Commit 91857aa2 authored by Pavel KORSHUNOV's avatar Pavel KORSHUNOV
Browse files

Merge branch 'harmonize_algorithms' into 'master'

Harmonizing algorithms from bob.pad.face and bob.pad.voice, fixes issue#16

See merge request !28
parents d9fcf2da f1a35ae3
Pipeline #16820 passed with stages
in 11 minutes and 52 seconds
from .utils import *
from . import database
from . import algorithm
from . import tools
......
#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Fri Aug 25 09:29:02 2017
@author: Olegs Nikisins
"""
#==============================================================================
# Import what is needed here:
from bob.pad.base.algorithm import Algorithm
from bob.bio.video.utils import FrameContainer
import numpy as np
from sklearn import linear_model
import bob.io.base
from bob.pad.base.utils import convert_frame_cont_to_array, convert_list_of_frame_cont_to_array, mean_std_normalize, \
norm_train_data
#==============================================================================
# Main body :
class LogRegr(Algorithm):
"""
This class is designed to train Logistic Regression classifier given Frame Containers
with features of real and attack classes. The procedure is the following:
1. First, the input data is mean-std normalized using mean and std of the
real class only.
2. Second, the Logistic Regression classifier is trained on normalized
input features.
3. The input features are next classified using pre-trained LR machine.
**Parameters:**
``C`` : :py:class:`float`
Inverse of regularization strength in LR classifier; must be a positive.
Like in support vector machines, smaller values specify stronger
regularization. Default: 1.0 .
``frame_level_scores_flag`` : :py:class:`bool`
Return scores for each frame individually if True. Otherwise, return a
single score per video. Default: ``False``.
``subsample_train_data_flag`` : :py:class:`bool`
Uniformly subsample the training data if ``True``. Default: ``False``.
``subsampling_step`` : :py:class:`int`
Training data subsampling step, only valid is
``subsample_train_data_flag = True``. Default: 10 .
``subsample_videos_flag`` : :py:class:`bool`
Uniformly subsample the training videos if ``True``. Default: ``False``.
``video_subsampling_step`` : :py:class:`int`
Training videos subsampling step, only valid is
``subsample_videos_flag = True``. Default: 3 .
"""
def __init__(self,
C=1,
frame_level_scores_flag=False,
subsample_train_data_flag=False,
subsampling_step=10,
subsample_videos_flag=False,
video_subsampling_step=3):
Algorithm.__init__(
self,
C=C,
frame_level_scores_flag=frame_level_scores_flag,
subsample_train_data_flag=subsample_train_data_flag,
subsampling_step=subsampling_step,
subsample_videos_flag=subsample_videos_flag,
video_subsampling_step=video_subsampling_step,
performs_projection=True,
requires_projector_training=True)
self.C = C
self.frame_level_scores_flag = frame_level_scores_flag
self.subsample_train_data_flag = subsample_train_data_flag
self.subsampling_step = subsampling_step
self.subsample_videos_flag = subsample_videos_flag
self.video_subsampling_step = video_subsampling_step
self.lr_machine = None # this argument will be updated with pretrained LR machine
self.features_mean = None # this argument will be updated with features mean
self.features_std = None # this argument will be updated with features std
# names of the arguments of the pretrained LR machine to be saved/loaded to/from HDF5 file:
self.lr_param_keys = ["C", "classes_", "coef_", "intercept_"]
#==========================================================================
def train_lr(self, real, attack, C):
"""
Train LR classifier given real and attack classes. Prior to training
the data is mean-std normalized.
**Parameters:**
``real`` : 2D :py:class:`numpy.ndarray`
Training features for the real class.
``attack`` : 2D :py:class:`numpy.ndarray`
Training features for the attack class.
``C`` : :py:class:`float`
Inverse of regularization strength in LR classifier; must be a positive.
Like in support vector machines, smaller values specify stronger
regularization. Default: 1.0 .
**Returns:**
``machine`` : object
A trained LR machine.
``features_mean`` : 1D :py:class:`numpy.ndarray`
Mean of the features.
