This research tackles a critical challenge in DNN testing: how to effectively test computer vision systems when simulators cannot provide ground-truth data for verification.

• Introduces a GAN-enhanced approach that generates synthetic inputs for testing computer vision DNNs
• Enables cost-effective testing of DNNs in safety-critical applications
• Creates a simulation-driven framework that works without ground-truth data
• Bridges the gap between simulation capabilities and verification requirements

For security professionals, this research provides a novel solution for verifying neural networks in autonomous systems where traditional testing approaches fail due to ground-truth limitations, helping ensure system reliability in critical applications.

GAN-enhanced Simulation-driven DNN Testing in Absence of Ground Truth