After uploading model predictions to the Model product, and analyzing model predictions and model metrics, you may find labeling mistakes that need rework. Labelbox allows you to send data from a model run to re-labeling.

How to send data to re-labeling

Step 1: Select data rows with labeling mistakes

You can find tips about how to find data rows with labeling mistakes here.

Step 2: Open the data rows in Catalog

Open the data rows in Catalog by clicking on Manage selection and then View in catalog.

Step 3: Add metadata to the data rows

Once in Catalog, select the data rows by clicking on Select all, and then add metadata to the data rows by clicking on X selected and Add metadata.

Add metadata to identify these data rows. For example, you can add a tag metadata with the value to re-label.

Step 4: Open the labeling project containing the labels to rework

Navigate to the labeling project in Annotate, which contains the labels to rework.

Then, go to the Data rows tab and filter the data rows that contain the metadata tag you created in Step 3. In our example, we filter on the metadata tag and the value to re-label.

Step 5: Send these data rows to rework

Select the data rows to rework and send them to the Rework task.

Step 6: Rework the labeling mistakes

The data rows will show up in the Rework task. Click Start reworking to fix the labeling mistakes.

During rework, already-existing labels are used as the starting point.