Deep Reinforcement Learning Attention Selection For Person Re-Identification

摘要

Identity alignment models assume precisely annotated images manually. Humanlabelling is unrealistic on large sized imagery data. Detection modelsintroduce varying amount of noise and hamper identity alignment performance. Inthis work, we propose to refine images by removing the undesired pixels. Thisis achieved by learning to eliminate less informative pixels in identityalignment. To this end, we formulate a method of automatically detecting andremoving identity class irrelevant pixels in auto-detected bounding boxes.Experiments validate the benefits of our model in improving identity alignment.