Quantifying and Bridging the Fidelity Gap: A Decisive-Feature Approach to Comparing Synthetic and Real Imagery

Virtual testing with synthetic data is crucial for autonomous vehicle safety, but pixel-level fidelity doesn't guarantee real-world transfer. This paper introduces Decisive Feature Fidelity (DFF), an SUT-specific metric using Explainable AI to compare causal evidence driving decisions across real and synthetic domains, and proposes a DFF-guided calibration scheme to enhance simulator fidelity.