Pseudo-Pair based Self-Similarity Learning for Unsupervised Person Re-identification

TL;DR

This paper introduces a novel unsupervised person re-identification method that leverages pseudo pairs and self-similarity learning to improve local feature discrimination and cross-image patch alignment without human annotations.

Contribution

It proposes a pseudo-pair based self-similarity learning framework that constructs patch surrogate classes and updates pseudo labels during training for unsupervised re-ID.

Findings

Outperforms state-of-the-art methods on benchmark datasets.

Effectively learns local discriminative features via intra-similarity.

Accurately aligns patches across images using inter-similarity.

Abstract

Person re-identification (re-ID) is of great importance to video surveillance systems by estimating the similarity between a pair of cross-camera person shorts. Current methods for estimating such similarity require a large number of labeled samples for supervised training. In this paper, we present a pseudo-pair based self-similarity learning approach for unsupervised person re-ID without human annotations. Unlike conventional unsupervised re-ID methods that use pseudo labels based on global clustering, we construct patch surrogate classes as initial supervision, and propose to assign pseudo labels to images through the pairwise gradient-guided similarity separation. This can cluster images in pseudo pairs, and the pseudos can be updated during training. Based on pseudo pairs, we propose to improve the generalization of similarity function via a novel self-similarity learning:it learns local discriminative features from individual images via intra-similarity, and discovers the patch correspondence across images via inter-similarity. The intra-similarity learning is based on channel attention to detect diverse local features from an image. The inter-similarity learning employs a deformable convolution with a non-local block to align patches for cross-image similarity. Experimental results on several re-ID benchmark datasets demonstrate the superiority of the proposed method over the state-of-the-arts.