'Smart_aug' objective is to make data augmentation differentiable, thus allowing to learn its parameters, with 'Higher' objects, such as 'Data_aug' classes. The meta-learning of the data augmentation parameter is performed jointly with the training of the model. Thus it minimize the overhead compared to other data augmentation learning techniques.
Updated 2024-08-20 11:55:02 +02:00
Updated 2020-06-30 17:56:51 +02:00