Transfer Learning for Image Classification — (6) Build and Fine-tune the Transfer Learning Model

Figure (A): Transfer Learning (image by author)

We are ready to apply the transfer learning technique. This chapter will show you how to get the core part from VGG-16 and build a new neural network. You will learn the procedure to build a transfer learning model. It is the same as building any neural network. I will train the model on the car damage images. I will show you how to measure the performance. Finally, I will also show you how to fine-tune a model by opening up more layers. The Python notebook is available via this link.

(A) The VGG-16 Cannot Recognize the Car Damage Images

How good is VGG-16 in recognizing the images of car damages? Our images are of three classes of damages: “car front bumper damage”, “car rear bumper damage”, and “car side damage”. I will feed the model with car damage images. I hope the model can recognize them.