Homemade Dataset for Drones

The UAV dataset used in this study is specifically designed for the detection of small, densely distributed, and occluded objects in UAV imagery. This dataset contains high-resolution images annotated with small object labels, including objects such as vehicles, pedestrians, and other relevant targets commonly found in UAV-captured environments. The dataset includes projection-based annotations, enabling accurate localization and size estimation of small objects in real-world metric units. These annotations make it particularly useful for metric-based evaluation of object detection models, providing valuable data for evaluating performance in practical, real-world settings. This dataset was used to train and evaluate SSDM-YOLO, a lightweight framework that enhances semantic representation and robustness under varying visual conditions. The dataset's focus on small object detection in UAV imagery allows researchers to benchmark their methods and algorithms, particularly for applications such as UAV-based traffic monitoring, environmental analysis, and visual perception tasks. The dataset is publicly available for research purposes, and researchers are encouraged to use it to test models for small-object detection and to explore further improvements in UAV-based vision systems.