Person Reidentification by Deep Structured Prediction—A Fully Parameterized Approach

Abstract

Existing efforts on person reidentification (re-ID) either ignore the structural interactions among person images or require a highly crafted re-ID structure as a priori information. In contrast, our approach formulates person re-ID as a deep structured prediction problem that outperforms the state-of-the-art methods by utilizing neural-style-transfer-based structure sampling and fully parameterized energy networks.