Do GAN's work well with small datasets?


I am currently working on creating synthetic data for electric poles. I have very less data in my training set. Do GAN’s work well with smaller datasets? Is there any alternative to GAN’s that work better with smaller datasets?

Hi, @vva21001!

I’m not particularly experienced in training GANs, but you can check several things like CycleGAN (Zhu, Park, Isola & Efros), differentiable augmentation (Zao et al.), or evolutionary training (Toutouh, Hemberg & O’Reilly).

Nevertheless, Diffusion Models (Dhariwal & Nichol) have superseded GANs for generative purposes. It may be worth giving them a try.


Thanks a lot @alvaroramajo. I will try using the diffusion models and the differentiable Augmentation. This was really helpful to me.

1 Like