Set up your team's workflow
Your first opportunity to manage how your team approaches their tasks comes on page 3 of adding a new dataset, where you'll be prompted to enter instructions for your team.
In the following page, you'll be able to define the annotation workflow for your team. You can choose from the Basic Workflow template which consists of an initial Annotate stage, followed by a Review stage, or clone a workflow from an existing dataset.
Both options will take you to the workflow creation wizard, where the workflow can be configured.
Drag and drop stages your workflow will require, and plug them together. The available stages that can be used are:
|Dataset||The first stage of any workflow - the dataset stage can be linked to the dataset which contains the files that will enter the workflow.|
|Annotate||A simple human annotation stage, the annotate stage should be followed by at least 1 review stage.|
|Review||Similar in functionality to the annotate stage, the key difference with a review stage is the ability to send files to two workflow stages: One where files will be sent when approved, and one where files will be sent when rejected.|
|Model||A stage that can be injected into a workflow to create a model-assisted workflow. When adding a model stage, any trained model in the Neural Networks page can be used to create annotations. Simply match the classes that the model was trained on to the corresponding classes in the dataset.|
|Complete||The final stage of a workflow. Once in a complete stage, files can’t be edited unless they are moved to a previous stage.|
Though they appear in the list of stages in the workflow creation wizard, code stages are not yet available for use with workflows. Code stages are one of many stage types that will be added to V7 in the coming delivery cycles.
Workflows can be configured from the dataset creation wizard, or from the Workflows page.
Once created, assignment rules for the workflow can be set in the Settings page of the dataset.
Name your workflows
Workers in your team will access data from the Workflows page of their account. Be sure to give your workflow a recognizable name to ensure that work takes place from the correct workflow.
Next up we'll take a look at how annotators can self-assign batches, as well as how you can manually assign tasks from your dataset.
Updated 26 days ago