``features_std`` : 1D :py:class:`numpy.ndarray`
Standart deviation of the features.
"""
real, attack, features_mean, features_std = norm_train_data(
real, attack)
# real and attack - are now mean-std normalized
X = np.vstack([real, attack])
Y = np.hstack([np.zeros(len(real)), np.ones(len(attack))])
machine = linear_model.LogisticRegression(C=C)
machine.fit(X, Y)
return machine, features_mean, features_std
#==========================================================================
def save_lr_machine_and_mean_std(self, projector_file, machine,
features_mean, features_std):
"""
Saves the LR machine, features mean and std to the hdf5 file.
The absolute name of the file is specified in ``projector_file`` string.
**Parameters:**
``projector_file`` : :py:class:`str`
Absolute name of the file to save the data to, as returned by
``bob.pad.base`` framework.
``machine`` : object
The LR machine to be saved. As returned by sklearn.linear_model
module.
``features_mean`` : 1D :py:class:`numpy.ndarray`
Mean of the features.
``features_std`` : 1D :py:class:`numpy.ndarray`
Standart deviation of the features.
"""
f = bob.io.base.HDF5File(projector_file,
'w') # open hdf5 file to save to
for key in self.lr_param_keys: # ["C", "classes_", "coef_", "intercept_"]
data = getattr(machine, key)
f.set(key, data)
f.set("features_mean", features_mean)
f.set("features_std", features_std)
del f
#==========================================================================
def subsample_train_videos(self, training_features, step):
"""
Uniformly select subset of frmae containes from the input list
**Parameters:**
``training_features`` : [FrameContainer]
A list of FrameContainers
``step`` : :py:class:`int`
Data selection step.
**Returns:**
``training_features_subset`` : [FrameContainer]
A list with selected FrameContainers
"""
indexes = range(0, len(training_features), step)
training_features_subset = [training_features[x] for x in indexes]
return training_features_subset
#==========================================================================
def train_projector(self, training_features, projector_file):
"""
Train LR for feature projection and save them to files.
The ``requires_projector_training = True`` flag must be set to True
to enable this function.
**Parameters:**
``training_features`` : [[FrameContainer], [FrameContainer]]
A list containing two elements: [0] - a list of Frame Containers with
feature vectors for the real class; [1] - a list of Frame Containers with
feature vectors for the attack class.
``projector_file`` : :py:class:`str`
The file to save the trained projector to, as returned by the
``bob.pad.base`` framework.
"""
# training_features[0] - training features for the REAL class.
# training_features[1] - training features for the ATTACK class.
if self.subsample_videos_flag: # subsample videos of the real class
real = convert_list_of_frame_cont_to_array(
self.subsample_train_videos(
training_features[0],
self.video_subsampling_step)) # output is array
else:
real = convert_list_of_frame_cont_to_array(
training_features[0]) # output is array
if self.subsample_train_data_flag:
real = real[range(0, len(real), self.subsampling_step), :]
if self.subsample_videos_flag: # subsample videos of the real class
attack = convert_list_of_frame_cont_to_array(
self.subsample_train_videos(
training_features[1],
self.video_subsampling_step)) # output is array
else:
attack = convert_list_of_frame_cont_to_array(
training_features[1]) # output is array
if self.subsample_train_data_flag:
attack = attack[range(0, len(attack), self.subsampling_step), :]
# Train the LR machine and get normalizers:
machine, features_mean, features_std = self.train_lr(
real=real, attack=attack, C=self.C)
# Save the LR machine and normalizers:
self.save_lr_machine_and_mean_std(projector_file, machine,
features_mean, features_std)
#==========================================================================
def load_lr_machine_and_mean_std(self, projector_file):
"""
Loads the machine, features mean and std from the hdf5 file.
The absolute name of the file is specified in ``projector_file`` string.
**Parameters:**
``projector_file`` : :py:class:`str`
Absolute name of the file to load the trained projector from, as
returned by ``bob.pad.base`` framework.
**Returns:**
``machine`` : object
The loaded LR machine. As returned by sklearn.linear_model module.
