Developer guide: Create a project via the Python SDK
The project is where you orchestrate all of your labeling operations. Use this guide to learn how to create a project and configure all available project settings.
To create a project via the Labelbox UI, follow these steps:
Under the top-level Annotate page, select New project. This will open a modal where you can configure your project type. Note that you will only be able to attach data rows that match the data type you set here. For example, you cannot select video as the data type upon project creation and then send image data rows to this project later.
After selecting a project type, name your project, optionally enter a description and select the type of data you will be labeling. If creating an LLM data generation project, you will also input a number of data rows to be generated by labelers.
After you select the data type and name your project, select a quality setting from the following options:
Benchmarks: Select benchmarks if you know that each data row you send to this labeling project will only need to be labeled once. Benchmarks allow you to designate, if desired, certain labeled data rows as a gold standard and work as a mechanism for measuring the performance of other labelers.
Consensus: Select consensus if you know that each data row will need to be labeled multiple times and, hence, should be labeled by multiple labelers.
If you are not sure where to start, select Benchmarks. Customers select this option for greater than 90% of projects.
Use project tags to enable easier organization and retrieval of the project. Click Create to establish the project and move on to more configurations.
Labelbox allows you to send a batch or batches of data rows to a project for labeling. When you click Queue data for labeling you will be sent to Catalog, where you can select a batch of data rows and send it to the project you are configuring.
In this step, you can filter and search your data rows to curate a batch. Click Send n data rows to create the batch.
Next, you will be prompted to name your batch and assign the data row priority. Click Submit batch to send this batch to your labeling project.
The final required step is to select the labeling interface (editor) that your project will use. To learn more about the various editors available, see our docs on labeling editors.
The Standard editor (recommended) is the native labeling interface and has the most robust labeling and quality analysis tools. However, if you need to set up a custom editor, select the Custom Editor.
If you choose to use the Standard Editor you will then be taken to a screen that lists all the ontologies created in Labelbox that match your project's data type
For instructions on creating an ontology, see Working with ontologies.
Once you finish these steps, the Start labeling button will be activated. However, there are two remaining optional setup steps available.
View quality assurance tools like benchmarks, consensus, and reviews (optional): This will take you to the quality settings page where you can set up (benchmarks), (consensus) and (Workflow) for reviews.
Review and manage members of this project (optional): This will take you to the Members page within the Settings section. For information on member permissions, see Add members or groups to a project.
Many customers need to use a dedicated external data labeling team. Labelbox offers a premium data labeling service integrated tightly into the Labelbox Annotate platform. See Workforce Boost for more details.
Currently, Workforce Boost is only available to customers who have purchased the software platform. You can get a free trial anytime by contacting [email protected]_.
To learn how to invite, onboard, and manage your own data labeling team, see Manage members.
Collaboration and rapid iteration are critical for achieving outstanding data quality
Our very best customers run their data labeling operations through real-time collaboration with the data labeling workforce, data engineers, and ML engineers. In Labelbox, you can easily augment your internal domain experts with external data labeling service providers.
To update project settings, go to the Settings tab from with a project.
Any time during the life of a data labeling project, you can easily update key project settings. Below are the most common things you will likely need to update:
- Add or remove data rows.
- Update the ontology.
- Manage the team members assigned to a project.
- Configure webhooks.
A project can only be deleted by an organization-wide admin. To delete a project, go to Settings > Danger zone, and proceed with the deletion.
Deleting a project will also delete all the annotations that have been submitted for a project.
Updated 3 months ago