Transfer Learning for Image Classification — (5) Get Image Data, Ready, and Go
Now you are about to build your transfer learning image model. What’s your topic? Maybe your model is to recognize any international cuisines, any Lego Minifigures or any actors in movies. You will need the data to build your model. Many enterprises can buy images in bulk and pay for image labeling services. Suppose you just want a few hundred images to build a proof-of-concept model. How do you even obtain hundreds of images to start with? How do you annotate images? How do you structure images for modeling? These obstacles are very practical. That’s why I’ve prepared this chapter. In this chapter, I present a hypothetical data science team who will solve an industry problem. After seeing their work, you will be able to build a model for any topic of your choice. The Python notebook is available via this link.
(A) The Hypothetical Data Science Team
Every year auto insurance companies receive millions of auto damage claims. When an auto insurer receives a claim, a representative needs to go on-site to assess the damages. Physical inspection can take a day or even a week if it is in a remote area. Often damages are minor or partial damages. The insurance company could have saved time and money if they just received photos of the damaged car. They can use an image classification model to…
