def add_volume_iou_metrics(inputs, outputs):
  """Computes the per-instance volume IOU.

  Args:
    inputs: Input dictionary of the voxel generation model.
    outputs: Output dictionary returned by the voxel generation model.

  Returns:
    names_to_values: metrics->values (dict).
    names_to_updates: metrics->ops (dict).

  """
  names_to_values = dict()
  names_to_updates = dict()
  print("*************************")
  print(inputs)
  print(outputs)
  labels = tf.greater_equal(inputs, 0.5)
  predictions = tf.greater_equal(outputs, 0.5)
  labels = (2 - tf.to_int32(labels))-1 #normalisieren auf 1,2 (we think 2 is empty/free), 1 is ground truth occupied
  predictions = (3 - tf.to_int32(predictions) * 2)-1 # 3=predicted free, 1=predicted occupied,
  #so we get intersection only for number 1, predicted occupied
  #labels= tf.Print(labels, [labels], message="This is the labels: ", summarize=100)
  #labels= tf.Print(labels, [predictions], message="This is the predictions: ",summarize=100)
  tmp_values, tmp_updates = tf.metrics.mean_iou(
      labels=labels,# we add -1 because the tf metrics expects it probably to start from zero
      predictions=predictions,
      num_classes=3)
  tmp_values=tf.Print(tmp_values,[tmp_values])
  names_to_values['volume_iou'] = tmp_values * 3.0
  names_to_updates['volume_iou'] = tmp_updates
  return names_to_values, names_to_updates