Hello,
Yes, it is possible to connect multiple ML backends to a single project in Label Studio. This allows you to utilize different models for various tasks within the same project, such as using EasyOCR for autolabeling and Tesseract for interactive labeling.
Here's how you can set up multiple ML backends in your project:

• Connect the First ML Backend (EasyOCR):

  • Go to your project in Label Studio.
  • Navigate to Settings > Machine Learning.
  • Click on Connect Model.
  • Enter the URL and details of your EasyOCR backend.
  • Click Validate and Save.
  • Ensure that the "Show predictions to annotators" option is enabled if you want to use it for autolabeling.

• Connect the Second ML Backend (Tesseract):

  • Still in Settings > Machine Learning, click on Connect Model again.
  • Enter the URL and details of your Tesseract backend.
  • Click Validate and Save.
  • For interactive labeling, make sure to enable the "Use for interactive preannotations" option.

• Verify Both Backends are Connected:

  • You should now see both the EasyOCR and Tesseract backends listed in the Machine Learning section of your project settings.
    Using Multiple ML Backends:

• When labeling tasks, you can select which ML backend's predictions to use.

• If both backends provide predictions for the same task, you can choose between them using the Model Version dropdown menu in the labeling interface.

• For interactive labeling with Tesseract, ensure that Auto-Annotation is enabled in the labeling interface.
  Note:

• Ensure that both ML backends are running and accessible to your Label Studio instance.

• If you encounter issues adding multiple ML backends via the UI, you can also start Label Studio from the command line with multiple backends using the --ml-backend option:

  label-studio start my_project --ml-backend http://localhost:9090 http://localhost:9091

Replace http://localhost:9090 and http://localhost:9091 with the URLs of your EasyOCR and Tesseract backends, respectively.

Thanks,

Tyler Conlee
Head of Support
HumanSignal

Comment by Tyler Conlee
Workflow Run