open-mmlab/mmdetection · error · ImportError

Please run "pip install instaboostfast" to install instaboos

Error message

Please run "pip install instaboostfast" to install instaboostfast first.

What it means

The transform() method of InstaBoost lazily re-imports instaboostfast at augmentation time (when the random aug_ratio check passes). If the package was uninstalled or became unimportable after init (e.g. broken env), the ImportError surfaces mid-training instead of at pipeline construction.

Source

Thrown at mmdet/datasets/transforms/instaboost.py:138

        instances.extend(ignore_anns)
        results['img'] = img
        results['instances'] = instances
        return results

    def transform(self, results) -> dict:
        """The transform function."""
        img = results['img']
        ori_type = img.dtype
        if 'instances' not in results or len(results['instances']) == 0:
            return results

        anns, ignore_anns = self._load_anns(results)
        if np.random.choice([0, 1], p=[1 - self.aug_ratio, self.aug_ratio]):
            try:
                import instaboostfast as instaboost
            except ImportError:
                raise ImportError('Please run "pip install instaboostfast" '
                                  'to install instaboostfast first.')
            anns, img = instaboost.get_new_data(
                anns, img.astype(np.uint8), self.cfg, background=None)

        results = self._parse_anns(results, anns, ignore_anns,
                                   img.astype(ori_type))
        return results

    def __repr__(self) -> str:
        repr_str = self.__class__.__name__
        repr_str += f'(aug_ratio={self.aug_ratio})'
        return repr_str

View on GitHub (pinned to cfd5d3a985)

Solutions

1. pip install instaboostfast and restart training so all dataloader workers import it cleanly
2. Verify the import in the same interpreter used for training: python -c "import instaboostfast"
3. Disable or set aug_prob/aug_ratio to 0 in the InstaBoost config entry to skip the code path

import importlib.util
assert importlib.util.find_spec('instaboostfast') is not None, 'instaboostfast missing in worker env'

try:
    out = instaboost_tf.transform(results)
except ImportError as e:
    # fall back to unaugmented results
    logger.warning(f'skipping instaboost: {e}')
    return results

When it happens

Trigger: Calling InstaBoost.transform() where np.random.choice selects augmentation (aug_ratio > 0) and `import instaboostfast` fails at runtime, typically because the environment changed between __init__ and training, or the object was constructed with the import succeeding but later runs in a different process/env.

Common situations: Training resumed from a pickle/copy of the pipeline in a worker process without the package; environment switched via SLURM/worker launch; partial installs where the .so import fails intermittently.

Related errors

• Please run "pip install instaboostfast" to install instaboos
• Invalid crop_type {crop_type}.
• LoadImageFromFile is not found in the test pipeline
• Visualization needs the "visualizer" termdefined in the conf
• Unsupported input type: {type(single_input)}

AI-assisted analysis of open-mmlab/mmdetection@cfd5d3a985 (2026-08-27). Data as JSON: /api/errors/c810d07c4b273229. Report an issue: GitHub.