Fine-tuning with Xception

The Xception network was made by the creator of Keras, François Chollet in Xception: Deep Learning with Depthwise Separable Convolutions (https://arxiv.org/abs/1610.02357). Xception is an extension of the Inception architecture, where the Inception modules are replaced with depthwise separable convolutions. In the previous recipe, we focused on training the top layers only while keeping the original Inception weights frozen during training. However, we can also choose to train all weights with a smaller learning rate. This is called fine-tuning. This technique can give the model a small performance boost by removing some biased weights from the original network. 

How to do it...

...