tensorflow/models · error · Exception

Unsupported datatype used in ground truth builder only {floa

Error message

Unsupported datatype used in ground truth builder only {float16, bfloat16, or float32}

What it means

Error "Unsupported datatype used in ground truth builder only {float16, bfloat16, or float32}" thrown in tensorflow/models.

Source

Thrown at official/projects/centernet/ops/target_assigner.py:295

        other entires are 0
      'box_mask': `Tensor` of shape [max_num_instances], where the first
        num_boxes entries are 1. All other entires are 0
      'box_indices': `Tensor` of shape [max_num_instances, 2], where the first
        num_boxes entries contain the y-center and x-center of a valid box.
        These are used to extract the regressed box features from the
        prediction when computing the loss

  Raises:
    Exception: if datatype is not supported.
  """
  if dtype == 'float16':
    dtype = tf.float16
  elif dtype == 'bfloat16':
    dtype = tf.bfloat16
  elif dtype == 'float32':
    dtype = tf.float32
  else:
    raise Exception(
        'Unsupported datatype used in ground truth builder only '
        '{float16, bfloat16, or float32}')

  # Get relevant bounding box and class information from labels
  # only keep the first num_objects boxes and classes
  num_objects = labels['groundtruths']['num_detections']
  # shape of labels['boxes'] is [max_num_instances, 4]
  # [ymin, xmin, ymax, xmax]
  boxes = tf.cast(labels['boxes'], dtype)
  # shape of labels['classes'] is [max_num_instances, ]
  classes = tf.cast(labels['classes'] - class_offset, dtype)

  # Compute scaling factors for center/corner positions on heatmap
  # input_size = tf.cast(input_size, dtype)
  # output_size = tf.cast(output_size, dtype)
  input_h, input_w = input_size[0], input_size[1]
  output_h, output_w = output_size[0], output_size[1]

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/projects/centernet/ops/target_assigner.py:295 when the library encounters an invalid state.

Common situations: See trigger scenarios.

AI-assisted analysis of tensorflow/models@e006f5f0d5 (2026-08-24). Data as JSON: /api/errors/f91119ec17802df0. Report an issue: GitHub.