Method __init__

training/dataset.py:469–509

(self, cfg, name)

class ValidationDataset(PreprocessedDataset):
  def __init__(self, cfg, name):
    super(ValidationDataset, self).__init__(cfg, name)
    self.tiles = []

    if self.num_images == 0:
      return

    input_channel_indices = get_channel_indices(self.channels, self.all_channels)

    # Split the images into tiles
    for sample_index in range(self.num_images):
      # Get the input image
      input_name,  _ = self.samples[sample_index]
      input_image, _ = self.images[input_name]

      # Get the size of the image
      height = input_image.shape[0]
      width  = input_image.shape[1]
      if height < self.tile_size or width < self.tile_size:
        error('image is smaller than the tile size')

      # Compute the number of tiles
      num_tiles_y = height // self.tile_size
      num_tiles_x = width  // self.tile_size

      # Compute the start offset for centering
      start_y = (height % self.tile_size) // 2
      start_x = (width  % self.tile_size) // 2

      # Add the tiles
      for y in range(num_tiles_y):
        for x in range(num_tiles_x):
          oy = start_y + y * self.tile_size
          ox = start_x + x * self.tile_size

          if self.main_feature == 'sh1':
            for k in range(0, 9, 3):
              ch = input_channel_indices[k:k+3] + input_channel_indices[9:]
              self.tiles.append((sample_index, oy, ox, ch))
          else:
            self.tiles.append((sample_index, oy, ox, input_channel_indices))

  def __len__(self):
    return len(self.tiles)