a multi-task network which learns a re-ID embedding and at the same time predicts pedestrian
attributes.
our work systematically investigates how person re-ID and attribute recognition benefit each other by a jointly learned network
In fact, many attributes usually co-occur for a person, and the correlations of attributes may be helpful to re-weight the prediction of each attribute. For example, the attributes “skirt and “handbag" are highly related to “female rather than male".
ReID survey
PA survey
Multi-task related
google scholar search list some significant cited