NAVE	Networked Augmented Virtual Environment (NAVE) Group

Publication:Yuanzhe Yang, Xin Zhang, Qichuan Geng, Chengxiang Chu and Zhou Zhou. Multi-Task Feature Decomposition Based Marginal Distribution for Person Search[C]. 2022 IEEE International Conference on Multimedia and Expo (ICME), Taipei, Taiwan, 18-22 July 2022. (CCF-B conference) pdf          Person search is a composite task, aiming at locating and identifying a query person from uncropped images. It requires jointly solving Pedestrian Detection and Person Reidentification. One major challenge in person search is the contradictory goals of detection and re-identification. The model has to simultaneously model the universality and specificity of persons. In this paper, we propose a novel parameterfree approach called Feature Decomposition Person Search (FDPS) to separate various tasks. FDPS decomposes the ROI feature map to extract sub-features based on the marginal distribution for different tasks. Also, we find that the Online Instance Match loss pays imbalanced attention to positive and negative categories. We present a Balance Online Instance Match (BOIM) loss to enhance the contribution of negative categories during training. Our method achieves the state-ofthe-art performance in one-step methods on two prevailing benchmarks, with high efficiency. create by admin at 2022-10-31 19:14:54