Learning to Disentangle Scenes for Person Re-identification
The proposed method employs a divide-and-conquer strategy for the ReID task. Concretely, two self-supervision operations are used to generate new images with the characteristics of occluded and scale variation scenes.
BibTex:
