We want to perform transfer learning on a StyleGAN, or just train a StyleGAN from scratch; however, the StyleGAN implementation is in TFv1.14 (presumably for backwards compatibility with other NVIDIA code bases), is not linted or tested, and is not readily usable in another context. Add an architecture module in mlops that contains tested reimplementations of useful architectures like StyleGAN, so that we as an ML community can have modular code.
We want to perform transfer learning on a StyleGAN, or just train a StyleGAN from scratch; however, the StyleGAN implementation is in TFv1.14 (presumably for backwards compatibility with other NVIDIA code bases), is not linted or tested, and is not readily usable in another context. Add an architecture module in
mlops
that contains tested reimplementations of useful architectures like StyleGAN, so that we as an ML community can have modular code.