def get_next(config, model_config, lstm_config, unroll_length):
data_augmentation_options = [
preprocessor_builder.build(step)
for step in train_config.data_augmentation_options
]
return seq_dataset_builder.build(config,
model_config,
lstm_config,
unroll_length,
data_augmentation_options,
batch_size=train_config.batch_size)
def get_next(config, model_config, lstm_config, unroll_length):
data_augmentation_options = [
preprocessor_builder.build(step)
for step in train_config.data_augmentation_options
]
return seq_dataset_builder.build(
config,
model_config,
lstm_config,
unroll_length,
data_augmentation_options,
batch_size=train_config.batch_size)
def test_raises_error_without_input_paths(self):
input_reader_text_proto = """
shuffle: false
num_readers: 1
load_instance_masks: true
"""
input_reader_proto = input_reader_pb2.InputReader()
text_format.Merge(input_reader_text_proto, input_reader_proto)
configs = self._get_model_configs_from_proto()
with self.assertRaises(ValueError):
_ = seq_dataset_builder.build(input_reader_proto,
configs['model'],
configs['lstm_model'],
unroll_length=1)
def test_video_input_reader(self, video_input_type):
input_reader_proto = input_reader_pb2.InputReader()
text_format.Merge(self._get_input_proto(video_input_type),
input_reader_proto)
configs = self._get_model_configs_from_proto()
tensor_dict = seq_dataset_builder.build(input_reader_proto,
configs['model'],
configs['lstm_model'],
unroll_length=1)
all_dict = self._create_training_dict(tensor_dict)
self.assertEqual((1, 32, 32, 3), all_dict['image0'].shape)
self.assertEqual(4, all_dict['groundtruth_boxes0'].shape[1])
def test_raises_error_without_input_paths(self):
input_reader_text_proto = """
shuffle: false
num_readers: 1
load_instance_masks: true
"""
input_reader_proto = input_reader_pb2.InputReader()
text_format.Merge(input_reader_text_proto, input_reader_proto)
configs = self._get_model_configs_from_proto()
with self.assertRaises(ValueError):
_ = seq_dataset_builder.build(
input_reader_proto,
configs['model'],
configs['lstm_model'],
unroll_length=1)
def test_video_input_reader(self, video_input_type):
input_reader_proto = input_reader_pb2.InputReader()
text_format.Merge(
self._get_input_proto(video_input_type), input_reader_proto)
configs = self._get_model_configs_from_proto()
tensor_dict = seq_dataset_builder.build(
input_reader_proto,
configs['model'],
configs['lstm_model'],
unroll_length=1)
all_dict = self._create_training_dict(tensor_dict)
self.assertEqual((1, 32, 32, 3), all_dict['image0'].shape)
self.assertEqual(4, all_dict['groundtruth_boxes0'].shape[1])
def test_build_with_data_augmentation(self):
input_reader_proto = input_reader_pb2.InputReader()
text_format.Merge(
self._get_input_proto('tf_record_video_input_reader'),
input_reader_proto)
configs = self._get_model_configs_from_proto()
data_augmentation_options = [
preprocessor_builder.build(
self._get_data_augmentation_preprocessor_proto())
]
tensor_dict = seq_dataset_builder.build(
input_reader_proto,
configs['model'],
configs['lstm_model'],
unroll_length=1,
data_augmentation_options=data_augmentation_options)
all_dict = self._create_training_dict(tensor_dict)
self.assertEqual((1, 32, 32, 3), all_dict['image0'].shape)
self.assertEqual(4, all_dict['groundtruth_boxes0'].shape[1])
def test_build_with_data_augmentation(self):
input_reader_proto = input_reader_pb2.InputReader()
text_format.Merge(
self._get_input_proto('tf_record_video_input_reader'),
input_reader_proto)
configs = self._get_model_configs_from_proto()
data_augmentation_options = [
preprocessor_builder.build(
self._get_data_augmentation_preprocessor_proto())
]
tensor_dict = seq_dataset_builder.build(
input_reader_proto,
configs['model'],
configs['lstm_model'],
unroll_length=1,
data_augmentation_options=data_augmentation_options)
all_dict = self._create_training_dict(tensor_dict)
self.assertEqual((1, 32, 32, 3), all_dict['image0'].shape)
self.assertEqual(4, all_dict['groundtruth_boxes0'].shape[1])
def get_next(config, model_config, lstm_config, unroll_length):
return seq_dataset_builder.build(config, model_config, lstm_config,
unroll_length)
def get_next(config, model_config, lstm_config, unroll_length):
return seq_dataset_builder.build(config, model_config, lstm_config,
unroll_length)