JayjieL

hey, as u showed in paper's sec4.3: results fine-tuning on cls network always perform better . So, why don't use it in prediction. If not, why we train it.