How to Build a Custom Synthetic Data Pipeline to Train AI Perception Models

Omniverse Replicator SDK is highly extensible framework build on a a scalable Omniverse platform that enables physically accurate 3D synthetic data generation to accelerate the training and boost the performance of AI perception networks. In this session we are building upon core concepts of Omniverse Replicator APIs we introduced during Spring GTC 2022 by discussing a typical process of creating a custom workflow for generating synthetic data for training a perception network. Training a perception model with a synthetic dataset is a multi-step process that requires considerations around simulation-ready assets, contextual scene generation, plausible randomizations, different annotators, photo-realistic rendering, and writing data that is usable by downstream machine learning frameworks. We walk through this process with an example use case and along the way introduce enhancements we are bringing to the product.
We are joined by our partners who have been building with us using Replicator SDK on the Omnivese platform and are accelerating what is possible.