tensorflow/models · error · ValueError

share_classification_heads cannot be set as True when att_pr

Error message

share_classification_heads cannot be set as True when att_prediction_tower_name is specified.

What it means

Error "share_classification_heads cannot be set as True when att_prediction_tower_name is specified." thrown in tensorflow/models.

Source

Thrown at official/vision/modeling/heads/dense_prediction_heads.py:255

        )
      elif att_type == 'classification':
        att_predictor_kwargs.update(
            {
                'bias_initializer': tf.constant_initializer(
                    -np.log((1 - 0.01) / 0.01)
                )
            }
        )
      else:
        raise ValueError(
            'Attribute head type {} not supported.'.format(att_type)
        )

      if (
          att_prediction_tower_name
          and self._config_dict['share_classification_heads']
      ):
        raise ValueError(
            'share_classification_heads cannot be set as True when'
            ' att_prediction_tower_name is specified.'
        )

      if not self._config_dict['use_separable_conv']:
        att_predictor_kwargs.update({
            'kernel_initializer': tf_keras.initializers.RandomNormal(
                stddev=1e-5
            ),
            'kernel_regularizer': self._config_dict['kernel_regularizer'],
        })
      self._attribute_kwargs.append(att_predictor_kwargs)

  def _apply_prediction_tower(self, features, convs, norms) -> tf.Tensor:
    x = features
    for conv, norm in zip(convs, norms):
      x = conv(x)
      x = norm(x)

View on GitHub (pinned to e006f5f0d5)

Solutions

  1. Set share_classification_heads to False when att_prediction_tower_name is specified.
  2. Remove att_prediction_tower_name if you want to share classification heads across attributes.

When it happens

Trigger: Thrown at official/vision/modeling/heads/dense_prediction_heads.py:255 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/0cf631a00904ee14. Report an issue: GitHub.