Hyperbolic Hierarchical Alignment for Video-Based Visible-Infrared Person Re-Identification

Video-based visible-infrared person re-identification (VVI-ReID) aims to learn robust video-level representations under modality discrepancy. However, existing methods typically rely on Euclidean geometry, which is suboptimal for modeling the complex temporal dynamics within visible and infrared tracklets, as it inevitably distorts the intrinsic hierarchical structure inherent in diverse temporal variations (e.g., occlusion, pose). In this paper, we propose Hyperbolic Hierarchical Alignment (HHA), which unifies spatio-temporal modeling and cross-modality alignment on the Poincaré ball. HHA employs a Hyperbolic Hierarchical Spatio-Temporal Aggregator (HHSA) to organize time-varying cues into low-distortion hierarchical representations via Hyperbolic Geometry Interaction (HGI) and Dual-Geometry Fusion (DGF). Furthermore, we introduce Geometry-Aware Modality Alignment (GMA), which integrates Hyperbolic Modality Alignment (HMA) to couple modality centroids for geometric consistency and Hyperbolic Prototype Alignment (HPA) to anchor both modalities to shared identity prototypes for robust discrimination. Experiments on HITSZ-VCM and BUPTCampus demonstrate state-of-the-art performance. The code will be available at https://github.com/Visuang/HHA.

People can often recognize someone in a video by combining what they look like with how they move. This task becomes much harder when videos come from different types of cameras, such as regular visible-light cameras and infrared cameras used at night or in low-light conditions. The same person may look very different across these cameras, and their appearance can also change over time because of pose changes, occlusion, and movement. This paper introduces a new method for matching people across visible and infrared videos. Instead of treating a video as a flat collection of visual features, our method organizes changing appearance and motion cues in a hierarchical space, making it easier to describe complex changes over time. It also helps bring visible and infrared videos of the same person closer together, so the system can identify people more reliably across camera types. Our experiments show that this approach improves person matching on two challenging video datasets. This work may help build more reliable person retrieval systems for low-light, nighttime, or cross-camera scenarios.