Jonathan Fly 👾さんのイラストまとめ

Probably my single favorite GauGAN 'upscaled' game. Getting closer to real time neural nets radically transforming old media like Zelda: Ocarina of Time.

Segmentation here is way off mechanically - this specific version would be unplayable - but an emulator could do it.

DiverseDepth predicts depth in a broader range of scenes, with more detail, than most depth models do in their best dataset. So they made a harder dataset.

DiverseDepth: Affine-invariant Depth Prediction Using Diverse Data

abs: https://t.co/wvy4FOzNsR

StyleGAN goodness: a full reimplementation, non-square image support, fun tools for visualizing, interacting, and animating with models. StyleGAN really can do more than just faces. Woah.

blog: https://t.co/3jkd6oAEgB
code: https://t.co/qfDe96vxlM

There's a new version of the #GauGAN demo. New model, new categories -- the output is so different it's hard to use old segmaps to even compare. Formerly-hidden categories like flowers work a lot better. I like to put them in the sky though. 🌺

https://t.co/2H3XYyZeza

Training the StyleGAN portrait model on Pikachu and oh my gosh... all the human portraits get creepy Pikachu eyes with just a few minutes training! Thanks @roadrunning01 for making this so easy to play with.