Abstract
Person search is a challenging computer vision task that handles and optimizes both pedestrian detection and person re-identification simultaneously. Person search is also closer to real-world applications compared to person re-identification. Existing person search works mainly focused on refining loss functions, using more complex network structures or redefining the person search as another task. However, few of them attempted to solve this problem from a feature representation perspective. In this paper, we embark on this point and present a novel method called FLAG to learn a better feature representation for person search. Specifically, partition pooling and cross-level feature hybridization are proposed to guide the model to learn more discriminative person features. Experiments show that the proposed method achieves encouraging performance improvement and outperforms similar end-to-end person search methods.
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have