この資料をSMS送信: Weakly supervised scale-invariant learning of models for visual recognition