Region selection for occluded person re-identification via policy gradient

Bolei Xu

Abstract: Highlights•It is the first paper to adopt reinforcement learning approach to address the occluded person re-identification problem.•A novel region selection learning strategy is proposed to remove irrelevant and noisy image regions.•A triplet-ViT network is proposed to learn discriminate features to distinguish non-target person occlusion.•Significant results were achieved for addressing the occluded person re-identification task.