Improving Anomaly Segmentation with Multi-Granularity Cross-Domain Alignment

TL;DR

This paper introduces MGCDA, a novel framework that enhances anomaly segmentation by aligning features across synthetic and real domains at multiple levels, improving robustness and accuracy in autonomous driving applications.

Contribution

The paper proposes a multi-granularity cross-domain alignment framework with domain adversarial training and anomaly-aware contrastive learning, advancing domain adaptation in anomaly segmentation.

Findings

Outperforms existing methods on Fishyscapes and RoadAnomaly datasets

Achieves superior domain generalization and robustness

Enables parameter-free inference with various architectures

Abstract

Anomaly segmentation plays a pivotal role in identifying atypical objects in images, crucial for hazard detection in autonomous driving systems. While existing methods demonstrate noteworthy results on synthetic data, they often fail to consider the disparity between synthetic and real-world data domains. Addressing this gap, we introduce the Multi-Granularity Cross-Domain Alignment (MGCDA) framework, tailored to harmonize features across domains at both the scene and individual sample levels. Our contributions are twofold: i) We present the Multi-source Domain Adversarial Training module. This integrates a multi-source adversarial loss coupled with dynamic label smoothing, facilitating the learning of domain-agnostic representations across multiple processing stages. ii) We propose an innovative Cross-domain Anomaly-aware Contrastive Learning methodology.} This method adeptly selects challenging anchor points and images using an anomaly-centric strategy, ensuring precise alignment at the sample level. Extensive evaluations of the Fishyscapes and RoadAnomaly datasets demonstrate MGCDA's superior performance and adaptability. Additionally, its ability to perform parameter-free inference and function with various network architectures highlights its distinctiveness in advancing the frontier of anomaly segmentation.