March 2, 2023

Release notes


  • We now offer a data row proxy for our more security-conscious customers. This feature hides the URL of the data row and instead shows a URL of where the real data row URL can only be accessed with a valid cookie. If you would like access to this feature, please contact [email protected].
  • Now you can view nested classifications in the annotation overlays, allowing you to easily view the classifications associated with each object.
  • Workflows are now compatible with the custom editor. To enable this compatibility, the Accept and Reject buttons have been moved from the editor interface to the data row browser. This small change is only enabled for projects that have a custom editor configured.
  • You can now filter on metadata in the Model tab to analyze model metrics based on metadata (e.g., how well does the model perform on camera 1 vs camera 2). For example, you can filter by metadata to surface problematic data rows, save the filter, and send those problematic data rows for relabeling. To learn more, visit this page.
  • The video editor now has AI-based bounding box tracking. This automation tool is designed to speed up your video labeling. The default number of automatically labeled frames is 10, however, you can adjust this number. To learn more, visit this page.
  • If you are a brand new free tier customer, you will see a Catalog start page that provides resources on how to upload data into Labelbox. It displays documentation on cloud integrations, data connectors, and direct file upload.
  • The buttons for adjusting playback speed and timeline zoom are now more prominent in the video editor. These features are designed to make it easier to navigate long videos.
  • You can save search queries in Model as a slice. This means that you no longer have to re-create the same search queries over and over again. Slices are dynamic and are automatically shared across model runs. To learn more, visit this page.
  • Labelbox now automatically generates some slices for you in Model. Examples of auto-generated slices include false positives and low-confidence predictions. To learn more, visit this page.
  • You can now add and remove multiple keyframes at a time for object annotations via a UI button or the hotkey k. This feature does not yet support classifications. This feature is helpful for labeling long videos. Click here to watch a demo.
  • In the video editor, you now have a way to jump to a specific frame in one hop. In other words, you can now jump to every nth frame, in both directions (e.g., jump to the 15th or 37th frame) using hotkeys or a UI button. Click here to watch a demo.
  • The detailed view in Model and Catalog now displays nested classifications and confidence scores.
  • Visitors can now access a read-only version of the Labelbox product, without the need to create an account. Visit to try it out.

Python SDK

The latest version of our Python SDK is v3.39.0. See our full changelog in Github for more details on what was added recently.

Version 3.39.0 (2023-02-28)


  • New method Project.task_queues() to obtain the task queues for a project.
  • New method Project.move_data_rows_to_task_queue() for moving data rows to a specified task queue.
  • Added more descriptive error messages for metadata operations
  • Added Task.errors_url for async tasks that return errors as separate file (e.g. export_v2)


  • Updated ProjectExportParams.labels to ProjectExportParams.label_details
  • Removed media_attributes from DataRowParams
  • Added deprecation warnings for LabelList and removed its usage
  • Removed unused arguments in Project.export_v2 and ModelRun.export_v2


  • Fixed examples/label_export/images.ipynb notebook metadata
  • Removed unused lb_serializer imports
  • Removed uuid generation in NDJson annotation payloads, as it is now optional
  • Removed custom embeddings usage in examples/basics/data_row_metadata.ipynb
  • New notebook examples/basics/custom_embeddings.ipynb for custom embeddings
  • Updated examples/annotation_import/text.ipynb to use TextData and specify Text media type

Version 3.38.0 (2023-02-15)


  • All imports are available via import labelbox as lb and import labelbox.types as lb_types.
  • Attachment_name support to create_attachment()


  • LabelImport.create_from_objects(), MALPredictionImport.create_from_objects(), MEAPredictionImport.create_from_objects(), Project.upload_annotations(), ModelRun.add_predictions() now support Python Types for annotations.


  • Removed NDJsonConverter from example notebooks
  • Simplified imports in all notebooks
  • Fixed nested classification in examples/annotation_import/image.ipynb
  • Ontology (instructions --> name)

Version 3.37.0 (2023-02-08)


  • New last_activity_start param to project.export_labels() for filtering which labels are exported. See docstring for more on how this works.


  • Rename Classification.instructions to


  • Retry connection timeouts