Description based person retrieval using outfit characteristics and human gender

Nowadays, video surveillance is widely deployed for security. Additionally, they serve for person retrieval. However, person retrieval manually can take long times. Thus, person retrieval using deep learning methods is analyzed in the research community. So we submit a person retrieval system using outfit characteristics and human gender. Our system includes YOLACT++ for clothes segmentation, YOLOR for human detection, and Efficientnet-B2 for clothes types and color classification. Experimental results are demonstrated on customized dataset and achieve an accuracy of 80.65% for the overall system.