I am facing the problem of overfitting on training data. I can easily get 100 % accuracy but only 60 % on test data. I have used l1, l2 regularisation, dropout and data augmentation but nothing is solving my problem. I am training a using pre-trained vGG 16 for image classification. Please help