SFSORT: A Computationally Efficient Real-Time Multi-Object Tracking System Leveraging Scene Features
This paper introduces SFSORT, a computationally efficient real-time multi-object tracking system that leverages scene features to enhance object-track association and improve track post-processing.
Self-Supervised Multi-Object Tracking with Robust Long-Distance Object Matching
The core message of this paper is that the novel concept of path consistency can be used as a reliable self-supervised signal to train a robust object matching model capable of associating objects over long temporal distances, enabling consistent tracking in the event of occlusion.
Improving Multi-Object Tracking Performance through Representation Alignment Contrastive Regularization
The core message of this work is to introduce a lightweight and detector-free module called Representation Alignment Module (RAM) that can effectively model spatio-temporal relationships and improve the performance of multi-object tracking algorithms through contrastive regularization based on representation alignment rules.
© 2024 by Linnk AI