``features_mean`` : 1D :py:class:`numpy.ndarray`
Mean of the features.
``features_std`` : 1D :py:class:`numpy.ndarray`
Standart deviation of the features.
"""
f = bob.io.base.HDF5File(projector_file,
'r') # file to read the machine from
# initialize the machine:
machine = linear_model.LogisticRegression()
# set the params of the machine:
for key in self.lr_param_keys: # ["C", "classes_", "coef_", "intercept_"]
data = f.read(key)
setattr(machine, key, data)
features_mean = f.read("features_mean")
features_std = f.read("features_std")
del f
return machine, features_mean, features_std
#==========================================================================
def load_projector(self, projector_file):
"""
Loads the machine, features mean and std from the hdf5 file.
The absolute name of the file is specified in ``projector_file`` string.
This function sets the arguments ``self.lr_machine``, ``self.features_mean``
and ``self.features_std`` of this class with loaded machines.
The function must be capable of reading the data saved with the
:py:meth:`train_projector` method of this class.
Please register `performs_projection = True` in the constructor to
enable this function.
**Parameters:**
``projector_file`` : :py:class:`str`
The file to read the projector from, as returned by the
``bob.pad.base`` framework. In this class the names of the files to
read the projectors from are modified, see ``load_machine`` and
``load_cascade_of_machines`` methods of this class for more details.
"""
lr_machine, features_mean, features_std = self.load_lr_machine_and_mean_std(
projector_file)
self.lr_machine = lr_machine
self.features_mean = features_mean
self.features_std = features_std
#==========================================================================
def project(self, feature):
"""
This function computes a vector of scores for each sample in the input
array of features. The following steps are apllied:
1. First, the input data is mean-std normalized using mean and std of the
real class only.
2. The input features are next classified using pre-trained LR machine.
Set ``performs_projection = True`` in the constructor to enable this function.
It is assured that the :py:meth:`load_projector` was **called before** the
``project`` function is executed.
**Parameters:**
``feature`` : FrameContainer or 2D :py:class:`numpy.ndarray`
Two types of inputs are accepted.
A Frame Container conteining the features of an individual,
see ``bob.bio.video.utils.FrameContainer``.
Or a 2D feature array of the size (N_samples x N_features).
**Returns:**
``scores`` : 1D :py:class:`numpy.ndarray`
Vector of scores. Scores for the real class are expected to be
higher, than the scores of the negative / attack class.
In this case scores are probabilities.
"""
# 1. Convert input array to numpy array if necessary.
if isinstance(
feature,
FrameContainer): # if FrameContainer convert to 2D numpy array
features_array = convert_frame_cont_to_array(feature)
else:
features_array = feature
features_array_norm, _, _ = mean_std_normalize(
features_array, self.features_mean, self.features_std)
scores = self.lr_machine.predict_proba(features_array_norm)[:, 0]
return scores
#==========================================================================
def score(self, toscore):
"""
Returns a probability of a sample being a real class.
**Parameters:**
``toscore`` : 1D :py:class:`numpy.ndarray`
Vector with scores for each frame/sample defining the probability
of the frame being a sample of the real class.
**Returns:**
``score`` : [:py:class:`float`]
If ``frame_level_scores_flag = False`` a single score is returned.
One score per video. This score is placed into a list, because
the ``score`` must be an iterable.
Score is a probability of a sample being a real class.
If ``frame_level_scores_flag = True`` a list of scores is returned.
One score per frame/sample.
"""
if self.frame_level_scores_flag:
score = list(toscore)
else:
score = [np.mean(toscore)] # compute a single score per video
return score
#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Mon Aug 28 16:47:47 2017
@author: Olegs Nikisins
"""
# ==============================================================================
# Import what is needed here:
from bob.pad.base.algorithm import Algorithm
from bob.bio.video.utils import FrameContainer
import numpy as np
import bob.io.base
from sklearn import mixture
from bob.pad.base.utils import convert_frame_cont_to_array, mean_std_normalize
# ==============================================================================
# Main body :
class OneClassGMM(Algorithm):
"""
This class is designed to train a OneClassGMM based PAD system. The OneClassGMM is trained
using data of one class (real class) only. The procedure is the following:
1. First, the training data is mean-std normalized using mean and std of the
real class only.
