def train_classifier(path):
#user_names = glob(path + "/*")
metadata = load_metadata(path)
classifier = Classifier(method='svc')
'''This is for training'''
return classifier.trainClassifier(metadata)
def grid_search(self, ml_method, classifier_name, grid, splits, n_pca):
print(
compose_configuration('Grid Search', self.config['filter_latent'],
self.config['standardization'], n_pca,
self.name))
print(ml_method, classifier_name)
classifier = Classifier(self.config_filepath, ml_method,
classifier_name)
classifier.grid_search(self.name, grid, splits, n_pca)
def _fit_all(self, n_pca):
print(
compose_configuration('Fitting features',
self.config['filter_latent'],
self.config['standardization'], n_pca,
self.name))
for ml_method in self.config['ML-method-options']:
for classifier_name in self.config['Classifier-options']:
classifier = Classifier(self.config_filepath, ml_method,
classifier_name)
scores = classifier.fit(self.name, n_pca)
if ml_method in self.results:
self.results[ml_method][classifier_name] = scores
else:
self.results[ml_method] = {classifier_name: scores}
#faceCascade = cv2.CascadeClassifier(cascadePath)
font = cv2.FONT_HERSHEY_SIMPLEX
#path = "/Users/siddhartham/Sid/PycharmProjects/Face-Recognition/dataset"
path = "/Users/siddhartham/Sid/PycharmProjects/Deep-Face-Recognition-Using-Inception-Model-Keras-OpenCV-Dlib/FaceRecognition_dataset"
def assure_path_exists(path):
dir = os.path.dirname(path)
if not os.path.exists(dir):
os.makedirs(dir)
metadata = load_metadata(path)
classifier = Classifier(method='svc')
'''This is for training'''
#classifier.trainClassifier(metadata)
from glob import glob
#img_mask = path + '/**/' + '*.jpg'
user_names = glob(path + "/*")
#cam = cv2.VideoCapture('/Users/siddhartham/Sid/PycharmProjects/CNN-face-recognition/id-face_turn.MOV')
#cam = cv2.VideoCapture(0)
imagePath = '/Users/siddhartham/Sid/PycharmProjects/Deep-Face-Recognition-Using-Inception-Model-Keras-OpenCV-Dlib/Edmund_photos/Edmund_id_2.png'
# Loop
while True:
# Read the video frame
#ret, img =cam.read()
num_sents=0
for doc in testing_doc_objs.values():
for sent in doc.sentence_obj_list:
num_sents+=1
for event in sent.set_entities:
if event.type in Globals.SPECIFIC_CLASSIFIER_TYPES:
count +=1
print("NUMBER OF SUBSTANCE ABUSE EVENTS:"+ str(count))
print("NUMBER OF SENTENCES: " + str(num_sents))
##########################################
#### EVENT CLASSIFICATION #########
########################################
# Train classifiers
training_feat_extractor = FeatureExtractor(training_doc_objs)
classifiers, feature_maps, sent_info = Classifier.train_models(training_feat_extractor)
# Classify sentences - Substance abuse general, and abuse type classifications
testing_feat_extractor = FeatureExtractor(testing_doc_objs)
sent_classification_info = Classifier.get_classifications(classifiers, feature_maps, testing_feat_extractor)
# How to use:
print("\nSentence Objects with substance info:\n" + str(sent_classification_info.get_sentences_w_info(Globals.SUBSTANCE)))
print("Sentence Objects with alcohol info:\n" + str(sent_classification_info.get_sentences_w_info(Globals.ALCOHOL)))
results_file = "classifier_results.txt"
sent_classification_info.evaluate_classifications(results_file, TEST_FOLD)
########################################
#### STATUS CLASSIFICATION ##########
######################################
def __init__(self):
super(FinalModel, self).__init__()
# MULTI ATTENTION SUBNET
self.vgg_features_initial = tf.keras.applications.VGG19(
input_shape=IMG_SHAPE, include_top=False,
weights='imagenet') # VGG_feature()
self.vgg_features_initial.trainable = False
self.kmeans = Kmeans(clusters_n=2, iterations=10)
self.average_pooling_0 = Average_Pooling()
self.average_pooling_1 = Average_Pooling()
"""
self.fc_0 = Fc(CHANNELS)
self.fc_1 = Fc(CHANNELS)
"""
self.initializer = tf.keras.initializers.glorot_normal()
self.fc1_1 = tf.keras.layers.Dense(512,
input_shape=(512, ),
activation="relu",
kernel_initializer=self.initializer)
self.fc1_2 = tf.keras.layers.Dense(512,
input_shape=(512, ),
activation="relu",
kernel_initializer=self.initializer)
self.fc2_1 = tf.keras.layers.Dense(512,
activation="sigmoid",
kernel_initializer=self.initializer)
self.fc2_2 = tf.keras.layers.Dense(512,
activation="sigmoid",
kernel_initializer=self.initializer)
self.bn0 = tf.keras.layers.BatchNormalization()
self.bn1 = tf.keras.layers.BatchNormalization()
self.weighted_sum0 = WeightedSum()
self.weighted_sum1 = WeightedSum()
self.c_init = tf.random_normal_initializer(0, 0.01)
self.C = tf.Variable(initial_value=self.c_init(shape=(N_CLASSES,
SEMANTIC_SIZE),
dtype=tf.float32),
trainable=True)
# cropping net
self.crop_net0 = RCN(hidden_unit=14, map_size=14, image_size=448)
self.crop_net1 = RCN(hidden_unit=14, map_size=14, image_size=448)
