Mastering Computer Vision with TensorFlow 2.x

Coding an hourglass model

The coding blocks of the hourglass model can be seen in the following diagram:

Let's take a moment to understand the preceding diagram since we will be coding it in the following section:

• The front module takes the image as input and the number of channels (the third dimension of each layer, with the first two dimensions being width and height).
• The image passes through the different layers of the front module and the final module is connected to the left half-block.
• The left half-block has four bottleneck convoluted blocks – f1, f2, f4, and f8 – each at 1, 1/2, 1/4, and 1/8 resolutions respectively. This should ...