tensorflow/models · error · ValueError

Class directory is empty

Error message

Class directory is empty: {class_path}

What it means

Raised by compute_class_weights when a class subdirectory exists but contains no regular files. Class weights are computed from per-class file counts; a zero count would cause a division by zero when normalizing, so the guard rejects it. Usually means a partially copied/extracted dataset, an empty class folder, or files nested in another subdirectory level that the per-file iterdir count misses.

Solutions

1. Ensure every class folder contains at least one image file directly (not nested in further subdirectories)
2. Re-download or re-extract the dataset if some class folders were left empty
3. Flatten nested structure so images sit directly under their class folder
4. Audit the dataset tree with a quick per-class file count before training

When it happens

Trigger: Thrown at official/projects/waste_identification_ml/fine_tuning/Dinov3_image_classifier/datasets.py:237 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/e35c537282de232d. Report an issue: GitHub.

Appendix: source

Thrown at official/projects/waste_identification_ml/fine_tuning/Dinov3_image_classifier/datasets.py:237

      ValueError: If `train_directory` contains no class subdirectories or if
        any class subdirectory contains no files.
  """
  class_names = sorted(
      entry.name for entry in train_directory.iterdir() if entry.is_dir()
  )
  if not class_names:
    raise ValueError(
        f"No class subdirectories found in train_directory: {train_directory}"
    )

  class_counts = []
  for class_name in class_names:
    class_path = train_directory / class_name
    number_of_files = sum(
        1 for entry in class_path.iterdir() if entry.is_file()
    )
    if number_of_files == 0:
      raise ValueError(f"Class directory is empty: {class_path}")
    class_counts.append(number_of_files)

  number_of_classes = len(class_counts)
  total_samples = sum(class_counts)
  weights = [
      total_samples / (number_of_classes * count) for count in class_counts
  ]

  _LOGGER.info("Class counts: %s", dict(zip(class_names, class_counts)))
  _LOGGER.info("Class weights: %s", dict(zip(class_names, weights)))

  return torch.tensor(weights, dtype=torch.float32)

View on GitHub (pinned to e006f5f0d5)