How to generate more random and different outputs?

[yingzwang]

I trained pix2pix on only one or two image pairs. My application is to generate watercolor paintings from pencil sketches #416 .
The GAN is supposed to learn a distribution conditioned on the input, not a deterministic mapping from input to output. But my output seems quite deterministic. Is it due to my limited training data? Or due to the limitation of pix2pix?
How can I generate more random and different output samples?

[junyanz]

See this post for more details.

[yingzwang]

Thanks @junyanz . Found the answer there. BicycleGAN is exactly what I’m looking for.
I want a pix2pix model which can produce different watercolor paintings from the same pencil sketch.
If I have some followup questions, should I ask in the BicycleGAN page?

[junyanz]

Yeah, please ask in the BicycleGAN page.