Implicit Motion State Modeling for Efficient and Effective Video-Level Object Tracking

Xianxin Jia

doi:10.1016/j.eswa.2026.132083

Implicit Motion State Modeling for Efficient and Effective Video-Level Object Tracking

2026-03-18

0

PRE

AI

DOI：10.1016/j.eswa.2026.132083

Original

Original request for help

Save

Abstract

En 中文

• Introduce an implicit motion state that replaces discrete update mechanisms. • Develop IMSETrack, a unified tracker combining implicit motion and spatial cues. • Use xLSTM to maintain a compact recurrent state capturing continuous motion dynamics. • Achieve 75.2% AUC on LaSOT with 69.6M parameters, 44 GFLOPs, and 55 FPS speed. • Demonstrate strong robustness and generalization across eight challenging benchmarks.

Keywords:

Implicit Motion State

Video Object Tracking

xLSTM

Motion Dynamics

Efficient Tracking

Implicit Motion State Modeling for Efficient and Effective Video-Level Object Tracking

Abstract

Journal

Organization