2. Second, the OneClassGMM with ``n_components`` Gaussians is trained using samples
of the real class.
3. The input features are next classified using pre-trained OneClassGMM machine.
**Parameters:**
``n_components`` : :py:class:`int`
Number of Gaussians in the OneClassGMM. Default: 1 .
``random_state`` : :py:class:`int`
A seed for the random number generator used in the initialization of
the OneClassGMM. Default: 7 .
``frame_level_scores_flag`` : :py:class:`bool`
Return scores for each frame individually if True. Otherwise, return a
single score per video. Default: False.
"""
def __init__(self,
n_components=1,
random_state=3,
frame_level_scores_flag=False):
Algorithm.__init__(
self,
n_components=n_components,
random_state=random_state,
frame_level_scores_flag=frame_level_scores_flag,
performs_projection=True,
requires_projector_training=True)
self.n_components = n_components
self.random_state = random_state
self.frame_level_scores_flag = frame_level_scores_flag
self.machine = None # this argument will be updated with pretrained OneClassGMM machine
self.features_mean = None # this argument will be updated with features mean
self.features_std = None # this argument will be updated with features std
# names of the arguments of the pretrained OneClassGMM machine to be saved/loaded to/from HDF5 file:
self.gmm_param_keys = [
"covariance_type", "covariances_", "lower_bound_", "means_",
"n_components", "weights_", "converged_", "precisions_",
"precisions_cholesky_"
]
# ==========================================================================
def train_gmm(self, real, n_components, random_state):
"""
Train OneClassGMM classifier given real class. Prior to the training the data is
mean-std normalized.
**Parameters:**
``real`` : 2D :py:class:`numpy.ndarray`
Training features for the real class.
``n_components`` : :py:class:`int`
Number of Gaussians in the OneClassGMM. Default: 1 .
``random_state`` : :py:class:`int`
A seed for the random number generator used in the initialization of
the OneClassGMM. Default: 7 .
**Returns:**
``machine`` : object
A trained OneClassGMM machine.
``features_mean`` : 1D :py:class:`numpy.ndarray`
Mean of the features.
``features_std`` : 1D :py:class:`numpy.ndarray`
Standart deviation of the features.
"""
features_norm, features_mean, features_std = mean_std_normalize(
real)
# real is now mean-std normalized
machine = mixture.GaussianMixture(
n_components=n_components,
random_state=random_state,
covariance_type='full')
machine.fit(features_norm)
return machine, features_mean, features_std
# ==========================================================================
def save_gmm_machine_and_mean_std(self, projector_file, machine,
features_mean, features_std):
"""
Saves the OneClassGMM machine, features mean and std to the hdf5 file.
The absolute name of the file is specified in ``projector_file`` string.
**Parameters:**
``projector_file`` : :py:class:`str`
Absolute name of the file to save the data to, as returned by
``bob.pad.base`` framework.
``machine`` : object
The OneClassGMM machine to be saved. As returned by sklearn.linear_model
module.
``features_mean`` : 1D :py:class:`numpy.ndarray`
Mean of the features.
``features_std`` : 1D :py:class:`numpy.ndarray`
Standart deviation of the features.
"""
f = bob.io.base.HDF5File(projector_file,
'w') # open hdf5 file to save to
for key in self.gmm_param_keys:
data = getattr(machine, key)
f.set(key, data)
f.set("features_mean", features_mean)
f.set("features_std", features_std)
del f
# ==========================================================================
def train_projector(self, training_features, projector_file):
"""
Train OneClassGMM for feature projection and save it to file.
The ``requires_projector_training = True`` flag must be set to True
to enable this function.
**Parameters:**
``training_features`` : [[FrameContainer], [FrameContainer]]
A list containing two elements: [0] - a list of Frame Containers with
feature vectors for the real class; [1] - a list of Frame Containers with
feature vectors for the attack class.