Frequently Asked Questions#

Answers to frequently asked questions about the cleanvision open-source package.

  1. What kind of machine learning tasks can I use CleanVision for?

CleanVision is independent of any machine learning tasks as it directly works on images and does not require and labels or metadata to detect issues in the dataset. The issues detected by CleanVision are helpful for all kinds of machine learning tasks.

  1. Can I check for specific issues in my dataset?

Yes, you can specify issues like light or blurry in the issue_types argument when calling find_issues()

imagelab.find_issues(issue_types={"light": {}, "blurry": {}})
  1. What dataset formats does CleanVision support?

Apart from plain image files stored locally or in the cloud, CleanVision also works with HuggingFace and Torchvision datasets. You can use the dataset objects as is with the image_key argument.

imagelab = Imagelab(hf_dataset=dataset, image_key="image")

For more detailed usage instructions and examples, check the Tutorials.

Commonly encountered errors#

  • RuntimeError: An attempt has been made to start a new process before the current process has finished its bootstrapping phase.

This probably means that you are not using fork to start your
child processes and you have forgotten to use the proper idiom
in the main module:

    if __name__ == '__main__':

The "freeze_support()" line can be omitted if the program
is not going to be frozen to produce an executable.

To fix this issue, refer to the "Safe importing of main module"
section in

The above issue is caused by multiprocessing module working differently for macOS and Windows platforms. A detailed discussion of the issue can be found here. A fix around this issue is to run CleanVision in the main namespace like this

if __name__ == "__main__":

    imagelab = Imagelab(data_path)

OR use n_jobs=1 to disable parallel processing: