Out-of-Distribution Segmentation via Wasserstein-Based Evidential Uncertainty

TL;DR

This paper introduces a Wasserstein-based evidential framework for out-of-distribution segmentation in deep neural networks, enhancing the detection and segmentation of unknown objects in open-world scenarios for safer AI applications.

Contribution

It proposes a novel Wasserstein loss with regularization and structural consistency for improved OOD segmentation, advancing beyond existing uncertainty-based methods.

Findings

Improved OOD segmentation accuracy over baseline methods

Effective detection of unknown objects in semantic segmentation

Enhanced safety in open-world AI applications

Abstract

Deep neural networks achieve superior performance in semantic segmentation, but are limited to a predefined set of classes, which leads to failures when they encounter unknown objects in open-world scenarios. Recognizing and segmenting these out-of-distribution (OOD) objects is crucial for safety-critical applications such as automated driving. In this work, we present an evidence segmentation framework using a Wasserstein loss, which captures distributional distances while respecting the probability simplex geometry. Combined with Kullback-Leibler regularization and Dice structural consistency terms, our approach leads to improved OOD segmentation performance compared to uncertainty-based approaches.