Cityscape Human Detection

The CityscapesSubset dataset comprises 200 annotated images derived from the Cityscapes validation set [1,2], designed for evaluating multi-class pedestrian and vehicle detection in urban environments. It includes 140 training, 40 validation, and 20 test images, each with a resolution of 2048 \u00d7 1024 pixels, capturing diverse scenes with occlusions, varying scales, and cluttered backgrounds. Annotations are provided in YOLO format for two classes: Pedestrian (class_id 0) and Vehicle (class_id 1), addressing challenges in multi-class robustness and false positive susceptibility, as explored in the IEEE Access manuscript *\Pedestrian Detection with YOLOv5 and GhostNet for Edge Device Deployment in Dynamic Environments\* [3]. This subset is ideal for training and testing lightweight object detection models on resource-constrained devices, such as those used in autonomous driving and smart surveillance. Users are encouraged to combine with the full Cityscapes dataset for broader multi-class analysis.[1] M. Cordts et al., \u201cThe Cityscapes Dataset for Semantic Urban Scene Understanding,\u201d CVPR, 2016.  [2] S. Zhang et al., \u201cCityPersons: A Diverse Dataset for Pedestrian Detection,\u201d CVPR, 2017.  [3] N. Salam, J. T. Jebaseeli, A. D. Durai C, \u201cPedestrian Detection with YOLOv5 and GhostNet for Edge Device Deployment in Dynamic Environments,\u201d IEEE Access, 2025.