ECCV2018paper,Fine-grainedimagerecognition,proposeanovelself-supervisionmechanismtoeffectivelylocalizeinformativeregionswithouttheneedofbounding-box/partannotations