Cross-view Asymmetric Metric Learning for Unsupervised Person Re-identification

Hong-Xing Yu, Ancong Wu, Wei-Shi Zheng

Sun Yat-Sen University, China

Abstract

While metric learning is important for Person reidentification (RE-ID), a significant problem in visual surveillance for cross-view pedestrian matching, existing metric models for RE-ID are mostly based on supervised learning that requires quantities of labeled samples in all pairs of camera views for training. However, this limits their scalabilities to realistic applications, in which a large amount of data over multiple disjoint camera views is available but not labelled. To overcome the problem, we propose unsupervised asymmetric metric learning for unsupervised RE-ID. Our model aims to learn an asymmetric metric, i.e., specific projection for each view, based on asymmetric clustering on cross-view person images. Our model finds a shared space where view-specific bias is alleviated and thus better matching performance can be achieved. Extensive experiments have been conducted on a baseline and five large-scale RE-ID datasets to demonstrate the effectiveness of the proposed model. Through the comparison, we show that our model works much more suitable for unsupervised RE-ID compared to classical unsupervised metric learning models. We also compare with existing unsupervised RE-ID methods, and our model outperforms them with notable margins. Specifically, we report the results on large-scale unlabelled RE-ID dataset, which is important but unfortunately
less concerned in literatures.


Figure 1. Illustration of view-specific interference/bias and our idea.

Images from different cameras suffer from view-specific interference, such as occlusions in Camera-1, dull illumination in Camera-2, and the change of viewpoints between them. These factors introduce bias in the original feature space, and therefore unsupervised re-identification is extremely challenging. Our model structures data by clustering and learns view-specific projections jointly, and thus finds a shared space where view-specific bias is alleviated and better performance can be achieved.


Figure 2. Illustration of how symmetric and asymmetric metric clustering structure data using our method for the unsupervised RE-ID problem.

One shape indicates samples from one camera view, while one color indicates samples of one person. (a) Original distribution. (b) distribution in the common space learned by symmetric metric clustering (c) distribution in the shared space learned by asymmetric metric clustering. In the common space learned by symmetric metric clustering, the bias is still significant. In contrast, in the shared space learned by asymmetric metric clustering, the bias is alleviated and thus the data is better characterized according to the identities of the persons.

The ExMarket Dataset

MATLAB code for constructing/evaluating the ExMarket dataset: ExMarket_Package.rar

(Due to some internal problems, we temporarily redirect the link to our github page.)

Project Demo

MATLAB code for demo of our model: DEMO.rar

If there is any problem, please feel free to contact us.

(Due to some internal problems, we temporarily redirect the link to our github page.)

Citation

Please kindly cite the our paper if our work is helpful in your research: