Create and configure projects
All labeling activities in Label Studio occur in the context of a project.
Project setup workflow
Select a template to configure the labeling interface for your dataset. Set up the labeling interface for your project.
(Optional) Set up annotation settings for your project.
(Optional) Set up review settings for your project.
Import data into Label Studio.
For small projects, testing, or proof of concept work, you can import local files directly into Label Studio. However, for larger projects, we recommend setting up cloud storage. For more information, see Get data into Label Studio and Sync data from external storage.
After you have imported your data, then you can set up the labeling interface for your project.
Next, you will want to configure what information annotators and reviewers can see, and how automated you want their workflow to be.
By default, any annotator who is a member of project can begin labeling as soon as it is published, and each task only requires one annotator before being considered complete. To customize this, see the Annotation section of the project settings.
By default, reviewers do not need to be assigned to completed tasks, and each task only needs one accepted annotation. To customize this, see the Review section of the project settings.
Create a project
From Label Studio, click Create Project in the upper right. A window opens with three tabs:
- Project Name
-
This is the only required section.
Here, select your workspace, enter a project name, and (optionally) a project description.
Once complete, you can click Save to create the project, or you can complete the other tabs.
- Data Import
-
From here, you can upload files into Label Studio. You can do this now or after the project has been created.
However, for larger projects, we recommend setting up cloud storage or using a different import method. For more information, see Get data into Label Studio and Sync data from external storage.
- Labeling Setup
-
You can select a template to begin your labeling configuration. For easier setup, select a template. You can later customize template to meet your needs. See Configure labeling.
You can do this now or after the project has been created.
When you’re done, click Save.
Tip
Rather than importing data directly into the project, you can create a dataset. From here, you can use an AI-powered search to refine your data, which can then be added to different projects as tasks. For more information, see Data Discovery overview.
Configure high-impact settings
By design, Label Studio is highly customizable and there are numerous configuration options for a project. To configure project settings, open a project and click Settings in the upper right.
To avoid getting overwhelmed, focus on the following settings. They have the most impact on your labeling experience.
Annotation settings
Annotators are the users who are labeling project tasks.
Distribute labeling tasks
Located under Annotation, this determines whether annotators must be manually assigned to a task in order to label it.
If you are using Auto distribution, project members can begin labeling as soon as the project is published. Otherwise, they must be manually assigned.
Allow empty annotations
Located under Annotation, this determines whether annotators can complete tasks without first adding a label.
By default, annotators are allowed to submit empty annotations. You can change this setting so that all tasks require a label.
Annotations per task minimum
(This is only applicable if you are using Auto distribution).
Located under Quality, this determines how many annotators must submit a task before the task is considered completed.
By default, each task only requires one annotator to submit. You can configure a higher task overlap, meaning that each task must have a minimum number of annotators before being considered complete.
For a description of all the settings available for annotators, see Project settings - Annotation and Project settings - Quality.
See the following video for a brief overview of automated task assignments:
Review settings
Reviewers are the users are reviewing the annotators’ submissions for accuracy.
Mark task as reviewed after it has at least 1 accepted annotation or
Mark task as reviewed after all annotations are processedConfigure what action determines that a task has been reviewed. By default, the reviewer only needs to accept one annotation for the task. You can change this so that the reviewer must accept or reject submissions from all annotators.
Review only manually assigned tasks
By default, reviewers are able to begin reviewing as soon as there are labeled tasks available to review. However, you can change this so that reviewers must be manually assigned to a task.
For a description of all the settings available for reviewers, see Project settings - Review.
Add members to a project
You can add members to a project in two ways:
- Add members at the workspace level. Workspace membership is inherited by projects.
- Add members at the project level.
To add members to a specific project:
Navigate to the project settings and select Members.
Use the search functionality to locate the user that you want to add to the project.
Select the checkbox next to their name and click the
>
arrow to add them. Click the opposite arrow<
to remove members.If the user’s organization-level role is Annotator or Reviewer, you can use the drop-down menu to assign them a role specific to this project.
Project-level roles are Annotator or Reviewer. So, for example, a user can be an Annotator in one project and a Reviewer in another project.
Click Save.
Users are not sent notifications when they are added to a project.
For more information, see Project settings - Members.
Publish a project
In Label Studio Enterprise, you can hide projects from annotators so that you can fully configure the project before anyone can start labeling. When you’re ready for annotators to start labeling, publish the project.
- Open a project and select Dashboard in the upper right.
- Click Publish.