In this chapter, we learned how to use TensorFlow/Keras to develop transfer learning code for deep learning models we studied in the last chapter. We learned how to import trained images from a directory consisting of multiple classes and use them to train the models and make predictions with them. We then learned how to keep the base layer of the model frozen, remove the top layer, and replace it with our own top layer and train the resultant model with it.
We studied the importance of visual search and how transfer learning can be used to augment visual search methods. Our example consists of furniture of three different classes—we understood the accuracy of the model and how to improve the resulting losses. In this chapter, we also learned how to use the TensorFlow tf.data input pipeline for faster image processing during training.
In the next chapter, we will...
