A Re-ranking Method using K-nearest Weighted Fusion for Person Re-identification

Abstract

In person re-identification, re-ranking is a crucial step to enhance the overall accuracy by refining the initial ranking of retrieved results. Previous studies have mainly focused on features from single-view images, which can cause view bias and issues like pose variation, viewpoint changes, and occlusions. Using multi-view features to present a person can help reduce view bias. In this work, we present an efficient re-ranking method that generates multi-view features by aggregating neighbors' features using K-nearest Weighted Fusion (KWF) method. Specifically, we hypothesize that features extracted from re-identification models are highly similar when representing the same identity. Thus, we select K neighboring features in an unsupervised manner to generate multi-view features. Additionally, this study explores the weight selection strategies during feature aggregation, allowing us to identify an effective strategy. Our re-ranking approach does not require model fine-tuning or extra annotations, making it applicable to large-scale datasets. We evaluate our method on the person re-identification datasets Market1501, MSMT17, and Occluded-DukeMTMC. The results show that our method significantly improves Rank@1 and mAP when re-ranking the top M candidates from the initial ranking results. Specifically, compared to the initial results, our re-ranking method achieves improvements of 9.8%/22.0% in Rank@1 on the challenging datasets: MSMT17 and Occluded-DukeMTMC, respectively. Furthermore, our approach demonstrates substantial enhancements in computational efficiency compared to other re-ranking methods. Code is available at https://github.com/chequanghuy/Enhancing-Person-Re-Identification-via-UFFM-and-AMC.

Figures (9)

• Figure 1: Comparing the computational cost and Rank@1 performance of various reranking methods on the Market1501 and Occluded-DukeMTMC datasets. The y-axis shows Rank@1, the x-axis represents GPU memory usage, and the circle size indicates evaluation time, with larger circles representing longer evaluation times.
• Figure 2: Person re-identification datasets often include eight images of the same person taken from different viewpoints and cameras. The Market1501 and MSMT17 datasets are known for their diverse viewpoints, lighting conditions, and backgrounds. On the other hand, the Occluded-DukeMTMC dataset poses extra challenges with its complex environments and frequent partial occlusions, complicating the re-identification process.
• Figure 3: Our person identification pipeline consists of two stages. In the first stage, gallery images are ranked according to the cosine distance between the query and gallery images. In second stage, the top 5 candidates from the first stage are re-ranked using multi-view features.
• Figure 4: The process of generating multi-view features from single-view features: First, the single-view feature calculates the distance to all features in the gallery to find the K nearest neighbors. Then, the KWF method aggregates the neighboring features to generate a multi-view feature.
• Figure 5: The process involves transforming single-view features into multi-view features using the KWF method. In each figure, the red star represents the single-view feature, gray circles represent the nearest neighboring features (with K = 6), and the green quadrilateral represents the generated multi-view feature.