Beyond ClimateGAN

ClimateGAN's contributions are mostly around the data set and the Masker model. The "Painter" is but a GauGAN with small tweaks.

In this space, I explore how in-painting with Stable Diffusion as a Painter can turn ClimateGAN into ClimateDiffGAN (the Masker is still trained with a GAN approach to Domain Adaptation (AdvEnt / DADA)).

It runs in 30s on T4s which I pay for myself (it does not run on CPUs unfortunately) so this can't stay up for long... except for a community grant!