Python util.GetFilesRecursively方法代码示例

# 需要导入模块: from utils import util [as 别名]
# 或者: from utils.util import GetFilesRecursively [as 别名]
def main(_):
  # Parse config dict from yaml config files / command line flags.
  config = util.ParseConfigsToLuaTable(FLAGS.config_paths, FLAGS.model_params)
  num_views = config.data.num_views

  validation_records = util.GetFilesRecursively(config.data.validation)
  batch_size = config.data.batch_size

  checkpointdir = FLAGS.checkpointdir

  # If evaluating a specific checkpoint, do that.
  if FLAGS.checkpoint_iter:
    checkpoint_path = os.path.join(
        '%s/model.ckpt-%s' % (checkpointdir, FLAGS.checkpoint_iter))
    evaluate_once(
        config, checkpointdir, validation_records, checkpoint_path, batch_size,
        num_views)
  else:
    for checkpoint_path in tf.contrib.training.checkpoints_iterator(
        checkpointdir):
      evaluate_once(
          config, checkpointdir, validation_records, checkpoint_path,
          batch_size, num_views) 

# 需要导入模块: from utils import util [as 别名]
# 或者: from utils.util import GetFilesRecursively [as 别名]
def evaluate(self):
    """Runs `Estimator` validation.
    """
    config = self._config

    # Get a list of validation tfrecords.
    validation_dir = config.data.validation
    validation_records = util.GetFilesRecursively(validation_dir)

    # Define batch size.
    self._batch_size = config.data.batch_size

    # Create a subclass-defined training input function.
    validation_input_fn = self.construct_input_fn(
        validation_records, False)

    # Create the estimator.
    estimator = self._build_estimator(is_training=False)

    # Run validation.
    eval_batch_size = config.data.batch_size
    num_eval_samples = config.val.num_eval_samples
    num_eval_batches = int(num_eval_samples / eval_batch_size)
    estimator.evaluate(input_fn=validation_input_fn, steps=num_eval_batches) 

# 需要导入模块: from utils import util [as 别名]
# 或者: from utils.util import GetFilesRecursively [as 别名]
def get_labeled_tables(config):
  """Gets either labeled test or validation tables, based on flags."""
  # Get a list of filenames corresponding to labeled data.
  mode = FLAGS.mode
  if mode == 'validation':
    labeled_tables = util.GetFilesRecursively(config.data.labeled.validation)
  elif mode == 'test':
    labeled_tables = util.GetFilesRecursively(config.data.labeled.test)
  else:
    raise ValueError('Unknown dataset: %s' % mode)
  return labeled_tables 

# 需要导入模块: from utils import util [as 别名]
# 或者: from utils.util import GetFilesRecursively [as 别名]
def main(_):
  # Parse config dict from yaml config files / command line flags.
  config = util.ParseConfigsToLuaTable(FLAGS.config_paths, FLAGS.model_params)

  # Get tables to embed.
  query_records_dir = FLAGS.query_records_dir
  query_records = util.GetFilesRecursively(query_records_dir)

  target_records_dir = FLAGS.target_records_dir
  target_records = util.GetFilesRecursively(target_records_dir)

  height = config.data.raw_height
  width = config.data.raw_width
  mode = FLAGS.mode
  if mode == 'multi':
    # Generate videos where target set is composed of multiple videos.
    MultiImitationVideos(query_records, target_records, config,
                         height, width)
  elif mode == 'single':
    # Generate videos where target set is a single video.
    SingleImitationVideos(query_records, target_records, config,
                          height, width)
  elif mode == 'same':
    # Generate videos where target set is the same as query, but diff view.
    SameSequenceVideos(query_records, config, height, width)
  else:
    raise ValueError('Unknown mode %s' % mode) 

# 需要导入模块: from utils import util [as 别名]
# 或者: from utils.util import GetFilesRecursively [as 别名]
def train(self):
    """Runs training."""
    # Get a list of training tfrecords.
    config = self._config
    training_dir = config.data.training
    training_records = util.GetFilesRecursively(training_dir)

    # Define batch size.
    self._batch_size = config.data.batch_size

    # Create a subclass-defined training input function.
    train_input_fn = self.construct_input_fn(
        training_records, is_training=True)

    # Create the estimator.
    estimator = self._build_estimator(is_training=True)

    train_hooks = None
    if config.use_tpu:
      # TPU training initializes pretrained weights using a custom train hook.
      train_hooks = []
      if tf.train.latest_checkpoint(self._logdir) is None:
        train_hooks.append(
            InitFromPretrainedCheckpointHook(
                config[config.embedder_strategy].pretrained_checkpoint))

    # Run training.
    estimator.train(input_fn=train_input_fn, hooks=train_hooks,
                    steps=config.learning.max_step)