Experimenting with Seeds

So far, we’ve thought of the seed that feeds the generator as just a random number. After a GAN has been trained, that seed can gain some interesting properties. Let’s see for ourselves.

Approach 1: Choosing values between $\;seed\; 1\;$ and $\;seed\;2$

Imagine two different seeds, $seed\;1$ and $seed\;2$. We can generate images from both of these seeds. Now imagine a seed that was half way between $seed\;1$ and $seed\;2$. What would the generated image look like? In fact, what would images look like for seeds at regular intervals between $seed\;1$ and $seed\;2$?