tensorflow/models · error · ValueError

Variables to load is empty.

Error message

Variables to load is empty.

What it means

Sanity check inside the checkpoint-restore scaffold function: after building the assignment map from the model variables and the checkpoint's var_to_shape_map, no variables remained to load. This means none of the model's variable names matched the checkpoint entries under the given prefix/skip regex.

Solutions

1. Verify the checkpoint path points to a compatible pretrained checkpoint
2. Check that the prefix matches the variable naming in the checkpoint (e.g. the backbone prefix)
3. Relax or remove skip_variables_regex if it is excluding all variables
4. Confirm the model was built before restore is attempted so variables exist

When it happens

Trigger: Thrown at official/legacy/detection/modeling/checkpoint_utils.py:139 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/32e28355153e383b. Report an issue: GitHub.

Appendix: source

Thrown at official/legacy/detection/modeling/checkpoint_utils.py:139

  """

  def _restore_checkpoint_fn(keras_model):
    """Loads pretrained model through scaffold function."""
    if not checkpoint_path:
      logging.info('checkpoint_path is empty')
      return
    var_prefix = prefix
    if prefix and not prefix.endswith('/'):
      var_prefix += '/'
    var_to_shape_map = _get_checkpoint_map(checkpoint_path)
    assert var_to_shape_map, 'var_to_shape_map should not be empty'
    vars_to_load = _build_assignment_map(
        keras_model,
        prefix=var_prefix,
        skip_variables_regex=skip_regex,
        var_to_shape_map=var_to_shape_map)
    if not vars_to_load:
      raise ValueError('Variables to load is empty.')
    tf.compat.v1.train.init_from_checkpoint(checkpoint_path, vars_to_load)

  return _restore_checkpoint_fn

View on GitHub (pinned to e006f5f0d5)