def init_dataset(self):
dataset = Flags.dataset
if dataset == 'UCF101':
self.dataset = ucf_ts.ucf_dataset(
split_number=split,
is_training_split=False,
batch_size=self.batch_size,
epoch=1,
test_crop=test_crop,
eval_type=eval_type,
video_split=self.video_split,
new_length=new_length,
image_size=self._IMAGE_SIZE,
frame_counts=self._FRAME_COUNTS,
prefetch_buffer_size=self.batch_size).test_dataset()
elif dataset == 'hmdb51':
self.dataset = hmdb_ts.hmdb_dataset(
split_number=split,
is_training_split=False,
batch_size=self.batch_size,
epoch=1,
test_crop=test_crop,
eval_type=eval_type,
video_split=self.video_split,
new_length=new_length,
image_size=self._IMAGE_SIZE,
frame_counts=self._FRAME_COUNTS,
prefetch_buffer_size=self.batch_size).test_dataset()
iter = tf.data.Iterator.from_structure(self.dataset.output_types,
self.dataset.output_shapes)
self.next_element = iter.get_next()
self.training_init_op = iter.make_initializer(self.dataset)
flow_train = tf.placeholder (tf.float32,
[None, _FRAME_COUNTS * factor, _IMAGE_SIZE, _IMAGE_SIZE, 20])
y_ = tf.placeholder (tf.float32, [None, _NUM_CLASSES])
dataset = Flags.dataset
if dataset == 'UCF101':
dataset = ucf_ts.ucf_dataset (split_number=split, is_training_split=False,
batch_size=batch_size, epoch=1,test_crop = test_crop,
eval_type=eval_type, frame_counts=_FRAME_COUNTS,
image_size=_IMAGE_SIZE,
prefetch_buffer_size=batch_size).test_dataset ()
elif dataset == 'hmdb51':
dataset = hmdb_ts.hmdb_dataset (split_number=split, is_training_split=False,
batch_size=batch_size, epoch=1,test_crop = test_crop,
eval_type=eval_type, frame_counts=_FRAME_COUNTS,
image_size=_IMAGE_SIZE,
prefetch_buffer_size=batch_size).test_dataset ()
iter = dataset.make_initializable_iterator ()
next_element = iter.get_next ()
if model_type == 'resnet50':
base_net = 'TS_resnet50'
elif model_type == 'resnet101':
base_net = 'ts_resnet101'
elif model_type == 'inception_v1':
base_net = 'ts_inception_v1'
with tf.variable_scope ('RGB', reuse=tf.AUTO_REUSE):