined by the ``n_train_samples`` argument.
+The data is also mean-std normalized, ``mean_std_norm_flag = True``.
+"""
+
+
diff --git a/doc/baselines.rst b/doc/baselines.rst
index 80057a25..b6d9b687 100644
--- a/doc/baselines.rst
+++ b/doc/baselines.rst
@@ -552,5 +552,20 @@ The ROC curves for the particular experiment can be downloaded from here:
------------
+.. _bob.pad.face.baselines.aggregated_db:
+
+Baselines on Aggregated Database
+--------------------------------------
+
+This section summarizes the results of baseline face PAD experiments on the Aggregated Database.
+The description of the database instance, which can be used to run face PAD experiments on the Aggregated Database is given
+here :ref:`bob.pad.face.resources.databases.aggregated_db`.
+To understand the settings of this database instance in more details you can check the
+corresponding configuration file: ``bob/pad/face/config/database/aggregated_db.py``.
+
+------------
+
+
.. include:: links.rst
+
diff --git a/doc/resources.rst b/doc/resources.rst
index 7d03b968..460737e0 100644
--- a/doc/resources.rst
+++ b/doc/resources.rst
@@ -98,3 +98,19 @@ Frame differences based features (motion analysis) + SVM for REPLAY-ATTACK
.. automodule:: bob.pad.face.config.frame_diff_svm
:members:
+
+
+.. _bob.pad.face.resources.face_pad.qm_svm_aggregated_db:
+
+Image Quality Measures as features of facial region + SVM for Aggregated Database
+===================================================================================
+
+.. automodule:: bob.pad.face.config.qm_svm_aggregated_db
+ :members:
+
+
+
+
+
+
+
diff --git a/setup.py b/setup.py
index ea22c19f..ce17ee72 100644
--- a/setup.py
+++ b/setup.py
@@ -75,7 +75,9 @@ setup(
# baselines:
'lbp-svm = bob.pad.face.config.lbp_svm',
'qm-svm = bob.pad.face.config.qm_svm',
+ 'qm-svm-aggregated-db = bob.pad.face.config.qm_svm_aggregated_db',
'frame-diff-svm = bob.pad.face.config.frame_diff_svm',
+ 'frame-diff-svm-aggregated-db = bob.pad.face.config.frame_diff_svm_aggregated_db',
],
# registered preprocessors:
--
GitLab