def set_cover(self, uploaded_image):
handler = ImageHandler()
images = handler.handle_cover_image(uploaded_image)
s3 = AmazonS3()
s3.upload_file(images['default'], '/u/%d/%s' % (self.musician.id, self.COVER_IMG_NAME))
if not self.media_list.filter(media_type__name = "cover_photo"):
self.media_list.add(Media.objects.create(media = "cover", media_type = self.__type("cover_photo")))
def set_avatar(self, uploaded_image):
handler = ImageHandler()
images = handler.handle_profile_images(uploaded_image)
s3 = AmazonS3()
s3.upload_file(images['default'], '/u/%d/%s' % (self.musician.id, self.AVATAR_IMG_NAME))
s3.upload_file(images['small'], '/u/%d/%s' % (self.musician.id, self.AVATAR_SMALL_IMG_NAME))
if not self.media_list.filter(media_type__name = "avatar"):
self.media_list.add(Media.objects.create(media = "avatar", media_type = self.__type("avatar")))
def train():
THRESHOLD = 0.95 # Proportion of representation when choosing the best eigen vectors
args = arguments.get_arguments()
# Training set characteristics
TRAINING_SET_SIZE = args.training_set_size #Amount of training images for each individual
# Testing set characteristics
TESTING_SET_SIZE = args.testing_set_size #Amount of testing images for each individual
# Load images
if (args.verbose):
print('Loading images')
training_images, training_classes, testing_images, testing_classes, profile, profiles_map = imageHandler.load_images(training_size=TRAINING_SET_SIZE, testing_size=TESTING_SET_SIZE, image_dir=args.images)
# Create matrix out of images
matrix = (np.matrix(training_images)).T / 255.
# Calculate mean different faces
mean = matrix.mean(axis=1)
imageHandler.save_image(mean, "mean.pgm")
# Center matrix
centered_matrix = (matrix - mean)
# Calculate the covariance matrix
# Calculate centered matrix * transposed centered matrix to get a similar matrix to the one of the covariance
if(args.verbose):
print('Calculating covariance matrix')
covariance_matrix = (centered_matrix.T).dot(centered_matrix)
# Calculate eigen values and eigen vectors
if (args.verbose):
print('Calculating eigen values and eigen vectors')
if args.type == "pca":
eig_values, eig_vectors = calculate_eigen(matrix=covariance_matrix, qr_method=args.qr_method, eig_method=args.eig_method)
elif args.type == "kpca":
eig_values, eig_vectors = calculate_kernel_eigen(matrix=covariance_matrix, qr_method=args.qr_method, eig_method=args.eig_method)
else:
raise ValueError("The type is not supported")
# Get best eigenvalues
if(args.verbose):
print('Getting representative eigen values')
best_eig_vectors = get_best_eig_vectors(eig_values, eig_vectors, THRESHOLD)
# Calculate images
# http://blog.manfredas.com/eigenfaces-tutorial/
eigen_faces = centered_matrix.dot(best_eig_vectors)
# Normalize eigen faces optimization
row_sums = np.linalg.norm(eigen_faces, axis=0)
eigen_faces = np.divide(eigen_faces,row(row_sums))
# Project values on eigen vectors
if(args.verbose):
print('Projecting values on eigen vectors')
projected_values = eigen_faces.T.dot(centered_matrix)
# Write image files
imageHandler.save_images(images=eigen_faces.T)
# Generate matrices from loaded images
test_matrix = np.matrix(testing_images).T/255.
test_matrix = test_matrix - mean
testing_set = eigen_faces.T.dot(test_matrix)
#Test images
clf = svm.LinearSVC()
# Training classifier with provided data set+group
clf.fit(projected_values.T, training_classes)
classifications = clf.score(testing_set.T, testing_classes)
return mean, eigen_faces, clf, classifications, profile, profiles_map
def train_kpca():
THRESHOLD = 0.95 # Proportion of representation when choosing the best eigen vectors
args = arguments.get_arguments()
# Testing set characteristics
TESTING_SET_SIZE = args.testing_set_size
# Training set characteristics
TRAINING_SET_SIZE = args.training_set_size # Amount of training images for each individual
# Load images
if (args.verbose):
print('Loading images')
training_images, training_classes, testing_images, testing_classes, profiles, profile_map = imageHandler.load_images(
training_size=TRAINING_SET_SIZE,
testing_size=TESTING_SET_SIZE,
image_dir=args.images)
TRAINING_IMAGES = len(training_images)
# Create matrix out of images
matrix = ((np.matrix(training_images)) - 127.5) / 127.5
# Calculate K matrix and get eigen values and eigen vectors
if (args.verbose):
print('Calculating eigen values and eigen vectors')
eig_values, eig_vectors, K = calculate_kernel_eigen(
matrix=matrix,
qr_method=args.qr_method,
eig_method=args.eig_method,
TRAINING_IMAGES=TRAINING_IMAGES)
# Get best eigenvalues
if (args.verbose):
print('Getting representative eigen values')
eigen_faces = get_best_eig_vectors(eig_values, eig_vectors, THRESHOLD)
if (args.verbose):
print('number of eigen faces used: ', eigen_faces.shape[1])
training_projection = np.dot(K.T, eigen_faces)
clf = svm.LinearSVC()
clf.fit(training_projection, training_classes)
testing_projection = get_testing_data(matrix, eigen_faces, testing_images,
K)
classifications = clf.score(testing_projection, testing_classes)
return clf, eigen_faces, matrix, K, classifications, profiles, profile_map
