Changelog (v26.2.12)

New

• Semantic segmentation: new Brush, Eraser and Lasso tools. Use the Brush to paint a mask, the Eraser to remove parts of it, and the Lasso to outline a region freehand.
• Consensus labeling now runs in parallel: multiple labelers can work on the same consensus asset at the same time, without waiting for the previous labelers to submit. Labeling remains blind, and consensus is computed once all required labels are submitted.
• Queue filters: the Predicted classes filter now offers "No predictions" and "Has predictions", and the Label classes filter offers "No value" and "Has value", to find assets with or without any prediction or label across all jobs.
• Queue filters: new "is not" and "is neither of" operators on the Predicted classes and Label classes filters, to exclude assets with specific classes.
• On Audio projects, asset metadata is now displayed in the labeling interface.

Change

• Smart tracking now supports multiple objects: select several Bounding Box or Semantic objects in the labeling interface, then launch tracking to track them all at once.
• Interactive segmentation is faster, especially on repeated clicks on the same image and on very large images.
• Python SDK: assign_assets_to_labelers() and the other queue action methods now return the succeeded, declined and failed assets.

Fix

• In the Labeling Jobs Editor, the OCR option for transcription sub-jobs is now only available under Bounding Box tools, the only geometry supported by OCR. Plain text transcription remains available for all geometries.
• Smart tracking is now only offered on Bounding Box and Semantic tools.
• Deleting a workflow step and then adding a new one no longer fails.
• Up to 8 levels of nested jobs are supported again on image, PDF and text projects.
• Video labels with very long object tracks now load correctly.
• An asset unassigned from a labeler no longer remains in their queue.
• On Audio projects, filtering on incomplete jobs works again, and labels can be submitted when optional jobs are left empty.
• Python SDK: using isUsedForConsensus with update_properties_in_assets now returns a clear message pointing to update_asset_consensus.
• Python SDK: the first parameter now returns the expected number of results.