||Existing Re-Identification networks focus on retrieving manually cropped query images from manually cropped gallery images. Real-world applications include finding criminals, cross-camera target tracking, person activity analysis, etc. Processing with manually cropped images is not a fit for real-world applications. Person search focuses on retrieving a
query image from whole gallery images. This study handles pedestrian detection and person search jointly in a single convolutional neural network. In the original work, an Online Instance Matching loss is proposed to train the network effectively. To validate the approach a largescale benchmark dataset is collected (CHUK-SYSU) which contains 18,184 images, 8,432 identities, and 96, 143 pedestrian bounding boxes. The results showed the superiority of
Online Instance Matching Loss over conventional Softmax loss.