Towards Understanding the Effect of Pretraining Label Granularity

Open in new window