UREPTrack: Unified RGB-Event Visual Tracking via PoolFormer Backbone

Abstract Visual Object Tracking (VOT) faces significant challenges under conditions such as fast motion, motion blur, and extreme illumination. RGB-only trackers often degrade in these scenarios, while event cameras provide microsecond latency and high dynamic range but lack rich spatial semantics. We introduce UREPTrack, a unified, single-stage, attention-free RGB-event tracker built on a lightweight PoolFormer backbone. Raw event data are voxelized into compact spatiotemporal tensors and, together with RGB template and search patches, embedded and concatenated into a single token stream processed by a shared backbone. A fully con- volutional head jointly predicts classification confidence, center offsets, and box size, eliminating the need for multi-branch Siamese pipelines and costly self-attention. UREPTrack achieves state-of-the-art performance, setting new benchmarks on COESOT (S 64.4, P 77.5, NP 76.2, BOC 23.7) at 170 FPS, VisEvent (S 55.46, SR0.5 67.01, SR0.75 46.96, P 71.58, NP 75.22), and FE108 (P 94.3, S 65.9). Ablation studies confirm (i) the complementarity of RGB and event modalities, (ii) the superiority of event voxelization over image-like alternatives, and (iii) favorable accuracy and effciency scaling across PoolFormer sizes. UREPTrack provides a practical, high-speed solution for real-time, multi-modal tracking on resource-constrained hardware. Our codes will be publicly released in https://github.com/HamadYA/UREPTrack.