Dual-mimic learning for occluded pedestrian detection

Abstract It is still a great challenge to detect pedestrians in crowded scenes. Recent works mainly focus on enhancing the feature representation of visible region. Although this strategy effective, it fails fully exploit appearance information contained full-body To end, we propose Dual-Mimic learning model, which consists occluded unoccluded (Occ-Unocc) and (Full-Vis) mimic branches. Both branches can reduce intra-class variances by different mimicking strategies. Besides, define an occlusion attribute describe pattern each detection. Based attribute, Occlusion-aware NMS (Occ-NMS), effectively remain instance with low confidence. We conduct extensive experiments standard benchmarks CityPersons Caltech, our method achieves new state-of-the-art performance at 36.34% MR −2 25.07% heavy sets.