Subscribe to release notesStatus pageCommunity
Back to All

January 5, 2023

over 1 year ago by Alex

Release notes

App

  • You can now use Model-assisted labeling to import entity (NER) predictions to PDF assets in Labelbox. For instructions, visit this end-to-end demo notebook here.
  • You can now click and drag keyframes containing classifications in the video editor. This capability allows you to easily move keyframes around on the timeline to adjust the size of classification segments.
  • Our document editor is now GA. This GA release includes the ability to add custom text layers to your PDFs. Visit this guide to learn about the improvements we made to our document editor.
  • We added a new organization-level role called “data admin”. The data admin has full access to the complete set of Catalog functionalities (e.g., upload data, create new datasets, create slices), but does NOT have access to any of the annotation projects. To learn more about the permissions for the data admin role, visit our docs.
  • We added a new organization-level role called “read-only admin”. The read-only admin can access all parts of the platform but in a read-only manner. To learn more about the permissions for the read-only admin role, visit our docs.
  • The cuboid tool in the image editor is now available to all customers. You can use this tool to create training data for 3D models or models that require some knowledge of depth on top of images. To learn more, visit our cuboid docs.
  • All customers now have access to the new analytics tools in the Model product. These new model analytics tools enable you to find the distribution of annotations and predictions in every model run via histograms. To learn more, visit our docs on model analytics.
  • All customers now have access to auto-generated metrics in the Model product. Model auto metrics enable you to:
    • Upload your model predictions and ground truths to a model run. Labelbox will then automatically generate the following metrics for that model run: precision, recall, F1-score, TP/TN/FP/FN, and a confusion matrix.
    • Click on any cell of the confusion matrix to find a specific type of misprediction.
    • Upload confidence scores on every prediction. This allows you to filter and sort data based on model confidence, not just on model metrics.
    • Change the confidence and IOU thresholds in the user interface.
    • To learn more, visit our docs on auto metrics.
  • We improved our segmentation masks in the Model product by making it more obvious where mask predictions and ground truths agree or disagree.

Python SDK

The latest version of our Python SDK is v3.34.0. See our full changelog in Github for more details on what was added recently.

Version 3.34.0 (2022-12-22)

Added

  • Added get_by_name() method to MetadataOntology object to access both custom and reserved metadata by name.
  • Added support for adding metadata by name when creating datarows using DataRowMetadataOntology.bulk_upsert().
  • Added support for adding metadata by name when creating datarows using Dataset.create_data_rows(), Dataset.create_data_rows_sync(), and Dataset.create_data_row().
  • Example notebooks for auto metrics in models

Changed

  • Dataset.create_data_rows() max limit of DataRows increased to 150,000
  • Improved error handling for invalid annotation import content
  • String metadata can now be 1024 characters long (from 500)

Fixed

  • Broken urls in detectron notebook

Version 3.33.1 (2022-12-14)

Fixed

  • Fixed where batch creation limit was still limiting # of data rows. SDK should now support creating batches with up to 100k data rows

Version 3.33.0 (2022-12-13)

Added

  • Added SDK support for creating batches with up to 100k data rows
  • Added optional media_type to client.create_ontology_from_feature_schemas() and client.create_ontology()

Changed

  • String representation of DbObject subclasses are now formatted

Version 3.32.0 (2022-12-02)

Added

  • Added HTML Enum to MediaType. HTML is introduced as a new asset type in Labelbox.
  • Added PaginatedCollection.get_one() and PaginatedCollection.get_many() to provide easy functions to fetch single and bulk instances of data for any function returning a PaginatedCollection. E.g. data_rows = dataset.data_rows().get_many(10)
  • Added a validator under ScalarMetric to validate metric names against reserved metric names

Changed

  • In iou.miou_metric() and iou.feature_miou_metric, iou metric renamed as custom_iou