Oriole: Thwarting Privacy against Trustworthy Deep Learning Models