April 3, 2023
over 1 year ago
Release notes
App
- You can now upload up to 10 custom embeddings to Labelbox. Custom embeddings are helpful for improving data exploration and similarity search experience. To learn more, visit this page.
- In Catalog, you can use the new find text filter to narrow down text, conversational text, documents, and HTML data rows by keyword. To learn more, visit this page.
- With the new natural language search filter in Catalog and Annotate, you can now describe the data rows you're looking for in plain English and Labelbox will surface the relevant data rows. Note: this functionality currently works for images and geospatial data. For more information, visit this page.
- We added a brush tool to the video & DICOM editors. You can use the brush tool to quickly draw segmentation masks. This feature is in beta. To learn more about the brush tool, see brush tool in video editor and brush tool in DICOM editor.
- When you export from a project, you'll get a new and improved export format that gives you more flexibility to retrieve the most valuable information from your data rows. Exports v2 enables you to use filters to select and export a subset of data rows. You can also choose to exclude parameters from the export that you do not need. This feature is in beta. For more information, see Export from project.
- You can now export data from a model run. Use the filters to query data row label status, metadata, batch, annotations, and workflow history. You can also choose to exclude parameters from the export that you do not need. This feature is in beta. For more information, see Export model run data.
- You can now export everything about a data row from the Catalog UI including slice information, dataset name, annotations, and model run data. To learn more, visit Export from catalog.
- You can now send model predictions that are stored in a model run to a labeling project as pre-labels. This allows you to inspect the model predictions before sending them to a labeling project. For more information, see this page.
- If you are browsing your data rows in the Model product and you discover data rows that contain labeling mistakes, you can now send those data from a model run to a labeling project for re-labeling. To learn more, read this page.
- We enabled the ability to turn off linear interpolation in the video editor at the organization level. When interpolation is disabled, objects will maintain the geometry of a keyframe until the next keyframe, instead of linear transitions between keyframes. To enable this for your organization, contact [email protected].
- We enabled three new filters in the Data Rows tab: Has review (legacy), Reviewed by (legacy), and Reviewed at (legacy). These filters allow you to see your legacy review information in the new Data Rows tab. For more information, visit this page.
- All of your projects that have been migrated to the new workflows/batch queue/data row tab paradigm are marked with a “legacy” tag. To learn more, visit this page.
- In the Workflows tab, the “send to task” functionality now supports up to 10,000 data rows.
- The zoom experience in the labeling editor is less sensitive and is much easier to achieve specific zoom levels than before.
- We introduced a new, simplified two-step process for importing annotations to Labelbox. Now, you can upload data rows and annotations in one call using global keys. For more information, visit this page.
- When importing metadata, you can now import multiple enum options.
- When you are in the Model product, the URL now contains the state of filters, slices, and splits so you can more easily share exact views of the Model tab with your team.
- AI-based bounding box tracking in video is now available as GA. To learn more about bounding box tracking, visit this page.
- We enabled support for Select all on the Data Rows tab With this feature you can now perform this bulk action with a single click.
- You can now bulk add metadata to data rows via the Catalog UI. For instructions, visit this page.
- Word selection is now available in the text editor. This feature allows you to hover over and annotate full words with one click, rather than labeling text characters. For more information, see this page.
Python SDK
The latest version of our Python SDK is v3.42.0. See our full changelog in Github for more details on what was added recently.
Version 3.42.0 (2023-03-22)
Added
- Message based classifications with annotation types for conversations
- Video and raster segmentation annotation types
- Global key support to
ConversationEntity
,DocumentEntity
andDicomSegments
- DICOM polyline annotation type
- Confidence attribute to classification annotations
Changed
- Increased metadata string size limit to 4096 chars
- Removed
deletedDataRowGlobalKey
fromget_data_row_ids_for_global_keys()
Fixed
- Annotation data type coercion by Pydantic
- Error message when end point coordinates are smaller than start point coordinates
- Some typos in error messages
Notebooks
- Refactored video notebook to include annotation types
- Replaced data row ids with global keys in notebooks
- Replaced
create_data_row
withcreate_data_rows
in notebooks
Version 3.41.0 (2023-03-15)
Added
- New data classes for creating labels:
AudioData
,ConversationData
,DicomData
,DocumentData
,HTMLData
- New
DocumentEntity
annotation type class - New parameter
last_activity_end
toProject.export_labels()
Notebooks
- Updated
annotation_import/pdf.ipynb
with example use ofDocumentEntity
class
Version 3.40.1 (2023-03-10)
Fixed
- Fixed issue where calling
create_batch()
on exported data rows wasn't working
Version 3.40.0 (2023-03-10)
Added
- Support Global keys to reference data rows in
Project.create_batch()
,ModelRun.assign_data_rows_to_split()
- Support upserting labels via project_id in
model_run.upsert_labels()
media_type_override
param to export_v2last_activity_at
andlabel_created_at
params to export_v2- New client method
is_feature_schema_archived()
- New client method
unarchive_feature_schema_node()
- New client method
delete_feature_schema_from_ontology()
Changed
- Removed default task names for export_v2
Fixed
- process_label() for COCO panoptic dataset
Notebooks
- Updated
annotation_import/pdf.ipynb
with more examples - Added
integrations/huggingface/huggingface.ipynb
- Fixed broken links for detectron notebooks in README
- Added Dataset QueueMode during project creation in
integrations/detectron2/coco_object.ipynb
- Removed metadata and updated ontology in
annotation_import/text.ipynb
- Removed confidence scores in
annotation_import/image.ipynb
- Updated custom embedding tutorial links in
basics/data_row_metadata.ipynb