Improving DNN Fault Tolerance in Semantic Segmentation Applications

Semantic segmentation of images is essential for autonomous driving and modern DNNs now achieve high accuracy. Automotive systems must comply with safety standards, requiring hardware fault detection. We present an analysis of the effect of faults using Google’s DeepLabV3+ network processing an industrial data-set. A new symptom-based fault detection algorithm is shown to detect >99% of critical faults with zero false positives and a compute overhead of 0.2%. Further, these faults can be masked, virtually eliminating all critical errors. To the authors’ knowledge this is the first fault tolerance study of a DNN semantic segmentation application.