HumanSignal/label-studio v1.4.1

Release 1.4.1

on GitHub

Notable changes

Starting in this version, when you add an ML backend, a webhook is created to send an event to your ML backend each time an annotation is updated or created. Alongside this change, the following has also changed:

• The completions parameter of the fit() method is deprecated and will be removed in the next release.
• "Start model training after any annotations are submitted or updated" is now unchecked by default.
• Requests to the /train API endpoint now include a timeout.

See train a model with your ML backend for more details about this change.

Changes from contributors

• Add a GitHub action to warn about stale issues and PRs to improve team responses. Thanks @twsl for your contribution in #1754.
• Fix an issue where the Relation tag was ignored by the parse_config function in the ML backend SDK. Thanks to @tomfat for your first contribution in #1811!
• @YuMS corrected the example syntax for the task import API in #1831. Thanks for your first contribution!
• @AaronDeRybelHowest added a Flair ML backend example in heartexlabs/label-studio-ml-backend#68. Thanks so much for your contribution!!
• @Fireblossom fixed an issue with the PyTorch ML backend in this PR: heartexlabs/label-studio-ml-backend#80. Thanks for contributing!
• Improved performance of the encode_rle function used to export brush masks thanks to contributions from @csaroff in heartexlabs/label-studio-converter#52.
• @changrq fixed an issue with YOLO coordinates deviating when a rectangle is rotated in heartexlabs/label-studio-converter#73. We appreciate your contribution!
• @raulbocanegra added a required project key to the payload of storage exports and imports in the SDK in heartexlabs/label-studio-sdk#14. Thanks for contributing!
• @denck007 fixed an issue where the get_users() function in the SDK didn't work for open source users because of a dependency on organization membership, which is an Enterprise-only functionality. Thanks for your contribution in heartexlabs/label-studio-sdk#23!
• @fabiolab extended the SDK to allow extra headers for HTTP requests. Thanks for your contributions in heartexlabs/label-studio-sdk#30.
• @laynr added the ability to sync storage to the SDK in heartexlabs/label-studio-sdk#20. Thanks for your consistent contributions!
• @Fkawala added the ability to use email and password credentials to log into the SDK, rather than using a token. Thanks for your contribution in heartexlabs/label-studio-sdk#29.
• @kaustuk added the ability to set a cursor color and width for the Audio tag. Thanks for your contributions in heartexlabs/label-studio-frontend#347 and requesting this feature in #1682.

Features

• Add auto-created webhooks when you add an ML backend to Label Studio. See more in the documentation.
• Add the option to reveal pre-annotations interactively
• Add the ability to enable experimental functionality with the EXPERIMENTAL_FEATURES environment variable.
• Add the ability to specify a lead_time to imported annotations.
• Add the ability to view created_at time for annotations and updated_at time for tasks on the data manager.
• Added an ELECTRA model example for the ML backend in heartexlabs/label-studio-ml-backend#76.
• Added a brush predictions example for the SDK.
• When performing active learning with Label Studio, the model version selected to display predictions on tasks defaults to the most recent model version.

Labeling and tag updates

• Add the ability to uncheck a selected checkbox when using the Choices tag.
• Add the overviewWidth parameter to the TimeSeries tag to set a default overview width in percents, requested by @TueLindhart in issue #1330.
• More fully support the Repeater tag.
• Add a vertical zoom control.
• Add a resolver parameter to the Text tag to retrieve data from a remote URL in PR #396.
• Add tooltips to zoom controls in audio labeling.

Notable bug fixes

• Fixed a bug where relations break the labeling interface when added as predictions, or when viewing annotations with relations.
• Fixed issues with permissions when setting up local file storage connections.
• Fixed issues with building and installing Label Studio taking a long time due to resolving various google dependencies.
• Fixed an issue where the selected=true option for the Choices tag wasn't being used. Thanks to @DougTrajano for reporting in issue #1848.
• Fixed issue #1757 reported by @uriel-tannen that prevented saving a labeling configuration without existing labels.
• Improved the functionality of the CONLL converter to be more robust as suggested by @tomasohara in issue #61, and also updated to handle empty labels and complicated configurations.
• Fixed an issue with Docker builds on MacOS reported by @simonlsk in issue #1859.
• Fixed an issue where the --version flag doesn't print the version reported by @magicaltoast in issue #1880.
• Fix an issue where shortcuts did not work when editing TextArea results.

Other bug fixes

• Fixed a bug with undo where it deselected a region instead of removing it for brush masks.
• Fixed bugs where unclosed tags and other errors when defining labeling configurations could cause Label Studio to break.
• Fixed an issue where bounding boxes failed to rotate.
• Fixed an issue where VOC export failed when trying to export using the snapshot functionality.
• Removed a dependency on the ground truth field for determining when labeling is complete.
• Fix performance issues when retrieving project storage.
• Fix crosshair functionality when zooming and rotating images.
• Improve appearance of tabs on the data manager when there are a large number.
• Fixed the ability to go back to a previously-skipped task and submit an annotation.
• Fix the ability to use custom service account credentials for GCE.
• Fixed timeouts occurring with the ML API and added environment variables to configure the timeout limits in #1978.
• Fixed an issue with the mmdetection ML backend example.
• Improved YOLO export to have sorted classes.
• Improved handling of empty labels in YOLO, COCO, and VOC export formats.
• Fix an issue rendering certain data types in the "view all" annotations page.

Changes to requirements and dependencies

• Bump Pillow from 8.3.2 to 9.0.0 in the ML backend.
• Bump nltk from 3.5 to 3.6.6 in the ML backend.
• Bump django from 3.1.13 to 3.1.14 in Label Studio.

Docs changes

• Added the ability to search Label Studio website by typing on any page, and to quickly copy code samples using a copy button.
• Updated all Templates documentation to match the product templates and include more details about use cases and functionality.
• Updated API documentation:
  • To be more clear about IDs
  • To remove deprecated information from pre-1.0.0 versions of Label Studio.
  • To improve export API specs
• Fixed issues with Docker Compose documentation.
• Add details about importing URLs and links using the valueType parameter in tags, and also improved details and troubleshooting when handling HTML labeling. See import data.
• Add spaCy export information, improved export format JSON details, and to the export docs.
• Add dedicated data management documentation.
• Add details to troubleshoot pre-annotations
• Updated the tags documentation to be the latest version.
• Removed support for Python 3.6 as a result of supporting Pillow 9.0.0.