Skip to content

Version 3.0.0
Compare
Choose a tag to compare
@msokoloff1 msokoloff1 released this 12 Aug 23:06
V3.0.0
e1d5882
This commit was created on GitHub.com and signed with GitHub’s verified signature. The key has expired.
GPG key ID: 4AEE18F83AFDEB23
Expired
Verified
Learn about vigilant mode.

Version 3.0.0 (2021-08-12)

Added

  • Annotation types
    • A set of python objects for working with Labelbox data
    • Creates a standard interface for both exports and imports
    • See example notebooks on how to use under examples/annotation_types
    • Note that these types are not yet supported for tiled imagery
  • Model Diagnostics Support
    • Model Diagnostics beta users can now just use the latest SDK release
  • Metadata support
    • New metadata features are now fully supported by the SDK
  • Easier export
    • project.export_labels() accepts a boolean indicating whether or not to download the result
    • Create annotation objects directly from exports with project.label_generator() or project.video_label_generator()
    • project.video_label_generator() asynchronously fetches video annotations
  • Retry logic on data uploads
    • Bulk creation of data rows will be more reliable
  • Datasets
    • Determine the number of data rows just by calling dataset.row_count.
    • Updated threading logic in create_data_rows() to make it compatible with aws lambdas
  • Ontology
    • OntologyBuilder, Classification, Option, and Tool can now be imported from labelbox instead of labelbox.schema.ontology

Removed

  • Deprecated:
    • project.reviews()
    • project.create_prediction()
    • project.create_prediction_model()
    • project.create_label()
    • Project.predictions()
    • Project.active_prediction_model
    • data_row.predictions
    • PredictionModel
    • Prediction
  • Replaced:
    • data_row.metadata() use data_row.attachments() instead
    • data_row.create_metadata() use data_row.create_attachments() instead
    • AssetMetadata use AssetAttachment instead

Fixes

  • Support derived classes of ontology objects when using from_dict
  • Notebooks:
    • Video export bug where the code would fail if the exported projects had tools other than bounding boxes
    • Model-assisted labeling demos were broken due to an image download failing.

Misc

  • Data processing dependencies are not installed by default to for users that only want client functionality.
  • To install all dependencies required for the data modules (annotation types and mea metric calculation) use pip install labelbox[data]
  • Decrease wait time between updates for BulkImportRequest.wait_until_done().
  • Organization is no longer used to create the LFO in Project.setup()
Assets 2
Loading