2017/07/19 by M. Saquib Sarfraz, Arne Schumann, Sarfraz, M. Saquib +5 · 8 citations
Computer Science · #Anomaly Detection Techniques and Applications #Artificial intelligence #Computer Vision and Pattern Recognition (cs.CV) #Computer science #Data mining #End-to-end principle #Exploit #FOS: Computer and information sciences #Human Pose and Action Recognition #Inference #Machine learning #Pedestrian #Search engine indexing #Task (project management) #Video Surveillance and Tracking Methods #cs.CV
paper · pdf · doi:10.48550/arxiv.1707.06089
published in arXiv (Cornell University) (Cornell University) · accepted BMVC 2017
arxiv created 2017/07/19 · openalex publication_date 2017/07/19 · arxiv updated 2017/07/20 · openalex created_date 2025/10/10 · openalex updated_date 2026/08/05
Pedestrian attribute inference is a demanding problem in visual surveillance\nthat can facilitate person retrieval, search and indexing. To exploit semantic\nrelations between attributes, recent research treats it as a multi-label image\nclassification task. The visual cues hinting at attributes can be strongly\nlocalized and inference of person attributes such as hair, backpack, shorts,\netc., are highly dependent on the acquired view of the pedestrian. In this\npaper we assert this dependence in an end-to-end learning framework and show\nthat a view-sensitive attribute inference is able to learn better attribute\npredictions. Our proposed model jointly predicts the coarse pose (view) of the\npedestrian and learns specialized view-specific multi-label attribute\npredictions. We show in an extensive evaluation on three challenging datasets\n(PETA, RAP and WIDER) that our proposed end-to-end view-aware attribute\nprediction model provides competitive performance and improves on the published\nstate-of-the-art on these datasets.\n