Abstract
As a crucial task in surveillance and security, person re-identification (re-ID) aims to identify the targeted pedestrians across multiple images captured by non-overlapping cameras. However, existing person re-ID solutions have two main challenges: the lack of pedestrian identification labels in the captured images, and domain shift issue between different domains. A generative adversarial networks (GAN)-based self-training framework with progressive augmentation (SPA) is proposed to obtain the robust features of the unlabeled data from the target domain, according to the preknowledge of the labeled data from the source domain. Specifically, the proposed framework consists of two stages: the style transfer stage (STrans), and self-training stage (STrain). First, the targeted data is complemented by a camera style transfer algorithm in the STrans stage, in which CycleGAN and Siamese Network are integrated to preserve the unsupervised self-similarity (the similarity of the same image between before and after transformation) and domain dissimilarity (the dissimilarity between a transferred source image and the targeted image). Second, clustering and classification are alternately applied to enhance the model performance progressively in the STrain stage, in which both global and local features of the target-domain images are obtained. Compared with the state-of-the-art methods, the proposed method achieves the competitive accuracy on two existing datasets.
Highlights
Person re-identification as a crucial task in surveillance and security strives to retrieve the same people across multiple images captured by non-overlapping cameras or across multi-scene images captured by the same camera
Despite the great success in person re-ID, some limitations still exist in practical applications, such as the acquisition of high-quality feature representation, the domain shift between training and testing data, and the difficulty of model migration from source domain to target domain
Existing person re-ID methods achieve high recognition rates on different types of single dataset, the great disparity exists between these person re-ID methods and practical applications, which is usually caused by the difference between the training and testing datasets [1,2,3,4,5,6,7,8,9]
Summary
Person re-identification (re-ID) as a crucial task in surveillance and security strives to retrieve the same people across multiple images captured by non-overlapping cameras or across multi-scene images captured by the same camera. Cross domain solutions ignore the global and local feature distribution of target-domain data, which is crucial for high-quality prediction. This paper proposes a GAN-based self-training framework with progressive augmentation (SPA) to solve the aforementioned two main challenges: the lack of labeled data and domain shift. Self-training stage (STrain): Clustering and classification are integrated to learn the robust features of the unlabeled target domain. As the progressive augmentation learning, both global and local features of the target-domain data are gradually enhanced by alternate clustering and classification. A two-stage (STrans and STrain) framework is proposed for unsupervised domain adaptive person re-ID, which can achieve good performance on both image style transformation and self-training. A progressive augmentation learning strategy integrates clustering and classification to obtain both global and local features of the target-domain data, and generates credible pseudo labels without any interventions.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
