VLA-ReID: Video-Level Association for Re-Identification in Multi-Object Tracking with Highly Similar Objects

2026-07-19

AI Summary

This work addresses the challenges of identity preservation and frequent identity switches in multi-object tracking, which are exacerbated by highly similar appearances and dense object distributions. To tackle these issues, the paper proposes a video-level association re-identification framework that formulates re-identification as a global trajectory matching task. By aggregating historical trajectory features with current detections for video-level association, the method enhances individual discriminability without requiring additional annotations. It further introduces two novel components: Frame-common Appearance Estimation (FCAE) and a Corresponding Appearance Suppression (CAS) mechanism. Evaluated on the BEE24 dataset, the approach achieves notable improvements, including a 1.1-point gain in HOTA, a 2.6-point increase in AssR, and a 28% reduction in identity switches, significantly outperforming existing state-of-the-art methods.

Abstract

Multi-object tracking (MOT) aims to localize multiple objects in videos while preserving their identities over time. Long-term identity preservation remains difficult when objects are small, densely distributed, and highly similar in appearance, as in bee swarm scenes. Existing trackers rely on re-identification (re-ID) models trained through single-instance assignment (instance-level querying). At inference, however, MOT requires global assignment between multiple trajectories and detections, corresponding to video-level querying. This training-inference mismatch can cause identity switches among visually similar objects. Existing approaches also often require substantial additional annotations to enhance appearance discrimination. We propose Video-Level Association re-ID (VLA-ReID), which reformulates re-ID as video-level association modeling. It uses aggregated historical trajectory features as queries and all current-frame detections as candidates, enabling direct optimization of their global association at each frame. In addition, Frame-Common Appearance Estimation (FCAE) estimates a common appearance direction from current-frame detections, while Common-Appearance Suppression (CAS) removes the corresponding component along this direction from trajectory and detection features. This amplifies discriminative differences among highly similar objects without additional annotations. Experiments on BEE24 show that VLA-ReID improves HOTA by 1.1, MOTA by 0.3, AssR by 2.6, AssA by 0.7, and IDF1 by 0.8 over state-of-the-art trackers, while reducing identity switches by 28%. These results demonstrate the effectiveness of video-level re-ID modeling for appearance-based association in MOT.

Problem

• multi-object tracking
• re-identification
• identity preservation
• highly similar objects
• training-inference mismatch

Innovation

• Video-Level Association
• Re-Identification
• Common-Appearance Suppression
• Multi-Object Tracking
• Appearance Discrimination