tensorflow/models · error · ValueError

A stage merge style in MOSAIC Decoder can only be concat_mer

Error message

A stage merge style in MOSAIC Decoder can only be concat_merge or sum_merge.

What it means

Error "A stage merge style in MOSAIC Decoder can only be concat_merge or sum_merge." thrown in tensorflow/models.

Source

Thrown at official/projects/mosaic/modeling/mosaic_head.py:142

            activation=activation,
            kernel_initializer=kernel_initializer,
            kernel_regularizer=kernel_regularizer,
            interpolation=interpolation)
        self._merge_stages.append(concat_merge_stage)
      elif stage_merge_style == 'sum_merge':
        sum_merge_stage = mosaic_blocks.DecoderSumMergeBlock(
            decoder_projected_depth=decoder_projected_filter,
            output_size=(0, 0),
            use_sync_bn=use_sync_bn,
            batchnorm_momentum=batchnorm_momentum,
            batchnorm_epsilon=batchnorm_epsilon,
            activation=activation,
            kernel_initializer=kernel_initializer,
            kernel_regularizer=kernel_regularizer,
            interpolation=interpolation)
        self._merge_stages.append(sum_merge_stage)
      else:
        raise ValueError(
            'A stage merge style in MOSAIC Decoder can only be concat_merge '
            'or sum_merge.')

    # Concat merge or sum merge does not require an additional classifer layer
    # unless the final decoder projected filter does not match num_classes.
    final_decoder_projected_filter = decoder_projected_filters[-1]
    if (final_decoder_projected_filter != num_classes and
        not use_additional_classifier_layer):
      raise ValueError('Additional classifier layer is needed if final decoder '
                       'projected filters does not match num_classes!')
    self._use_additional_classifier_layer = use_additional_classifier_layer
    if use_additional_classifier_layer:
      # This additional classification layer uses different kernel
      # initializers and bias compared to earlier blocks.
      self._pixelwise_classifier = tf_keras.layers.Conv2D(
          name='pixelwise_classifier',
          filters=num_classes,
          kernel_size=classifier_kernel_size,

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/projects/mosaic/modeling/mosaic_head.py:142 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/b4044e7db8178b5c. Report an issue: GitHub.