QTrack: Query-Driven Reasoning for Multi-modal MOT

TL;DR

QTrack introduces a novel query-driven multi-object tracking framework that localizes and tracks objects based on natural language queries, supported by a new benchmark and a multimodal reasoning model.

Contribution

The paper proposes a new paradigm for language-guided tracking, a large-scale benchmark RMOT26, and an end-to-end model QTrack with motion-aware reasoning strategies.

Findings

QTrack outperforms existing methods on the RMOT26 benchmark.

The approach effectively localizes and tracks objects based on natural language queries.

The Temporal Perception-Aware Policy improves reasoning accuracy.

Abstract

Multi-object tracking (MOT) has traditionally focused on estimating trajectories of all objects in a video, without selectively reasoning about user-specified targets under semantic instructions. In this work, we introduce a query-driven tracking paradigm that formulates tracking as a spatiotemporal reasoning problem conditioned on natural language queries. Given a reference frame, a video sequence, and a textual query, the goal is to localize and track only the target(s) specified in the query while maintaining temporal coherence and identity consistency. To support this setting, we construct RMOT26, a large-scale benchmark with grounded queries and sequence-level splits to prevent identity leakage and enable robust evaluation of generalization. We further present QTrack, an end-to-end vision-language model that integrates multimodal reasoning with tracking-oriented localization. Additionally, we introduce a Temporal Perception-Aware Policy Optimization strategy with structured rewards to encourage motion-aware reasoning. Extensive experiments demonstrate the effectiveness of our approach for reasoning-centric, language-guided tracking. Code and data are available at https://github.com/gaash-lab/QTrack