ARC-DVS Coastal Benchmark - Test split

Test split from the dataset presented in the paper "A Holistic Comparison of Performance and Efficiency between Von Neumann and Neuromorphic Computing in a Real-World Application" This split is part of an aerial beach dataset that captured the dynamic activity of surfers, swimmers, and beach users from drone-mounted vision sensors. It offers synchronized data obtained through conventional frame-based and event-based cameras for multi-modal analysis of real-world coastal environments. Dataset structure The split is organized into four top-level folders, one per beach. Inside each beach folder there are three modality subfolders: classic/ — frame images in PNG format events/ — event-based images in PNG format tensors/ — event-based tensors in PyTorch .pt format Annotations Each modality folder contains a labels_detection/ subfolder. Within labels_detection/, ZIP files are provided; each ZIP contains an annotations.xml file (CVAT format) with the bounding boxes and class labels corresponding to its paired data file (.png or .pt).