VoGle26 - Birds in Railway Environments / Vögel im Gleisumfeld

1. General description The dataset was used for the development of a workflow for the object detection and classification of birds in the railway environment, utilizing video and image data. The project is funded by the German Centre for Rail Traffic Research and the project consortium consists of Fraunhofer Institute for Computer Graphics Research and GICON Ecosystems GmbH. The dataset comprises image data paired with corresponding annotation data (provided as text files). The annotation data indicates which image data sets were utilized for training and validating the workflow. The image data was extracted and annotated from selected video recordings made between 2024 and 2025 at two locations in Mecklenburg-Western Pomerania, Germany using a total of three static cameras: Peene Bridge, Anklam (1 camera) Riekdahler Weg, Rostock (2 cameras) For object detection, only annotation with an object area of more than 512 pixels were taken into account. Annotations with an area of more than 1024 pixels were transferred to a separate classification dataset. 2. Description of image data Number of images: 975 Image format(s): JPEG & PNG Resolution(s): Variable Color space: RGB 3. Annotations for object detection Annotation format Format: COCO JSON Coordinate system: Absolute (pixels) Bounding box definition: (x_min, y_min, width, height) Classes / labels (List of all object classes, unique class IDs) Classification (The classes were grouped according to visual similarity and frequency of occurrence to enable a balanced classification.) Möwe / seagull Taube / pigeon Krähe / crow Star / starling Turmfalke / kestrel Sonstige (Vogel) / other bird 4. Data set structure Classification Train/Val split: 70%/30% distributed per class