tensorflow/models · error · ValueError

Length of class_weights should be {}

Error message

Length of class_weights should be {}

What it means

Error "Length of class_weights should be {}" thrown in tensorflow/models.

Source

Thrown at official/projects/panoptic/losses/panoptic_deeplab_losses.py:62

    # Assign pixel with ignore label to class 0 (background). The loss on the
    # pixel will later be masked out.
    labels = tf.where(valid_mask, labels, tf.zeros_like(labels))

    labels = tf.squeeze(tf.cast(labels, tf.int32), axis=3)
    valid_mask = tf.squeeze(tf.cast(valid_mask, tf.float32), axis=3)
    onehot_labels = tf.one_hot(labels, num_classes)
    onehot_labels = onehot_labels * (
        1 - self._label_smoothing) + self._label_smoothing / num_classes
    cross_entropy_loss = tf.nn.softmax_cross_entropy_with_logits(
        labels=onehot_labels, logits=logits)

    if not self._class_weights:
      class_weights = [1] * num_classes
    else:
      class_weights = self._class_weights

    if num_classes != len(class_weights):
      raise ValueError(
          'Length of class_weights should be {}'.format(num_classes))

    weight_mask = tf.einsum('...y,y->...',
                            tf.one_hot(labels, num_classes, dtype=tf.float32),
                            tf.constant(class_weights, tf.float32))
    valid_mask *= weight_mask

    if sample_weight is not None:
      valid_mask *= sample_weight

    cross_entropy_loss *= tf.cast(valid_mask, tf.float32)

    if self._top_k_percent_pixels >= 1.0:
      loss = tf.reduce_sum(cross_entropy_loss) / normalizer
    else:
      loss = self._compute_top_k_loss(cross_entropy_loss)
    return loss

View on GitHub (pinned to e006f5f0d5)

When it happens

Trigger: Thrown at official/projects/panoptic/losses/panoptic_deeplab_losses.py:62 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/8415cc6e576feaba. Report an issue: GitHub.