Data annotation is such a crucial part of machine learning. Just like when baking a cake: no matter how good you are at baking, without great ingredients, there's going to be a limit to how tasty your cake can be. Your training data, which you create using the platform's Annotation tool, are the ingredients in your machine learning recipe!
Because this is such an important component of the platform, we have a series of articles devoted to annotation. I'll link to those below—think of this article as the introduction plus a table of contents.
First, some terminology
There are a handful of terms that get used together or sometimes interchangeably when it comes to this type of annotation, so I want to quickly describe them. That way, when you come across them, you'll already be familiar with their meaning.
Annotation, or data annotation, is the term we prefer at CrowdAI to refer to the step of the machine learning lifecycle where you apply different types of labels to your media to create training data. This step is also often called data labeling (sometimes spelled labelling) because you're applying labels to your data.
Categories are what you're looking for when you annotate. Perhaps you're looking for a type of defect in an image from your production line (category =
defect), or you're searching geospatial imagery for evidence of flood damage (category =
flood-damage). These are often also referred to as classes in the machine learning world, but we prefer to use the term Categories, so you'll see that in the platform.
(At CrowdAI, we call these Categories, but elsewhere you may hear the term classes or even ontology, which have similar meanings. We think Categories is a bit easier to understand, though.)
Labels are the application of Categories to your media. They're the individual units of information you attach to your media to say "this is the thing I'm looking for in this image." An image or video might have zero, one, or many labels, depending on what you're looking for.
As you go through the Annotation process, you create training data, which is the combination of your media and all the labels that go with them.