Human-Centric Computer Vision with Synthetic Data

Companies are continuing to accelerate the adoption of computer vision to detect, identify and understand humans from camera imagery. We see these human-centric use cases in a growing range of applications including augmented reality, self-checkout in retail, automated surveillance and security, player tracking in sports, and consumer electronics. Creating robust solutions for human-centric computer vision applications requires large, balanced, carefully curated labeled data sets. But acquiring real-world image and video data of humans is challenging due to concerns around bias, privacy and safety. And labeling and curating real-world data is expensive and error-prone. Synthetic data provides an elegant and cost-effective alternative to these challenges. In this session, we will show how Unity’s tools and services can be used to quickly generate perfectly labeled, privacy-compliant, unbiased datasets for human-centric computer vision.