Method __init__

training/dataset.py:319–354

(self, cfg, name)

class PreprocessedDataset(Dataset):
  def __init__(self, cfg, name):
    super(PreprocessedDataset, self).__init__()

    if not name:
      self.samples = []
      self.num_images = 0
      return

    # Check whether the preprocessed images have all required features
    data_dir = get_preproc_data_dir(cfg, name)
    data_cfg = load_config(data_dir)

    self.tile_size = cfg.tile_size

    # Get the features
    self.features = cfg.features
    self.main_feature = get_main_feature(cfg.features)
    self.aux_features = get_aux_features(cfg.features)
    self.clean_aux = cfg.clean_aux and self.aux_features

    # Get the channels
    self.channels = get_dataset_channels(cfg.features)
    self.all_channels = get_dataset_channels(data_cfg.features)
    self.num_main_channels = len(get_model_channels(self.main_feature))

    # Get the image samples
    samples_filename = os.path.join(data_dir, 'samples.json')
    self.samples = load_json(samples_filename)
    self.num_images = len(self.samples)

    if self.num_images == 0:
      return

    # Create the memory mapping based image reader
    tza_filename = os.path.join(data_dir, 'images.tza')
    self.images = tza.Reader(tza_filename)

## -----------------------------------------------------------------------------
## Training dataset