self.crop0 = Crop()
self.crop1 = Crop()
# joint feature learning subnet
self.reshape_global = ReShape224()
self.reshape_local0 = ReShape224()
self.reshape_local1 = ReShape224()
self.vgg_features_global = tf.keras.applications.VGG19(
input_shape=(224, 224, 3), include_top=False,
weights='imagenet') # VGG_feature()
self.vgg_features_global.trainable = False
self.vgg_features_local0 = tf.keras.applications.VGG19(
input_shape=(224, 224, 3), include_top=False,
weights='imagenet') # VGG_feature()
self.vgg_features_local0.trainable = False
self.vgg_features_local1 = tf.keras.applications.VGG19(
input_shape=(224, 224, 3), include_top=False,
weights='imagenet') # VGG_feature()
self.vgg_features_local1.trainable = False
self.average_pooling_global = tf.keras.layers.GlobalAveragePooling2D(
) # Average_Pooling()
self.average_pooling_local0 = tf.keras.layers.GlobalAveragePooling2D(
) # Average_Pooling()
self.average_pooling_local1 = tf.keras.layers.GlobalAveragePooling2D(
) # Average_Pooling()
self.joint_net_global = JFL(N_CLASSES, SEMANTIC_SIZE, CHANNELS)
self.joint_net_local0 = JFL(N_CLASSES, SEMANTIC_SIZE, CHANNELS)
self.joint_net_local1 = JFL(N_CLASSES, SEMANTIC_SIZE, CHANNELS)
self.global_score = Scores()
self.score0 = Scores()
self.score1 = Scores()
self.classifier = Classifier(N_CLASSES)
from align import AlignDlib
from Preprocessing import convertToGrayscale, align_image, load_metadata, load_image
from Classification import Classifier
path = "/Users/siddhartham/Sid/PycharmProjects/Deep-Face-Recognition-Using-Inception-Model-Keras-OpenCV-Dlib/FaceRecognition_dataset"
def assure_path_exists(path):
dir = os.path.dirname(path)
if not os.path.exists(dir):
os.makedirs(dir)
metadata = load_metadata(path)
classifier = Classifier(method='svc')
classifier.trainClassifier(metadata)
from glob import glob
#img_mask = path + '/**/' + '*.jpg'
user_names = glob(path + "/*")
for user in user_names:
images = glob(user + "/*.jpg")
index = random.randrange(len(images))
img_path = images[index]
print(img_path)
img = load_image(img_path)
def __init__(self):
super(FinalModel, self).__init__()
# MULTI ATTENTION SUBNET
self.vgg_features_initial = tf.keras.applications.VGG19(
input_shape=IMG_SHAPE, include_top=False,
weights='imagenet') # VGG_feature()
self.vgg_features_initial.trainable = False
self.kmeans = Kmeans(clusters_n=2,
iterations=10,
batch_size=BATCH_SIZE)
self.average_pooling_0 = tf.keras.layers.GlobalAveragePooling2D()
self.average_pooling_1 = tf.keras.layers.GlobalAveragePooling2D()
self.fc1_1 = tf.keras.layers.Dense(CHANNELS,
input_shape=(512, ),
activation="relu")
self.fc1_2 = tf.keras.layers.Dense(CHANNELS,
input_shape=(512, ),
activation="relu")
self.fc2_1 = tf.keras.layers.Dense(CHANNELS, activation="sigmoid")
self.fc2_2 = tf.keras.layers.Dense(CHANNELS, activation="sigmoid")
self.bn0 = tf.keras.layers.BatchNormalization()
self.bn1 = tf.keras.layers.BatchNormalization()
self.weighted_sum0 = WeightedSum()
self.weighted_sum1 = WeightedSum()
self.c_init = tf.random_normal_initializer(0, 0.01)
self.C = tf.Variable(initial_value=self.c_init(shape=(N_CLASSES,
SEMANTIC_SIZE),
dtype=tf.float32),
trainable=True)
# cropping net
self.crop_net0 = RCN(hidden_unit=14, map_size=14, image_size=448)
self.crop_net1 = RCN(hidden_unit=14, map_size=14, image_size=448)
self.crop0 = Crop()
self.crop1 = Crop()
# joint feature learning subnet
'''shit'''
'''
self.reshape_global = ReShape224()
self.reshape_local0 = ReShape224()
self.reshape_local1 = ReShape224()
self.vgg_features_global = tf.keras.applications.VGG19(input_shape=(224, 224, 3),
include_top=False,
weights='imagenet') # VGG_feature()
self.vgg_features_global.trainable = False
self.vgg_features_local0 = tf.keras.applications.VGG19(input_shape=(224, 224, 3),
include_top=False,
weights='imagenet') # VGG_feature()
self.vgg_features_local0.trainable = False
self.vgg_features_local1 = tf.keras.applications.VGG19(input_shape=(224, 224, 3),
include_top=False,
weights='imagenet') # VGG_feature()
self.vgg_features_local1.trainable = False
self.global_score = Scores()
self.score0 = Scores()
self.score1 = Scores()
'''
self.vgg_features_global = tf.keras.applications.VGG19(
input_shape=(224, 224, 3),
include_top=False,
weights='imagenet',
pooling="avg")
self.W = tf.keras.layers.Dense(SEMANTIC_SIZE, activation="relu")
self.l2loss = tf.keras.layers.Lambda(lambda x: tf.keras.backend.sum(
tf.keras.backend.square(x[0] - x[1][:, 0]), 1, keepdims=True))
#self.fc = tf.keras.layers.Dense(N_CLASSES, activation="softmax")
# trainable class centers, C={c1, c2, ..., c_{n_classes}}
c_init = tf.random_normal_initializer(0, 0.01)
self.C = tf.Variable(initial_value=c_init(shape=(SEMANTIC_SIZE,
N_CLASSES),
dtype='float32'),
trainable=True,
name="C")
self.classifier = Classifier(N_CLASSES)