ROAM (Real-time Objective Animal Monitoring) Toolkit Labeled Video Dataset (video frames and labels for polar bears)

This is a condensed video dataset for the associated article, "Towards 24/7 behavioural monitoring: Automated real-time surveillance of animal behaviour from continuous video streams". The dataset is comprised of video frames and their associated behavioural labels from five behavioural categories that can be used to train models and test out functionalities (activity budgets, live on-screen monitoring, real-time alert system) associated with the ROAM toolkit. The annotated dataset (N = 8624 images with 13126annotated labels) for training the behavioural classification model in polar bears under managed care are comprised of 5 behavioural categories: Class 1: 1796 samples [RESTING]Class 2: 2251 samples [LOCOMOTION]Class 3: 2096 samples [SWIMMING]Class 4: 599 samples [HEAD SWINGING]Class 5: 6384 samples [FORAGING]*the dataset is imbalanced due to the inclusion of video frames that do not contain guests, protecting the confidentiality of guests of the accredited zoological where this work took place. The model, "best.pt", can be found in the associated GitHub repository linked below and reflects the object detection algorithm weights trained on the full dataset described in the associated article. Public video datasts and behaviour detection models for the Siberian Jay (Chan et al., 2025) and KABR (Kholiavchenkoet al., 2025) are cited below. names: ["RESTING", "LOCOMOTION", "SWIMMING", "HEAD SWINGING", "FORAGING"], Using the ROAM pipeline, bouts of stereotypical bouts of pacing and loop swimming are inferred using logic-based heuristics coupled with detections of locomotion or swimming, respectively. This dataset is licensed under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).Use permitted for academic and non-commercial research purposes only.Commercial use is prohibited without explicit permission from the authors. Additional information can be found on the GitHub repository for this project: https://github.com/lidunnchen/ROAM Siberian Jay Dataset: Chan, A.H.H., Putra, P., Schupp, H., Köchling, J., Straßheim, J., Renner, B., … & Kano, F., 2025. YOLO-Behaviour: A simple, flexible framework to automatically quantify animal behaviours from videos. Methods Ecol. Evol. 16, 760–774. https://doi.org/10.1111/2041-210X.14502 KABR (Kenyan Animal Behaviour Recognition) dataset: Kholiavchenko, M., Kline, J., Kukushkin, M., Brookes, O., Stevens, S., Duporge, I., ... & Stewart, C. V. (2025). Deep dive into kabr: a dataset for understanding ungulate behavior from in-situ drone video. Multimedia Tools and Applications, 84(21), 24563-24582. https://doi.org/10.1007/s11042-024-20512-4