Skip to main content

Video-Based Animal Re-Identification (VARe-ID) from Multiview Spatio-Temporal Track Clustering

This work presents a modular software pipeline and end-to-end workflow for video-based animal re-identification, which assigns consistent individual IDs by clustering multiview spatio-temporal tracks with minimal human intervention. Starting from raw video, the system detects and tracks animals, scores and selects informative left/right views, computes embeddings, clusters annotations by viewpoint, and then links clusters across time and varying perspectives using spatio-temporal continuity. Automated consistency checks resolve remaining ambiguities. Preliminary experiments demonstrate near-perfect identification accuracy with very limited manual verification. The workflow is designed to be generalizable across species. Currently, trained models support Grevy’s and Plains zebras, with plans to expand to a broader range of species.

GitHub Repo License

image

References​

Acknowledgements​

  • National Science Foundation (NSF) funded AI institute for Intelligent Cyberinfrastructure with Computational Learning in the Environment (ICICLE) (OAC 2112606).
  • Imageomics Institute (A New Frontier of Biological Information Powered by Knowledge-Guided Machine Learning) is funded by the US National Science Foundation's Harnessing the Data Revolution (HDR) program under Award (OAC 2118240).
  • Support from Rensselaer Polytechnic Institute (RPI).
  • Support from Finnish Cultural Foundation.
  • Resources from Ohio Supercomputer Center made it possible to train and test algorithmic components.

Citation​

Ankit K. Upadhyay, Ekaterina Nepovinnykh, S. M. Rayeed, Aidan Westphal, Lawrence Miao, Julian Bain, Jaeseok Kang, Tuomas Eerola, Heikki Kälviäinen, Charles V. Stewart. Animal Re-Identification via Multiview Spatio-Temporal Track Clustering. Rensselaer Polytechnic Institute, LUT University, Brno University of Technology, CV4Animals, CVPR 2025.