Ensuring maritime safety and optimizing traffic management in increasingly crowded and complex waterways require effective waterway monitoring. However, current methods struggle with challenges arising from multimodal data, such as dimensional disparities, mismatched target counts, vessel scale variations, occlusions, and asynchronous data streams from systems like the automatic identification system (AIS) and closed-circuit television (CCTV). Traditional multi-vessel association methods often struggle with these complexities, particularly in densely trafficked waterways. To overcome these issues, we propose a graph learning-driven multi-vessel association (named GMvA) method tailored for maritime multimodal data fusion. By integrating AIS and CCTV data, GMvA leverages time series learning and graph neural networks to capture the spatiotemporal features of vessel trajectories effectively. To enhance feature representation, the proposed method incorporates temporal graph attention and spatiotemporal attention, effectively capturing both local and global vessel interactions. Furthermore, a multi-layer perceptron-based uncertainty fusion module computes robust similarity scores, and the Hungarian algorithm is adopted to ensure globally consistent and accurate target matching. To validate the efficacy of our method, we have also constructed a new maritime multimodal dataset (termed MaritimeMmD) for data fusion. Extensive experiments demonstrate that GMvA delivers superior accuracy and robustness in multi-vessel association, outperforming existing methods even in challenging scenarios with high vessel density and incomplete or unevenly distributed AIS and CCTV data. The source dataset and code are available athttps://github.com/LouisYxLu/GMvA

Songjie Niu, Kaisen Yang, Rui Zhao, Yichao Liu, Zonglin Li, Hongning Wang, Wenguang Chen. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Vision-Language-Action (VLA) models for autonomous driving show promise but falter in unstructured corner case scenarios, largely due to a scarcity of targeted benchmarks. To address this, we introduce Impromptu VLA. Our core contribution is the Impromptu VLA Dataset: over 80,000 meticulously curated video clips, distilled from over 2M source clips sourced from 8 open-source large-scale datasets. This dataset is built upon our novel taxonomy of four challenging unstructured categories and features rich, planning-oriented question-answering annotations and action trajectories. Crucially, experiments demonstrate that VLAs trained with our dataset achieve substantial performance gains on established benchmarks—improving closed-loop NeuroNCAP scores and collision rates, and reaching near state-of-the-art L2 accuracy in open-loop nuScenes trajectory prediction. Furthermore, our Q&A suite serves as an effective diagnostic, revealing clear VLM improvements in perception, prediction, and planning. Our code, data and models are available at https://github.com/ahydchh/Impromptu-VLA