Working in batches

When setting up a new project, it’s helpful to start with a smaller batch of tasks before scaling it for production. For example, if your target dataset has 1,000 tasks, you could start with a batch of 100 tasks.

Working in smaller batches allows you to review annotations as well as labeller feedback for completed tasks. This helps you understand how to update the project’s instructions for the next batch.

Setting up tasks

The complexity of a project has a direct impact on the annotation quality. To ensure that labellers are able to clearly understand your project needs, it is helpful to break down complex tasks by:

1. Reducing the number of labels.
2. Reducing the number of annotations per image.