Researchers, developers, and enterprises need large, carefully labeled datasets to train high accuracy neural networks for perception AI applications. Academic datasets, while great for benchmarking, are very limited to the classes they provide. Acquiring ground truth data for new labels or tasks is costly, time-consuming, prone to errors, and may not capture all the edge cases—and sometimes, does not exist. All of these issues lead to delays and high costs for creating effective AI solutions.

Enter Omniverse Replicator, an open, modular SDK for building custom synthetic data tools and data sets. Omniverse Replicator is a highly extensible framework built on a scalable Omniverse platform that enables physically accurate 3D synthetic data generation to accelerate the training and performance of AI perception networks.

• Understand the difference between training with synthetic and real data, how to bridge the gap between the two.

• Get an overview of the Omniverse Replicator SDK, exposed in the new Omniverse Code IDE application.

• Utilize Omniverse Replicator’s randomizers to create multiple variations of a Universal Scene Description (USD) scene, store the desired ground truth of the scene with annotators and write the data in your desired format.

• Get your questions answered in a live Q&A session with the Omniverse Replicator technical product managers

Content

content goes here