GDRIVE: Adaptive Object Detection in Autonomous Vehicles via Graph-Based Feature Learning

Abstract

Navigating domain shifts in object detection is crucial for autonomous driving systems, particularly under varying weather conditions and diverse visual perspectives. Existing Cross-Domain Object Detection methods often struggle due to their reliance on broad semantic models, which can introduce biases and reduce detection accuracy in novel environments. To overcome these limitations, we propose GDRIVE, a novel framework tailored for robust object detection in autonomous driving contexts. GDRIVE employs a graph-driven domain adaptation strategy that achieves state-of-the-art performance by integrating a Feature Enhancement and Alignment module. This module effectively reconstructs noisy nodes and separates domain-specific styles from invariant object features. Furthermore, an Adaptive Graph Enhancement Module employs advanced probabilistic methods to minimize the influence of anomalous nodes, thus improving detection robustness. Experimental results demonstrate that GDRIVE significantly surpasses existing techniques in graph-driven domain adaptation for object detection in challenging autonomous driving scenarios.