0
$\begingroup$

In the Gan paper it is said page 3 Figure 1:

"The lower horizontal line is the domain from which z is sampled, in this case uniformly. The horizontal line above is part of the domain of x. The upward arrows show how the mapping x = G(z) imposes the non-uniform distribution pg on transformed samples"

For those who wants to see the figure: enter image description here

I wanted to know what that would mean in practical case. Let's say you are working with images that are normalized between [0,..,1] this would be the domain of x as referred in the paper right? Does this mean that I would have to sample my z from the domain of x, i.e: [0,..,1] ?

In most implementations I see people taking point randomly using things such as: np.random.randn(latent_dim)

$\endgroup$

1 Answer 1

0
$\begingroup$

Let's say you are working with images that are normalized between [0,..,1] this would be the domain of x as referred in the paper right?

No, the domain of X would be "images of [whatever they contain (e.g. dogs)] normalized between 0 and 1".

Does this mean that I would have to sample my z from the domain of x, i.e: [0,..,1] ?

No, they are both different domains and the generator $G$ maps between them.

In the paragraph you linked, the authors just point out that the generator $G$ is a function that maps the input data (i.e. random vectors following a uniform distribution in $[0, 1]$) to the output data (e.g. images of dogs) and that the mapping is non-regular, meaning that very different inputs may lead to similar outputs and vice versa.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.