def _load_data(self):
# Load all records
(x_train, y_train) = load_data(self.__dataset_dir, 'lfw_crop', self.__img_size)
# Merge them (we split them by classes)
x = x_train
y = y_train
# Reshape x for tensorflow
x = x.reshape((x.shape[0],) + self.get_data_shape())
# Normalize x to [0, 1]
x = self._scale_data(x)
# Split the records by classes and store it
data = {i: x[y == i] for i in np.unique(y)}
# Remove classes with to less records
if self.__min_images_per_class is not None and self.__min_images_per_class > 1:
new_data = {}
counter = 0
for i in data.keys():
if len(data[i]) >= self.__min_images_per_class:
new_data[counter] = data[i]
counter += 1
print("By using only classes with at least {} elements, the dataset was reduced from {} to {} records.".format(
self.__min_images_per_class, len(data), len(new_data)
))
data = new_data
return data
def _load_data(self):
# Load all records
(x, y) = load_data(self.__dataset_dir, 'tiny_image_net', self.__img_size)
# Reshape x for tensorflow
x = x.reshape((x.shape[0],) + self.get_data_shape())
# Normalize x to [-1, 1]
x = self._scale_data(x)
# x = x.astype(np.float32) / 255
# Split the records by classes and store it
y = y.reshape((y.shape[0],))
return {i: x[y == i] for i in np.unique(y)}
def _load_data(self):
# Load all records
(x_train, y_train) = load_data(self.__dataset_dir, 'facescrub', self.__img_size)
# Merge them (we split them by classes)
x = x_train
y = y_train
# Reshape x for tensorflow
x = x.reshape((x.shape[0],) + self.get_data_shape())
# Normalize x to [0, 1]
x = self._scale_data(x)
# Split the records by classes and store it
return {i: x[y == i] for i in np.unique(y)}