A huge amount of data is required to train an AI model. But sometimes, that data is not available from real world sources, so data scientists use synthetic data for that. In machine vision applications, this means creating different environments and objects to train robots or self-driving cars, for example. But while there are quite a few tools for creating virtual environments, there aren’t a lot of tools for creating virtual objects.
At its Re-Mars conference, Amazon today announced Synthetix in the Sagemaker Ground Truth to create a nearly unlimited number of images of a given object in a variety of conditions and under different lighting conditions, as well as different ratios and other variations. a new facility.
Along with WorldForge, the company already provides a tool for creating synthetic visuals. “Inested of producer whole worlds For Robot To Step around, this is Specific To Belongings either individual Component,” AWS VP of Engineering Bill Vas told me. He noted that the company itself needed such a tool because the millions of packages that Amazon itself ships still didn’t have enough images to train the robot.
“Andhat Ground Truth Synthetics does you are Start with 3D pattern In One number of Different formats He you can do pull This one inRa This will happen Synthetic Make photorealistic Images He match resolution of sensor you is,” he explained. And while some customers today intentionally disturb or break physical parts of a machine, for example, to train their models to take pictures of them – which can quickly become quite expensive – they now use virtual parts instead. and do this millions of times if need be.
He gave the example of a customer making chicken nuggets. That customer used the tool to train his model to simulate a lot of deformed chicken nuggets.
Vaas noted that Amazon is also partnering with 3D artists to help companies that may not have access to the kind of in-house talent to get started with the service, which by default. Uses the Unreal Engine, although it also supports Unity and the open-source Open 3D Engine. Using those engines, users can even begin to simulate the physics of how those objects would behave in the real world as well.