Step 6, Creating Synthetic Data Concepts

Understanding Concepts

This is the stage where you teach dataspan.ai "What" to insert and "Where" to insert it in images.

Note: The result of this stage will later be used to generate images at scale in the Batches stage (which is described later in Step 7, Generating Batches – Applying Concepts at Scale).

Creating a Concept means teaching dataspan.ai’s generative AI model about the type of foreground object to insert into background images.

Each Concept typically represents a single class, meaning a type of object (foreground) to insert on background images. For example, in regard to various types of lumber, you might want to insert resin or live knots – each is a class.

Clicking on a Concept displays the images that were synthetically generated by dataspan.ai, meaning images in which the foreground objects have been inserted into the background objects. For example, as shown below:

The Concept samples box specifies the name of the Dataset and class to be used to teach dataspan.ai about the objects to insert. For example, Lumber detects Datasets contains a folder named live_knots that contains 5 – 15 images of knot defects on lumber.

Clicking on an image zooms into it and provides tools for inspecting it. Selecting the Show Bounding Box option visualizes the bounding box or segmentation mask associated with this image around the object inserted by dataspan.ai. For example, as shown below: