Event-assisted Low-Light Video Object Segmentation

In the realm of video object segmentation (VOS), the challenge of operatingunder low-light conditions persists, resulting in notably degraded imagequality and compromised accuracy when comparing query and memory frames forsimilarity computation. Event cameras, characterized by their high dynamicrange and ability to capture motion information of objects, offer promise inenhancing object visibility and aiding VOS methods under such low-lightconditions. This paper introduces a pioneering framework tailored for low-lightVOS, leveraging event camera data to elevate segmentation accuracy. Ourapproach hinges on two pivotal components: the Adaptive Cross-Modal Fusion(ACMF) module, aimed at extracting pertinent features while fusing image andevent modalities to mitigate noise interference, and the Event-Guided MemoryMatching (EGMM) module, designed to rectify the issue of inaccurate matchingprevalent in low-light settings. Additionally, we present the creation of asynthetic LLE-DAVIS dataset and the curation of a real-world LLE-VOS dataset,encompassing frames and events. Experimental evaluations corroborate theefficacy of our method across both datasets, affirming its effectiveness inlow-light scenarios.