FSNet: A Failure Detection Framework for Semantic Segmentation

Semantic segmentation is an important task that helps autonomous vehicles understand their surroundings and navigate safely. However, during deployment, even the most mature models are vulnerable to various external factors can degrade performance with potentially catastrophic consequences for vehicle its surroundings. To address this issue, we propose a failure detection framework identify pixel-level misclassification. We do so by exploiting internal features of model training it simultaneously network. During detector flags areas in image where has failed segment correctly. evaluate proposed approach against state-of-the-art methods achieve 12.30%, 9.46%, 9.65% improvement AUPR-Error metric Cityscapes, BDD100k, Mapillary semantic datasets.