Skip to main content
Annotation is where you teach the model what to look for: draw a box around each object and assign it a class — the label your model will predict.

​
Set up your classes

Before labeling, define the classes your project detects:
  1. In your project, open Classes & Tags.
  2. Add a class for each thing you want to detect (for example, pass and fail), each with its own color.
Classes and tags

Each class gets a name and a color; you can also add free-form tags.

​
Label a batch

  1. Open Annotate and select a batch from the Unassigned column.
  2. For each image, draw a box around every object and pick its class.
  3. Use the arrows to move through the batch — your progress is tracked as you go.
Annotation editor

Draw a box around each object and assign it a class.

Label consistently — tight boxes, the right class every time. A small, clean dataset beats a large, noisy one.

​
Review and add to the dataset

As you finish images, mark them reviewed. Filter by Reviewed / Not reviewed or Has annotations / No annotations to check your work. When a batch is ready, choose Mark all reviewed & add to dataset and assign a train/validation/test split. The batch moves into the Dataset column, ready to snapshot.

​
Next steps

Create a dataset version

Freeze your labeled images into a reproducible snapshot.

Projects & batches

How batches move from collection through annotation to dataset.