def _build_detected_glyphs()

in moonlight/glyphs/convolutional.py [0:0]

• 22 lines of code
• 10 McCabe index (conditional complexity)

  def _build_detected_glyphs(self, predictions_arr):
    """Takes the convolutional output ndarray and builds the individual glyphs.

    At each staff and y position, looks for short runs of the same detected
    glyph, and then outputs a single glyph at the x coordinate of the center of
    the run.

    Args:
      predictions_arr: A NumPy array with the result of `staffline_predictions`.
        Shape (num_staves, num_stafflines, width).

    Returns:
      A 2D array of the glyph coordinates. Shape (num_glyphs, 4) with columns
          corresponding to base.GlyphsTensorColumns.
    """
    glyphs = []
    num_staves, num_stafflines, width = predictions_arr.shape
    for staff in range(num_staves):
      for staffline in range(num_stafflines):
        y_position = num_stafflines // 2 - staffline
        run_start = -1
        run_value = musicscore_pb2.Glyph.NONE
        for x in range(width + 1):
          if x < width:
            value = predictions_arr[staff, staffline, x]
          if x == width or value != run_value:
            if run_value > musicscore_pb2.Glyph.NONE:
              # Process the current run if it is at least run_min_length pixels.
              if x - run_start >= self.run_min_length:
                glyph_center_x = (run_start + x) // 2
                glyphs.append(
                    self._create_glyph_arr(staff, y_position, glyph_center_x,
                                           run_value))
            run_value = value
            run_start = x
    # Convert to a 2D array.
    glyphs = np.asarray(glyphs, np.int32)
    return np.reshape(glyphs, (-1, 4))