class IqmFacePad(FacePad):
def __init__(self, svm_model_path, threshold, name='IqmFacePad'):
self.features_extractor = IqmFeaturesExtractor()
self.classifier = RbfSvc(name='RbfSvc')
self.classifier.load(svm_model_path)
self.scores = []
self.threshold = threshold
self.finished = False
super(IqmFacePad, self).__init__(name, threshold)
def process(self, im):
dict_images = {'0': im}
dict_features = self.features_extractor.run(dict_images)
X = {'features': np.reshape(dict_features['0'], (1, dict_features['0'].shape[0]))}
X = self.classifier.run(X)
score = X['scores'][0]
self.scores.append(score)
self.finished = True
def is_finished(self):
return self.finished
def reset(self):
self.scores = []
self.finished = False
def get_decision(self):
average_score = np.mean(self.scores)
self.scores = []
if average_score < self.threshold:
return 'ATTACK', average_score
else:
return 'NO_ATTACK', average_score
def test_transform_calls_sklearn_transform(self):
features = self.X['features'].copy()
svc = RbfSvc()
svc.run(self.X)
np.testing.assert_array_equal(SVC.decision_function.call_args[0][0],
features)
def __init__(self, svm_model_path, threshold, name='IqmFacePad'):
self.features_extractor = IqmFeaturesExtractor()
self.classifier = RbfSvc(name='RbfSvc')
self.classifier.load(svm_model_path)
self.scores = []
self.threshold = threshold
self.finished = False
super(IqmFacePad, self).__init__(name, threshold)
def test_save_model(self):
svc = RbfSvc(name='TestSvc')
svc.save(self.base_path)
h5py.File.assert_called_once_with(
os.path.join(self.base_path, 'processors/TestSvc.h5'), 'w')
self.h5f.create_dataset.assert_called_once_with(
'data', data=np.array('Model dump'))
def test_load_model(self):
svc = RbfSvc(name='TestSvc')
svc.load(self.base_path)
h5py.File.assert_called_once_with(
os.path.join(self.base_path, 'processors/TestSvc.h5'), 'r')
if sys.version_info[0] < 3:
pickle.loads.assert_called_once_with('Model')
else:
pickle.loads.assert_called_once_with('Model', encoding='latin1')
self.assertEquals('Model', svc._model)
def test_fit_calls_sklearn_fit(self):
svc = RbfSvc()
svc.fit(self.X)
labels = copy.deepcopy(self.X['labels'])
expected_labels = copy.deepcopy(labels)
expected_labels[labels > 0] = 0
expected_labels[labels == 0] = 1
np.testing.assert_array_equal(SVC.fit.call_args[0][0],
self.X['features'])
np.testing.assert_array_equal(SVC.fit.call_args[0][1], expected_labels)
def get_pipeline_average_features_scaled_rbfsvc(name):
pipeline = Pipeline(
name,
[AverageFeatures(),
FeaturesScaleNormalizer(),
RbfSvc(name='RbfSvc')])
return pipeline
def test_constructor_calls_sklearn_constructor(self):
RbfSvc(c=0.94, gamma=0.1)
SVC.__init__.assert_called_once_with(C=0.94, gamma=0.1)