The size should vary based on the dataset. size=32 is the typical size used for CIFAR-10/100.
The scale is relatively less important. scale=(0.2, 1.0) should be a good default for different datasets. Of course, you can try a different lower/upper bound for parameter tuning.
The size should vary based on the dataset. size=32 is the typical size used for CIFAR-10/100. The scale is relatively less important. scale=(0.2, 1.0) should be a good default for different datasets. Of course, you can try a different lower/upper bound for parameter tuning.