Unsupervised learning of local features for person re-identification with loss function
by Lunzheng Tan; Guoluan Chen; Rui Ding; Limin Xia
International Journal of Autonomous and Adaptive Communications Systems (IJAACS), Vol. 16, No. 6, 2023

Abstract: Many methods for person re-identification focus on making full use of local features, which typically requires either comprehensive manual labelling or complex pretreatment. This paper proposes a novel loss function, termed feature channels dropout and de-similarity loss that drives the autonomous learning of discriminative local features in convolutional neural networks. The proposed loss function consists of two components. The first is a feature channels dropout component designed to compel each feature channel to be discriminative. A novel channel-dropout function and a cross-channel-element-max function are applied in this component. The second component is a de-similarity component that uses the Pearson correlation coefficient to constrain feature channels and ensure they differ from each other. This component is conducive to diverse local features in mining. Extensive experiments on three large-scale re-identification datasets demonstrate that feature channels dropout and de-similarity loss achieve superior performance compared with state-of-the-art methods.

Online publication date: Tue, 14-Nov-2023

The full text of this article is only available to individual subscribers or to users at subscribing institutions.

 
Existing subscribers:
Go to Inderscience Online Journals to access the Full Text of this article.

Pay per view:
If you are not a subscriber and you just want to read the full contents of this article, buy online access here.

Complimentary Subscribers, Editors or Members of the Editorial Board of the International Journal of Autonomous and Adaptive Communications Systems (IJAACS):
Login with your Inderscience username and password:

Username:        Password:          Forgotten your password?

Want to subscribe?
A subscription gives you complete access to all articles in the current issue, as well as to all articles in the previous three years (where applicable). See our Orders page to subscribe.

If you still need assistance, please email subs@inderscience.com