grddavies / stylegan2-ada

StyleGAN2 with adaptive discriminator augmentation (ADA) - Official TensorFlow implementation

Ey up (hello), I'm Gethin

About me

• 🏢 I'm currently working on a browser-based DICOM viewer at Motilent
• 🐣 I’m learning Rust 🦀 and Haskell 💜 when I have the time
• 🎨 I occasionally make generative art (have a look here) and music
• 📖 I'm reading The Book of Trespass by Nick Hayes
• 🦜 Hablo un poco de Español, e também falo Português... mas não muito.
• 💬 I like talking about (but claim no expertise in) programming language theory, time-series, synthesizers, music and other wiggly lines 📈

GitHub Stats

Technologies I find useful or am interested by...

VSCode	NeoVim	TypeScript	React	Python	R
Rust	WASM	Haskell	Solid JS	PyTorch	Ableton Live

Generative Art

CW // Unreality

Surreal cityscapes generated using styleGAN.

Network trained on aerial photographs scraped from Flickr. Animations generated by circilar interpolation of the constant layer input weights, and through the latent space. For details see here

Special effect glaze textures using styleGAN2

Network trained on photographs of ceramic glazes (DWIGn dataset). Animations generated by circilar interpolation of the constant layer input weights, and through the latent space.

Blended networks

Network produced by splicing layers from a network trained on the FFHQ (faces) dataset, and my own DWIGn (ceramic glazes) dataset. Low resolution layers are from the FFHQ network, retaining some facial features. Textures and colours are generated in higher resolution layers from the DWIGn dataset. Animations generated by circilar interpolation of the constant layer input weights, and through the latent space.