Object-centric multiple object tracking

Zhao Z, Wang J, Horn M, Ding Y, He T, Bai Z, Zietlow D, Carl-Johann Simon-Gabriel C-JS-G, Shuai B, Tu Z, Brox T, Schiele B, Fu Y, Locatello F, Zhang Z, Xiao T. Object-centric multiple object tracking. arXiv, 2309.00233.

Download (ext.)

Preprint | Submitted | English
Author
Zhao, Zixu; Wang, Jiaze; Horn, Max; Ding, Yizhuo; He, Tong; Bai, Zechen; Zietlow, Dominik; Carl-Johann Simon-Gabriel, Carl-Johann Simon-Gabriel; Shuai, Bing; Tu, Zhuowen; Brox, Thomas; Schiele, Bernt
All
Department
Abstract
Unsupervised object-centric learning methods allow the partitioning of scenes into entities without additional localization information and are excellent candidates for reducing the annotation burden of multiple-object tracking (MOT) pipelines. Unfortunately, they lack two key properties: objects are often split into parts and are not consistently tracked over time. In fact, state-of-the-art models achieve pixel-level accuracy and temporal consistency by relying on supervised object detection with additional ID labels for the association through time. This paper proposes a video object-centric model for MOT. It consists of an index-merge module that adapts the object-centric slots into detection outputs and an object memory module that builds complete object prototypes to handle occlusions. Benefited from object-centric learning, we only require sparse detection labels (0%-6.25%) for object localization and feature binding. Relying on our self-supervised Expectation-Maximization-inspired loss for object association, our approach requires no ID labels. Our experiments significantly narrow the gap between the existing object-centric model and the fully supervised state-of-the-art and outperform several unsupervised trackers.
Publishing Year
Date Published
2023-09-01
Journal Title
arXiv
Article Number
2309.00233
IST-REx-ID

Cite this

Zhao Z, Wang J, Horn M, et al. Object-centric multiple object tracking. arXiv. doi:10.48550/arXiv.2309.00233
Zhao, Z., Wang, J., Horn, M., Ding, Y., He, T., Bai, Z., … Xiao, T. (n.d.). Object-centric multiple object tracking. arXiv. https://doi.org/10.48550/arXiv.2309.00233
Zhao, Zixu, Jiaze Wang, Max Horn, Yizhuo Ding, Tong He, Zechen Bai, Dominik Zietlow, et al. “Object-Centric Multiple Object Tracking.” ArXiv, n.d. https://doi.org/10.48550/arXiv.2309.00233.
Z. Zhao et al., “Object-centric multiple object tracking,” arXiv. .
Zhao Z, Wang J, Horn M, Ding Y, He T, Bai Z, Zietlow D, Carl-Johann Simon-Gabriel C-JS-G, Shuai B, Tu Z, Brox T, Schiele B, Fu Y, Locatello F, Zhang Z, Xiao T. Object-centric multiple object tracking. arXiv, 2309.00233.
Zhao, Zixu, et al. “Object-Centric Multiple Object Tracking.” ArXiv, 2309.00233, doi:10.48550/arXiv.2309.00233.
All files available under the following license(s):
Copyright Statement:
This Item is protected by copyright and/or related rights. [...]

Link(s) to Main File(s)
Access Level
OA Open Access

Export

Marked Publications

Open Data ISTA Research Explorer

Sources

arXiv 2309.00233

Search this title in

Google Scholar