def get_data(self, tfrecord_dir, batch_size):
# Read video data having labels A to T.
self.dataset_trainAT = get_split_mfcc_lips('trainAT', tfrecord_dir)
_, self.videos_train, self.labels_trainAT = load_batch_mfcc_lips(
self.dataset_trainAT, batch_size=batch_size)
# Read audio data having labels U to Z.
self.dataset_trainUZ = get_split_mfcc_lips('trainUZ', tfrecord_dir)
self.mfccs_train, _, self.labels_trainUZ = load_batch_mfcc_lips(
self.dataset_trainUZ, batch_size=batch_size, is_training=False)
# Read data for test.
self.dataset_test = get_split_mfcc_lips('validation', tfrecord_dir)
_, self.videos_test, self.labels_test = load_batch_mfcc_lips(
self.dataset_test, batch_size=batch_size, is_training=False)
# Read all the data with labels A to T once before the beginning
# of training in order to do KNN later.
self.all_mfccs, self.all_videos, all_labels = load_batch_mfcc_lips(
self.dataset_trainAT,
shuffle=False,
batch_size=self.dataset_trainAT.num_samples)
# Some methods in `TrainClassify` use `self.dataset_train`.
# This is a small problem and should be changed.
self.dataset_train = self.dataset_trainAT
return self.dataset_trainAT
def get_data(self, tfrecord_dir, batch_size):
self.dataset_train = get_split_mfcc_lips('trainAT', tfrecord_dir)
_, self.videos_train, self.labels_train = load_batch_mfcc_lips(
self.dataset_train, batch_size=batch_size)
self.dataset_test = get_split_mfcc_lips('validation', tfrecord_dir)
_, self.videos_test, self.labels_test = load_batch_mfcc_lips(
self.dataset_test, batch_size=batch_size, is_training=False)
return self.dataset_train
def get_data(self, split_name, tfrecord_dir, batch_size, shuffle):
self.dataset = get_split_mfcc_lips(split_name, tfrecord_dir)
if batch_size is None:
batch_size = self.dataset.num_samples
_, self.videos, self.labels = load_batch_mfcc_lips(
self.dataset,
batch_size=batch_size,
shuffle=shuffle,
is_training=False)
return self.dataset