Persistent animal identification leveraging non-visual markers: Tracking and Identification Dataset - Detections Subset (TIDe-D)

This is the Detections subset of the Tracking and Identification Dataset (TIDe) as described in our paper "Persistent Animal Identification Leveraging Non-Visual Markers" [1] and used in the PhD Thesis "Automated Identification and Behaviour Classification for Modelling Social Dynamics in Group-Housed Mice" 2023 [2]. It grew out of a collaboration with the Mary Lyon Centre at MRC Harwell, with the need to automatically detect mice in single-channel infra-red videos of the home-cage. The challenge lies in the level of occlusion due to the group-housed mice in the enriched home-cage. We provide herein an annotated dataset containing video-frames (as jpeg images) and annotations of mouse and tunnel detections (in CoCo format). The dataset can be used to train and evaluate object detectors. Further details are available at https://github.com/michael-camilleri/TIDe. [1] M. P. J. Camilleri, L. Zhang, R. S. Bains, A. Zisserman, and C. K. I. Williams, “Persistent Animal Identification Leveraging Non-Visual Markers,” CoRR, vol. cs.CV, no. 2112.06809, Dec. 2021. [2] M. P. J. Camilleri, “Automated Identification and Behaviour Classification for Modelling Social Dynamics in Group-Housed Mice,” PhD Thesis, University of Edinburgh, 2023. It is aimed at training and evaluating mouse detectors. Further details are provided at https://github.com/michael-camilleri/TIDe.