CoTracker: it is better to track together

Conference item

We introduce CoTracker, a transformer-based model that tracks a large number of 2D points in long video sequences. Differently from most existing approaches that track points independently, CoTracker tracks them jointly, accounting for their dependencies. We show that joint tracking significantly improves tracking accuracy and robustness, and allows CoTracker to track occluded points and points outside of the camera view. We also introduce several innovations for this class of trackers, including using token proxies that significantly improve memory efficiency and allow CoTracker to track 70k points jointly and simultaneously at inference on a single GPU. CoTracker is an online algorithm that operates causally on short windows. However, it is trained utilizing unrolled windows as a recurrent network, maintaining tracks for long periods of time even when points are occluded or leave the field of view. Quantitatively, CoTracker substantially outperforms prior trackers on standard point-tracking benchmarks. Code and model weights are available at https://co-tracker.github.io/.

Authors: Karaev, N; Rocco, I; Graham, B; Neverova, N; Vedaldi, A

• Publication status: Published
• Peer review status: Peer reviewed
• Publisher: Springer
• Host title: Computer Vision – ECCV 2024: 18th European Conference, Milan, Italy, September 29–October 4, 2024, Proceedings, Part LXII
• Pages: 18-35
• Series: Lecture Notes in Computer Science
• Series number: 15120
• Place of publication: Cham, Switzerland
• Publication date: 2024-10-31
• Acceptance date: 2024-07-01
• Event title: 18th European Conference on Computer Vision (ECCV 2024)
• Event location: Milan, Italy
• Event website: https://eccv.ecva.net/
• Event start date: 2024-09-29
• Event end date: 2024-10-04
• DOI: 10.1007/978-3-031-73033-7_2
• EISSN: 1611-3349
• ISSN: 0302-9743
• EISBN: 9783031730337
• ISBN: 9783031730320
• Language: English
• Keywords: oiptical flow; point tracking; motion estimation