Out-of-Distribution Segmentation in Autonomous Driving: Problems and State of the Art

TL;DR

The paper tackles dense OoD segmentation for autonomous driving, focusing on road-obstacle detection under real-world domain shifts. It surveys major methodological families—outlier exposure, uncertainty estimation, generative models, and Mask2Former-based approaches—evaluating them on the SMIYC-OT and LostAndFound-NoKnown benchmarks to reveal strengths and limitations. A key takeaway is that improvements in OoD detection often accompany inlier degradation, with architecture dependence and threshold calibration posing challenges for real-world deployment. The work highlights the need for real-time, edge-friendly solutions, multimodal datasets, and integration with downstream planning to translate OoD segmentation advances into safer autonomous systems.

Abstract

In this paper, we review the state of the art in Out-of-Distribution (OoD) segmentation, with a focus on road obstacle detection in automated driving as a real-world application. We analyse the performance of existing methods on two widely used benchmarks, SegmentMeIfYouCan Obstacle Track and LostAndFound-NoKnown, highlighting their strengths, limitations, and real-world applicability. Additionally, we discuss key challenges and outline potential research directions to advance the field. Our goal is to provide researchers and practitioners with a comprehensive perspective on the current landscape of OoD segmentation and to foster further advancements toward safer and more reliable autonomous driving systems.