Abstract:Recent deep learning-based person re-identification (RE-ID) approaches mainly adopt supervised learning, by which the network is trained with labels and tested in the same dataset. It has drawn increasingly great amount of attentions and achieved significantly competitive results. However, supervised RE-ID requires a large amount of manual and unaffordable labels in the specific domain, retarding the large-scale applications. To solve this problem and make RE-ID more practical in the real-world scenarios, we propose a novel unsupervised RE-ID method named unified domanial learning (UDL). UDL is trained with data from a labeled source domain and an unlabeled target domain, and then tested in the target domain, which resembles real-world application requirements. Two main components of UDL are: (1) Inter domain transfer learning (IDTL) is proposed based on generative adversarial network (GAN). IDTL can align the data distributions of different datasets to alleviate the huge domain gap by transferring the source domain images into the style of the target domain. (2) Intra domain trait mining (IDTM) is put forward which further excavates the potential information in the unlabeled target dataset, i.e., camera information and appearance information. IDTM is composed of camera invariance learning (CIL) and appearance similarity learning (ASL). CIL is utilized to extract camera style information using GAN technique and ASL is put forward to learn the appearance information by optimizing feature distances. Extensive experimental results on several widely-used datasets, i.e., Market-1501, DukeMTMC-reID and CUHK03 effectively demonstrate that UDL outperforms state-of-the-art methods by a little margin.

Unsupervised person re-identification by hierarchical cluster and domain transfer

Unsupervised Person Re-Identification Based On Clustering And Domain-Invariant Network

Domain-Camera Adaptation for Unsupervised Person Re-Identification

Domain Camera Adaptation and Collaborative Multiple Feature Clustering for Unsupervised Person Re-ID

Dynamic Re-Weighting and Cross-Camera Learning for Unsupervised Person Re-Identification

Domain Adaptive Attention Learning for Unsupervised Person Re-Identification

Cross-domain Person Re-Identification by Hybrid Supervised and Unsupervised Learning

An Unsupervised Person Re-Identification Method Based on Heterogeneous Convolutional Neural Networks Ensemble

Unsupervised Horizontal Pyramid Similarity Learning for Cross-Domain Adaptive Person Re-Identification

Hierarchical Connectivity-Centered Clustering for Unsupervised Domain Adaptation on Person Re-Identification

A Novel Unsupervised Camera-aware Domain Adaptation Framework for Person Re-identification

Unsupervised Person Re-identification Via Cross-camera Similarity Exploration.

Style Transfer for Unsupervised Domain-Adaptive Person Re-Identification

Imitating Targets from All Sides: an Unsupervised Transfer Learning Method for Person Re-identification.

Threshold-Based Hierarchical Clustering for Person Re-Identification

A Camera Identity-guided Distribution Consistency Method for Unsupervised Multi-target Domain Person Re-identification

Unsupervised Person Re-identification Using Unified Domanial Learning

Towards Discriminative Representation Learning for Unsupervised Person Re-identification

Joint Visual and Temporal Consistency for Unsupervised Domain Adaptive Person Re-Identification

Distance Constraint Between Features for Unsupervised Domain Adaptive Person Re-Identification

Unsupervised Domain Adaptation for Person Re-identification Via Heterogeneous Graph Alignment.