Unified Coastal Waste Dataset (UCWD): UAV Imagery with YOLO and COCO Annotations for Geospatial Analytics

Overview The Unified Coastal Waste Dataset (UCWD) provides a comprehensive collection of UAV imagery and high-quality annotations designed for automated coastal waste monitoring, computer vision benchmarks, and geospatial analytics. The dataset bridges the gap between disparate, fragmented public repositories by consolidating, filtering, and standardizing raw aerial imagery into a single, unified framework optimized for domain-specific remote sensing. The UCWD is openly available under a CC‑BY 4.0 license, allowing widespread reuse, adaptation, and contribution to scalable solutions for global marine debris management. Technical Specifications & Dataset Structure Total Size: $\approx$ 12.5 GB of high-resolution aerial imagery and annotations. Format Duality: To ensure broad cross-framework compatibility and reproducibility, UCWD is released simultaneously in YOLO and COCO formats. Each format branch contains the full image corpus along with its respective annotations ($\approx$ 6.25 GB per format branch). Framework Interoperability: The YOLO format branch is pre-configured for seamless training with state-of-the-art architectures like YOLOv8 and YOLOv11. The COCO JSON format branch enables immediate integration with established frameworks such as Faster R-CNN, Detectron2, and MMDetection. Data Splits: Structurally organized into standard Training, Validation, and Test splits with explicit directory layouts and accompanying metadata. Instance Count: Consolidates exactly 26,258 annotated instances mapped to a standardized taxonomy optimized for marine litter analysis. Raw Data Provenance & Sources The UCWD was constructed by harmonizing raw imagery from five primary public sources. While this dataset provides updated, cleaned, and re-standardized annotations to eliminate historical label noise, please credit and acknowledge the original data creators if you utilize this work: The Red Sea UAV Beach Litter Dataset: Martin, C., Parkes, S., Przeslawski, R., AlAsmari, S., & Duarte, C. M. (2021). Object detection benchmarks for Marine debris on Sandy beaches. Mendeley Data, v1. DOI: 10.17632/z7v6v8z6z3.1 UAVVaste Dataset: Kraft, M., Piechocki, M., Piechocka, A., Cortesi, F., & Valente, M. (2021). A dataset for debris detection in aerial imagery. Remote Sensing, 13(6), 1072. DOI: 10.3390/rs13061072 Marine Litter Dataset: Roboflow Universe (2025). Marine Litter Dataset (Version 6rmtv). Roboflow Public Workspace. URL: https://universe.roboflow.com/test-vivq6/marine-litter-6rmtv Aerial Beach Waste Dataset: National Cheng Kung University (2024). Aerial Beach Waste Dataset for Coastal Remote Sensing Evaluation. Roboflow Universe. URL: https://universe.roboflow.com/national-cheng-kung-university-wjot1/aerial-beach-waste-dataset-xpzsi Litter-myfyp Dataset: Litter-myfyp Project Group (2023). Litter Detection Dataset using Fine-Grained UAV Categorization. Roboflow Public Workspace. URL: https://universe.roboflow.com Intended Applications By providing multiple annotation formats and a transparent processing pipeline, UCWD facilitates reproducibility and encourages cross‑framework experimentation. It is intended to advance research across: Computer Vision & Remote Sensing: Benchmarking deep learning object detection networks on small, highly dense, or overlapping targets in variable coastal environments. Geospatial Analytics: Mapping and analyzing macro-plastic accumulation zones and pollution distribution patterns along coastlines. Environmental Policy: Providing data-driven foundations for automated monitoring systems used by practitioners and policymakers.