Steps and Patch

A step in a workflow defines an action to take when a job goes through it. The steps can be combined to set up a workflow.

Reminder: A job represents one unit of data to be annotated. For example, an image or a video.

Types of step

There are eight different types of steps:

1. Start Step

It represents the start of the workflow, any job you create passes through the start step in the workflow. It is present in all the workflows you create by default.

2. Maker Step

When a job reaches the Maker Step, it gets assigned to an annotator to create annotations.

Task configuration

• In the Annotation Tools section, you can select all the annotation types and classes to be enabled.

• Features specific to different task types:

Segmentation

AI assistance

Create pixel-perfect annotations with just four clicks, to learn more: AI assistance

Task timer

Display a timer for annotations tasks in GT create

Brightness Feature

Enable brightness control in GT create, to learn more: Image Adjustments

Video and Images

Attribute Config

Enable or disable attribute selection in GT Create. Learn more about Attributes.

ML Proposals

Get pre-labeled bounding box annotations on images and videos by our model. Learn more about ML proposals.

👉 Note: You have to connect the AI step before the maker step to enable ML proposals

Linear Interpolation

Predict annotations between frames using interpolation. Learn more about Interpolation.

👉 Note: It is only applicable for bounding box annotations.

Bulk Attribute Change

Mark attribute value for a set of frames at once. It is beneficial for review tasks.

• The annotations made by the maker go to maker_response key in the Job build.

To learn more on the structure of the Job build, you can have a look here:

Job Build Structure

3. Review Step

Create a review task to review the work done by a Maker. A reviewer can:

a) Edit the job and correct the mistakes made by the Maker. The review step then analyses the maker's and reviewer's submission to calculate the maker's accuracy and determine the type of mistakes made by the maker. You can see the Annotator Performance tab to track their performance.

b) Reject the job and provide feedback to the Maker. The job then goes back to the maker.

The reviewer's selection goes to the editor_approval_status key in the job build. To connect a review step with other steps, you can configure the workflow routes using this key.

• If a reviewer edits the job, editor_approval_status is set True

• If a reviewer rejects the job, editor_approval_status is set False

To know how to configure workflow routes, you can have a look here:

Workflow Routes

To know how a review task works in GT create, you can have a look here:

• Reviewer Task for Image Segmentation

• Reviewer Task for 2D Video & Images

4. Manual Step

After the data is processed through one or more steps in the workflow, you can use a manual step to collect jobs at one step. You can then manually push these Jobs to the next step as required.

5. AI Step

Pre-label up to 80 common classes automatically for bounding box annotations. It dramatically reduces the amount of time spent per annotation. To enable ML proposals:

• Connect the AI step before any Maker step.

• Select the model in the AI step.

• Enable ML proposals in the Maker task and save the workflow.

To know how ML proposal work in GT Create, you can have a look here.

6. End Step

It is the last step in a data annotation pipeline. Once one or more jobs reach the end step, you can perform a Quality Check or export the annotated data.

7. Empty Step

Use an empty step to configure complex workflow route logic.

Maker Reviewer Patch

It is a standard workflow patch with a maker step, a review step, and a manual step.