Thansk for your work. I use your model to test data pitts 250k 。 top-1 88.2% top-5 95.4% top-10 96.7% Inconsistent with the paper。