CVPR 2023 Conference Tutorial
Exploring Synthetic data as an Enterprise Capability for Training and Validating CV Systems
Presenters
Nathan Kundtz, Ph.D, CEO Rendered.ai nathan@rendered.ai
Matt Robinson, Rendered.ai matt@rendered.ai
Dan Hedges, Rendered.ai dan@rendered.ai
Scheduled Date/Time
Monday June 19th, Afternoon 13:30-16:30
Tutorial Content
Presentation Slides (Coming soon)
Video Recording (Coming soon)
- Introduction to physics-based synthetic data
- Welcome and Introductions
- Value of synthetic data
- Physics-based synthetic data
- Discussion of Generative Synthetic data
- Workflows for engineering synthetic data for iterative improvement and innovation
- Running an experiment with synthetic data
- Introduction to the Rendered.ai Platform
- Overview of concepts and user interface
- Step-by-step tutorial on generating and accessing synthetic data
- Creating synthetic data for real world use cases
- Walkthrough of the process of building a custom synthetic data channel
- Demonstration of NVIDIA Omniverse Replicator running in the AWS cloud for synthetic data generation
- Walkthrough of using AWS SageMaker with Rendered.ai to integrate synthetic data into an AI training pipeline
- Using domain adaptation and dataset comparison to create 'better' synthetic data
- Discussion of CycleGAN domain adaptation and randomization techniques to match real data output while preserving training image semantics
- Adding custom and standard annotation to a dataset
- Using UMAP data characterization to explore explainability of AI training outcomes
- Brief case study with Eigen Innovation Image Twin
- Wrap up discussion
In partnership with
