With Model Run config (open beta), you can now configure, track, and compare essential model training hyperparameters alongside training data and data splits in a single place. This update allows you to easily track and reproduce model experiments and enables you to share best practices with your team.
We have added a new field to our label_* trigger webhooks called
labelCountInProject. It returns the number of labels that have been submitted for the Data Row the label is attached to. This will improve the ways our customers can understand when a Data Row with Consensus has been met.
We can now support Cartesian (aka pixel) level coordinates in our geospatial editor! With this option, you can map your annotations to the pixels of the max zoom level rather than a geospatial coordinate system.
With the new annotations count filter in Catalog, you can now filter Data Rows by the frequency of an annotation.
- For Workflow, as you go through multiple reviews, the system is eventually consistent and it can sometimes takes up to 15 minutes for the data row counts to stabilize.
- When using Label imports with Workflow, the data rows that have labels imported will end up in the first review step of the workflow.
- When working with the SDK, we will be disallowing projects to have quality setting of Consensus and queueing mode set to Batch.
- If you want to use Batch queueing mode while using the SDK, you will have to specify the Batch queueing mode as an explicit parameter as part of the project creation method call
The latest version of our Python SDK is v3.26.1. See our full changelog in Github for more details on what was added recently.
- Modifies get_config to return un-nested Model Run config
* Add training metadata config as a model run creation param
- Updates model run training metadata
- Resets model run training metadata
- Fetches model run training metadata
Batch.delete()which will delete an existing
Batch.delete_labels()which will delete all
Label’s created after a
Project’s mode has been set to batch.
- Note: Does not include labels that were imported via model-assisted labeling or label imports
- Support for creating model config when creating a model run
TEXT_FILEattachment types to replace the
- Label export will continue polling if the downloadUrl is None