Attribute tagging is the process of labeling specific characteristics or properties of an object that's already been identified in an image or video, like color, size, material, or condition, rather than just marking that the object exists. It adds a layer of detail on top of basic object detection.

‍

‍Why it matters

‍ Knowing that a bounding box contains a car is often not enough information for a model to actually be useful, a lot of real applications need to know the car's color, whether it's parked or moving, or whether it shows visible damage. Attribute tagging is what supplies that extra layer, without it, a dataset can technically identify objects correctly while still being far too generic for the specific task a model needs to perform.

‍

‍How it's done

Annotators typically tag attributes after the initial object detection or segmentation step is already complete, working from a predefined list of attribute options for each object type rather than describing objects freely in their own words. Tools like CVAT and Labelbox support this as an additional labeling layer on top of a bounding box or segmentation mask, and getting the attribute list right up front, deciding exactly which properties actually matter for the task, is usually more important than the tagging process itself.