a typical implementation of classification task
This is a typical implementation of vgg16 on cifar100 including data preprocessing based on tf.layers api (architecture: VGG convolutions + 512 dense + 0.5 dropout) with more regulariers and better models we may reach higher
the model could be trained from scratch to reach arround 60% test accuracy
data loader should be later extracted to be writen in yield format to allow data loading and training at the same time