Training a 3D-GAN is similar to training a vanilla GAN. The steps involved in the training of a 3D-GAN are as follows:

1.  Sample a 200-dimensional noise vector from a Gaussian (normal) distribution.
2. Generate a fake image using the generator model.
3. Train the generator network on real images (sampled from real data) and on the fake images generated by the generator network.
4. Use the adversarial model to train the generator model. Don't train the discriminator model.
5. Repeat these steps for a specified number of epochs.

We will explore these steps in detail in a later section. Let's move on to setting up a project.