This is an implementation of the fastai part1 v2 course in Keras lesson1 and lesson1-finetune2 jupyter notebooks go through the dogs and cats dataset
- lesson1 uses a finetune function that simply freezes early layers and makes the fc layer output predictions for 2 classes
- lesson1-finetune2 uses finetune2 function. It uses a few extra layers to enhance the model. It adds average and max pooling layers and concatenates them and then follows them with batchnorm, dropout, and dense layers. Note: that our function to find the optimal learning rate (LR_FIND) does not seem to work quite as well when we use finetune2.
Callback: LR_find