UAIV: A Large-Scale Low-Altitude Multimodal Dataset for Urban Intelligence

The data is collected through a self-operated UAV acquisition pipeline, covering real-world complex urban scenarios with high resolution, multi-scale characteristics, dynamic variations, and strong scene heterogeneity.At the modality level, the dataset not only includes remote sensing imagery and structured annotations, but also provides high-quality image–semantic aligned data, enabling deep integration between visual content and semantic representation to support cross-modal learning.At the task level, the dataset systematically covers fundamental vision tasks such as object detection and semantic segmentation, and further extends to change detection, adverse weather image restoration (rain/snow/fog), and person re-identification (Re-ID), forming a unified cross-task data paradigm.The dataset is designed for multimodal large models and advanced remote sensing analysis, aiming to address key challenges in complex urban governance scenarios, including insufficient semantic understanding, weak cross-modal alignment, and limited generalization ability.

It provides a high-value data foundation for urban event understanding, intelligent perception, and decision-making, with strong application potential and rarity.