Stylegan Dehua

A selection of machine-learning generated Dehua porcelain heads.

StyleGAN2-ada-pytorch on Google Colab with single NVIDIA V100 GPU.  Converted legacy network FFHQ 512x512 StyleGAN2 config f, then using 500 images of faces from dehua sculptures ran transfer learning for 200kimg (default augmentation, x-mirror).

Previous
Previous

Machine Learning for Artists

Next
Next

Recreating Ancient Glazes