Learning TensorFlow.js

Chapter 7. Model-Making Resources

“By seeking and blundering we learn.”

—Johann Wolfgang von Goethe

You’re not limited to the models from TensorFlow Hub. Every day there are new and exciting models being tweeted, published, and highlighted in the community spotlight. These models and ideas are shared outside the Google-approved hubs, and sometimes they are even outside the realm of TensorFlow.js.

You’re starting to advance beyond the garden walls and work with models and data in the wild. This chapter is specifically geared toward arming you with new ways to make models from existing models, as well as charge you with the challenges of gathering and understanding data.

We will:

• Introduce model conversion

• Introduce Teachable Machine

• Train a computer vision model

• Review where training data comes from

• Cover some key concepts of training

When you finish this chapter, you’ll be armed with a few ways of making a model and a better understanding of the process of using data to make a machine learning solution.

Out-of-Network Model Shopping

TensorFlow.js hasn’t been around for long. Consequently, the number of models available is limited, or at least there are fewer than other frameworks. That doesn’t mean you’re out of luck. You can often convert models that were trained on other frameworks to TensorFlow.js. Converting existing models to make new models that work in a new environment is a great way to find recently developed resources and create exciting and contemporary ...