On-prem changelog



  • MySQL bin logs were filling up the Persistent Volume. We set the expiration period to 2 days from the 30-day default. This would have eventually caused issues for customers.
  • Model Assisted Labeling and Label imports weren’t working due to private CA certificate issues. We altered the code to allow this and retest the entire product with a private CA to verify private CAs work with Labelbox.



  • Catevent allowed for maximum RabbitMQ queue streams larger than the Persistent Volume allocated for RabbitMQ. This would have eventually caused issues for customers.
  • Prediction import changes to allow for private certificates were missed in the last hotfix.


  • Added support for GCP GKE deployments
  • Added support for Azure AKS deployments



  • Default pod requests were reduced to allow smaller instance sizes
  • Previous changes to allow customers to use private CA certs were re-added back.


  • Added support for Openshift 4.10.X
  • Keycloak client secret can be changed in the Replicated console config screen.


To view the on-prem specific docs for on-prem v3.2.0, click here.



  • You now have the ability to see thumbnails and previews for our newer geospatial formats of NITF and COGs. This will make it easier for you to find the most relevant Data Rows to label.
  • In the Geospatial editor, you can now "freehand" draw Polygon shapes and use the pen/erase tool to create Polygon shapes.


  • You can now apply global classifications to an entire video clip in the Video editor.
  • The segmentation tool is now available in the Video editor
  • It is now possible to use Model-assisted labeling to upload video frame classifications where the answer changes throughout the frame sequence.
  • When a labeler identifies a duplicate instance of an object class in the video editor, they can use the Merge feature to combine them as long as there are no conflicting frames. Objects included are bounding boxes, polylines, points, and segmentation masks.


  • You can now create relationships between annotations in the Image editor. For more information, see Image annotation relationships.
  • You can use Segmentation auto-labeling tool to draw a box around the object you wish to label and Labelbox will auto-generate a segmentation mask for you. This allows you to speed up you labeling efforts.
  • Our Image editor can now support up to 1,000 annotations per Data Row. Now you can create significantly more annotations in the image editor with no lag in the UI.


  • The Conversational text editor is now in open beta and available for all customers in pro, trial, enterprise, and education tiers. For more information, see Conversational text (beta)


  • With the new Document editor, you can add PDF data to a project and create bounding boxes around key characters, concepts, or page sections for Optical Character Recognition (OCR).
  • Issues & comments is now available in the Document editor.
  • Now you can add relationships between your annotations in a PDF document. See our docs on image annotation relationships to learn more.
  • PDF thumbnails now appear in Catalog.


  • We launched a new and improved version of our Catalog. See our Catalog docs to learn more.
  • In Catalog, you can now see much richer information on top of your raw data in Catalog. Labelbox displays ground truths and predictions over the assets, in the thumbnail view and in the detailed view. This makes data exploration easier and more powerful in Catalog.
  • With our new Batches feature, you can select individual Data Rows from Catalog and send them to a Project for labeling, rather than having to attach an entirely new dataset.
  • You can now quickly import your Data Rows and attach your metadata with one or two lines of code. Then, you can filter based on metadata in Catalog. See our docs on importing metadata for more details.


  • Our new project creation flow helps you set up your projects faster.
  • User groups is now available. You can use this feature to manage large numbers of users that need to be added to the same set of projects.
  • You can now set filters to export subsets of your annotations rather than exporting the annotations from the entire project. Currently, only createdAt is supported.
  • A new project overview has been launched.
  • Our Data Rows tab got some new and improved search capabilities. These updates make the search experience in the Data Rows tab more intuitive and much faster.


  • We revamped our Python SDK reference section of the docs to make them more comprehensive. It is now much easier for developers to find SDK-related information.


  • The Superpixel tool has been removed from our platform.
  • Dropdown annotation type has been removed from our platform. Use Radio, Checklist, or Free-form text instead.

Python SDK

The Python SDK version for this release is v3.25.2. See our changelog in Github to learn more.


To view the on-prem specific docs for v3.1.0 and v3.1.1, click here.

  • Customers had issues upgrading to 3.1.0. The issue was due to 3 Replicated defects that have been addressed:
    • Airgapped migration from docker to containerd fails
    • Airgapped Kubernetes upgrade fails with image pull errors from kubeadm
    • Docker to containerd migration fails and reverts back to docker if script fails after migration but before removal of docker
  • The “Un-reviewed”, “Accepted”, and “Declined” data rows statistics links now work as expected.
  • Video editor now shows the label object descriptors.


  • Annotation relationships
  • Video polygons


The Python SDK version for this on-prem release is v3.15.0. View our changelog in Github to learn more.