Learning TensorFlow.js

Chapter 6. Advanced Models and UI

“It always seems impossible until it’s done.”

—Nelson Mandela

You have a baseline for understanding models. You’ve consumed and utilized models and even displayed the results in overlays. It might seem like the sky is the limit. However, you’ve already seen that models tend to return information in various and complex ways. For the Tic-Tac-Toe model, you wanted only one move, but it still returns all nine possible boxes, leaving some cleanup work for you before you could utilize the model’s output. As models get more complicated, this problem can compound. In this chapter, we will select a widespread and complex model type for object detection and work through the UI and concepts to give you a full sense of what kind of tasks might befall you.

Let’s review what your current workflow looks like. First, you select a model. Identify if it is a Layers or Graph model. Even if you didn’t have this information, you’d be able to figure it out by trying to load it one way or another.

Next, you’ll need to identify the inputs and outputs for the model—not just the shape, but what the data actually represents. You batch your data, call predict on the model, and the output is good to go, right?

Unfortunately, there’s just a little more you should know. Some of the latest and greatest models have significant differences from what you’ve come to expect. In many ways, they are far superior, and in other ways, they are more cumbersome. Don’t fret, because you’ve ...