What and Where?

When generating synthetic data, there are two critical questions to address: "What?" and "Where?"

What Foreground Objects will be Inserted?

What objects do you want to insert? For example, the anomalies or defects you need, such as cracks in concrete. You can provide image samples of the defect, and/or you can enter a textual description of what the defect looks like.

Where Will Objects be Inserted?

Where do you want to generate objects? This involves specifying the background images (canvases) into which to insert the foreground objects (described above). It may also involve specifying your preference for the regions where the objects should be placed within the image. For instance, if you want to generate a puncture in a car tire, placing the puncture on the tire itself makes sense, not on the rim or hub cap.

dataspan.ai requires this information in order to accurately insert an object that looks natural and real into the appropriate areas of the background data. This ensures the synthetic data generated is realistic and maintains the original data distribution.