def create_input_fn(dataset_dir):
prepare(dataset_dir)
train_data_file = RECORD_FILE_NAME_FORMAT.format(dataset_dir,
ModeKeys.TRAIN)
eval_data_file = RECORD_FILE_NAME_FORMAT.format(dataset_dir, ModeKeys.EVAL)
meta_data_filename = MEAT_DATA_FILENAME_FORMAT.format(dataset_dir)
train_input_fn = create_input_data_fn(mode=ModeKeys.TRAIN,
pipeline_config=PipelineConfig(
name='TFRecordImagePipeline',
dynamic_pad=False,
params={
'data_files':
train_data_file,
'meta_data_file':
meta_data_filename
}))
eval_input_fn = create_input_data_fn(mode=ModeKeys.EVAL,
pipeline_config=PipelineConfig(
name='TFRecordImagePipeline',
dynamic_pad=False,
params={
'data_files':
eval_data_file,
'meta_data_file':
meta_data_filename
}))
return train_input_fn, eval_input_fn
def create_sequence_dataset_input_fn(dataset_dir, prepare_fn,
record_file_name_format,
meta_data_file_name_format,
bucket_boundaries):
prepare_fn(dataset_dir)
train_data_file = record_file_name_format.format(dataset_dir, Modes.TRAIN)
eval_data_file = record_file_name_format.format(dataset_dir, Modes.EVAL)
meta_data_filename = meta_data_file_name_format.format(dataset_dir)
train_input_fn = create_input_data_fn(
mode=Modes.TRAIN,
pipeline_config=PipelineConfig(module='TFRecordSequencePipeline',
dynamic_pad=True,
bucket_boundaries=bucket_boundaries,
params={
'data_files': train_data_file,
'meta_data_file': meta_data_filename
}))
eval_input_fn = create_input_data_fn(
mode=Modes.EVAL,
pipeline_config=PipelineConfig(module='TFRecordSequencePipeline',
dynamic_pad=True,
bucket_boundaries=bucket_boundaries,
params={
'data_files': eval_data_file,
'meta_data_file': meta_data_filename
}))
return train_input_fn, eval_input_fn
def create_image_dataset_input_fn(dataset_dir, prepare_fn,
record_file_name_format,
meta_data_file_name_format):
prepare_fn(dataset_dir)
train_data_file = record_file_name_format.format(dataset_dir, Modes.TRAIN)
eval_data_file = record_file_name_format.format(dataset_dir, Modes.EVAL)
meta_data_filename = meta_data_file_name_format.format(dataset_dir)
train_input_fn = create_input_data_fn(mode=Modes.TRAIN,
pipeline_config=PipelineConfig(
module='TFRecordImagePipeline',
dynamic_pad=False,
params={
'data_files':
train_data_file,
'meta_data_file':
meta_data_filename
}))
eval_input_fn = create_input_data_fn(mode=Modes.EVAL,
pipeline_config=PipelineConfig(
module='TFRecordImagePipeline',
dynamic_pad=False,
params={
'data_files':
eval_data_file,
'meta_data_file':
meta_data_filename
}))
return train_input_fn, eval_input_fn
def create_dataset_predict_input_fn(dataset_dir, prepare_fn, record_file_name_format,
meta_data_file_name_format):
prepare_fn(dataset_dir)
test_data_file = record_file_name_format.format(dataset_dir, Modes.PREDICT)
meta_data_filename = meta_data_file_name_format.format(dataset_dir)
test_input_fn = create_input_data_fn(
mode=Modes.PREDICT,
pipeline_config=PipelineConfig(module='TFRecordImagePipeline', dynamic_pad=False,
num_epochs=1,
params={'data_files': test_data_file,
'meta_data_file': meta_data_filename})
)
return test_input_fn
def create_sequence_dataset_predict_input_fn(dataset_dir, prepare_fn,
record_file_name_format,
meta_data_file_name_format,
bucket_boundaries):
prepare_fn(dataset_dir)
test_data_file = record_file_name_format.format(dataset_dir, Modes.PREDICT)
meta_data_filename = meta_data_file_name_format.format(dataset_dir)
test_input_fn = create_input_data_fn(
mode=Modes.PREDICT,
pipeline_config=PipelineConfig(module='TFRecordSequencePipeline',
dynamic_pad=True,
num_epochs=1,
batch_size=4,
min_after_dequeue=0,
bucket_boundaries=bucket_boundaries,
params={
'data_files': test_data_file,
'meta_data_file': meta_data_filename
}))
return test_input_fn