Skip to main content

Introduction

Annotation, frequently called labeling or data labeling, is the process of transforming unlabeled data to labeled data. Since labeled data is frequently used to train machine learning applications, it's important that the labeling is done to a high standard -- annotations must strive to be completeand accurate.

When applied to the unstructured data modalities in computer vision, there may be several objectives depending on the specific problem at hand. Frequent goals include category labeling, and object detection and localization, including nested attribute annotation on specific object instances. Encord provides the tools to coordinate and monitor the performance of your labeling tams and projects with our task management system, the power to precisely describe your data with our flexible and nested ontologies and the ability to precisely annotate them with our configurable and powerful label editor.

Label editor#

The label editor is the primary way to view, create and edit annotations in Encord. The editor supports various computer vision modalities and is designed with maximal configurability and scalability in mind. The editor automatically adjusts the interface based on the data type tasked for labeling, and currently supports the following data types:

You can use all the annotation types available in Encord across all modalities such as the:

  • Bounding box
  • Polygon
  • Polyline
  • Point
  • Custom object primitives
  • Segmentation mask (via export)
  • Frame classification (radio classification, checklist classification, free-form text input)
  • Dynamic classification (video only - radio classification, checklist classification, free-form text input)

Classifications can be applied at the frame or image-level, as attributes in objects with localization properties, and as events in objects with localization and temporal properties (for videos & image sequences).

The image above depicts the editor interface for a single image. To get oriented to the powerful features and settings of label editor, head to the label editor overview page. Alternately, click any of the data modalities listed above to jump to detailed interface explanations for that specific modality.