20.03.2021 Views

Deep-Learning-with-PyTorch

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

364 CHAPTER 13 Using segmentation to find suspected nodules

to use a technique called upsampling, which takes an image of a given resolution and

produces an image of a higher resolution. Upsampling at its simplest just means

replacing each pixel with an N × N block of pixels, each with the same value as the

original input pixel. The possibilities only get more complex from there, with options

like linear interpolation and learned deconvolution.

13.3.1 The U-Net architecture

Before we end up diving down a rabbit hole of possible upsampling algorithms, let’s

get back to our goal for the chapter. Per figure 13.6, step 1 is to get familiar with a

foundational segmentation algorithm called U-Net.

The U-Net architecture is a design for a neural network that can produce pixelwise

output and that was invented for segmentation. As you can see from the highlight

in figure 13.6, a diagram of the U-Net architecture looks a bit like the letter U, which

explains the origins of the name. We also immediately see that it is quite a bit more

complicated than the mostly sequential structure of the classifiers we are familiar with.

We’ll see a more detailed version of the U-Net architecture shortly, in figure 13.7, and

learn exactly what each of those components is doing. Once we understand the model

architecture, we can work on training one to solve our segmentation task.

1. Segmentation

UNet

2. Update:

2a. Model

T/F

2b. Dataset

t

3. Results

2c. Training

Figure 13.6

The new model architecture for segmentation, that we will be working with

The U-Net architecture shown in figure 13.7 was an early breakthrough for image segmentation.

Let’s take a look and then walk through the architecture.

In this diagram, the boxes represent intermediate results and the arrows represent

operations between them. The U-shape of the architecture comes from the multiple

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!