This work was done as part of a research project with RISE and Smart Eye, where we created an app to generate realistic driver faces to be used for training and validation of autonomous-driving systems.

Features

• Models tested: We used StyleGAN based models with several loss functions, picking the best one.
• Customizable image parameters: The user specifices which facial features, expressions, and ethnicity should the image contain.
• End result: Provides real-world situations that improve AI accuracy in production scenarios

Impact

The solution enabled an autonomous-driving company to add to its training and validation data for its AD models. The end result reduced cost and time by generating diverse and rare data, which is hard to collect in natural settings.