Synthetic flood debris dataset (Flood-SHAB)

This repository contains a synthetic image dataset of a floating-debris for bridge and near-pier water-surface scenes. The dataset is designed for object-detection training (debris vs. no-debris) and for validating robustness under diverse bridge geometries, camera viewpoints, and water-surface conditions. All images are synthetic (AI-generated) and provided as JPG files. Object-detection annotations are provided in YOLO format. The dataset is synthetic; domain shift is expected. Dataset size and splitTotal images: 1,400- Images with debris (positive): 700- Images without debris (negative): 700 Splits (balanced 50/50):- train: 1,120 images (560 positive, 560 negative)- valid: 280 images (140 positive, 140 negative) Directory structureThe dataset follows a YOLO-style split: dataset.zip/ train/ images/ *.jpg labels/ *.txt valid/ images/ *.jpg labels/ *.txt Notes:- Each image have a corresponding label file in YOLO TXT format with the same filename stem.- Images without debris have empty label file.- Each line: <class_id> <x_center> <y_center> <width> <height>- All coordinates are normalised to [0, 1] relative to image width/height. Example:0 0.28125 0.615625 0.528125 0.4where class_id=0 (debris), x_center=0.28125, y_center=0.615625, width=0.528125, height=0.4. Definition of “debris” in this dataset:- Visible floating debris on or near the water surface that is operationally relevant to near-pier accumulation monitoring.- Bridge structure, shoreline vegetation, and unrelated background clutter are not annotated as debris.- Very small ambiguous artifacts (noise-like patterns) are generally not annotated. Ownership and redistribution:- All images are synthetic and were created using AI image generation tools. Based on the service terms reviewed at the time of dataset preparation, the dataset creator retains ownership of the input prompts and generated outputs. Attribution:This dataset includes images generated using multiple tools (Grok, OpenAI tools (e.g., Sora and ChatGPT image generation), and Canva AI). Grok requests attribution; accordingly, when redistributing or using the dataset publicly, please include the statement “Created by Grok”. This attribution requirement should be interpreted as applying to the subset of images generated with Grok and does not imply that every individual image was generated by Grok, nor does it imply endorsement by any tool provider. Citation: If you use this dataset, please cite this dataset DOI. In addition, please cite the associated article once published (bibliographic details will be added to a future version of this record). FundingThis work has received funding from the European Union’s Horizon Europe research and innovation programme under the Marie Skłodowska-Curie grant agreement No. 101154316. Views and opinions expressed are, however, those of the authors only and do not necessarily reflect those of the European Union or the European Research Executive Agency (REA). Neither the European Union nor REA can be held responsible for them.