July 5, 2023

Release notes


  • Auto-segment 2.0, powered by Meta's Segment Anything model, helps you draw draw segmentation masks much faster in the image editor. To use this new feature, you can use:
    • Hover-and-click: When you hover over your image in the editor, Labelbox highlights the object underneath the cursor. Select the highlighted object to create a segmentation mask.
    • Box mode: Draw a bounding box over an object in the image to automatically create a segmentation mask over the object.
  • You can now use a brush tool to draw segmentation masks in the image editor. You can adjust the size and shape of the brush.
  • Our new LLM data generation editor lets you prepare a dataset of prompts and responses to fine-tune large language models (LLMs).
  • The Performance dashboard has been improved to be more reliable and offers more metrics.
    Note: Performance dashboard updates are being rolled out gradually. If you do not see the changes at this time, check again in a day or two.

Python SDK

The latest version of our Python SDK is v3.49.0. See our full changelog in Github for more details on what was added recently.

Version 3.49.0 (2023-06-27)


  • Improved batch creation logic when more than 1000 global keys provided


  • Added example on how to access mark in export v2
  • Removed NDJSON library from examples/basics/custom_embeddings.ipynb
  • Removed queue_mode property from create_project() method call

Version 3.48.0 (2023-06-13)


  • Support for ISO format to exports V2 date filters
  • Support to specify confidence for all free-text annotations


  • Removed backports library and replaced it with python dateutil package to parse iso strings


  • Added predictions to model run example
  • Added notebook to run yolov8 and sam on video and upload to LB
  • Updated google colab notebooks to reflect raster segmentation tool being released on 6/13
  • Updated radio NDJSON annotations format to support confidence
  • Added confidence to all free-text annotations (ndjson)
  • Fixed issues with cv2 library rooting from the Geospatial notebook used a png map with a signed URL with an expired token