Neighbor-Based Feature and Index Enhancement for Person Re-Identification
Chao Yuan, Tianyi Zhang, Guanglin Niu
TL;DR
This work tackles the robustness of person re-identification by leveraging latent multi-order neighborhood information to enrich feature representations and computationally refine the distance-based ranking. It introduces two complementary modules: Dynamic Multi-Order Neighbor (DMON) Feature Enhancement, which aggregates first-, second-, and third-order neighborhood information with adaptive Gaussian weights, and Asymmetric Relationship Optimization (ARO), which independently refines cross-view and intra-gallery distances to produce a more discriminative similarity matrix. The approach, built on a ViT-TransReID baseline, achieves consistent improvements in Rank-1 accuracy and mAP across Market-1501, DukeMTMC, and MSMT17, and includes comprehensive ablations and parameter analyses demonstrating the contributions of each module. The method is scalable and adaptable to other Re-ID tasks, highlighting the practical impact of incorporating higher-order contextual cues and asymmetric distance optimization in retrieval systems.
Abstract
Person re-identification (Re-ID) aims to match the same pedestrian in a large gallery with different cameras and views. Enhancing the robustness of the extracted feature representations is a main challenge in Re-ID. Existing methods usually improve feature representation by improving model architecture, but most methods ignore the potential contextual information, which limits the effectiveness of feature representation and retrieval performance. Neighborhood information, especially the potential information of multi-order neighborhoods, can effectively enrich feature expression and improve retrieval accuracy, but this has not been fully explored in existing research. Therefore, we propose a novel model DMON-ARO that leverages latent neighborhood information to enhance both feature representation and index performance. Our approach is built on two complementary modules: Dynamic Multi-Order Neighbor Modeling (DMON) and Asymmetric Relationship Optimization (ARO). The DMON module dynamically aggregates multi-order neighbor relationships, allowing it to capture richer contextual information and enhance feature representation through adaptive neighborhood modeling. Meanwhile, ARO refines the distance matrix by optimizing query-to-gallery relationships, improving the index accuracy. Extensive experiments on three benchmark datasets demonstrate that our approach achieves performance improvements against baseline models, which illustrate the effectiveness of our model. Specifically, our model demonstrates improvements in Rank-1 accuracy and mAP. Moreover, this method can also be directly extended to other re-identification tasks.