AqME-RTDETR dataset

The VisDrone 2019 dataset is an authoritative resource for the international drone vision community. This dataset was collected by deploying multirotor drones in the airspace of 14 cities in China, covering various urban and rural environments, sparse to dense scenes, and lighting conditions (daytime and nighttime). It presents a wide range of diverse multi-scene and multi-task drone captures. It includes various objects across 10 categories, including pedestrians, vehicles, bicycles, and trucks. As shown in the figure below, most of the objects in the dataset have a pixel size smaller than 32 \u00d7 32 , classifying them as small objects, which presents significant challenges to the computational efficiency and scale adaptability of detection algorithms.For our experiments, we divide the images into three sets: a training set with 6,471 samples, a validation set with 548 samples, and a test set with 1,610 samples.