Model Conversion Using Model Optimizer

We have already discussed in detail why we need OpenVINO's optimization power. Now, let's understand how Model Optimizer helps with that. Generally speaking, OpenVINO's optimization takes place in two steps: one at the Model Optimizer stage and one at the Inference Engine stage. We will investigate the Inference Engine in the next section, but for now, let's focus on Model Optimizer.

Model Optimizer carries out two main operations – quantization and fusion.

Let's start by understanding quantization. As you may recall, we talked about precision in Exercise 8.01, Downloading the Pedestrian and Vehicle Detection Model, when we said that FP16 means that the number representation is going to use 16 bits of memory and will be represented in floating-point format. Quantization is closely related to that. To understand it, let's look at an example. Let's say we want to represent a number, 0.578156, in a...