StyleGAN2-ADA PyTorch for Grayscale Images

This repository is a fork & adaptation of

• NVlabs/stylegan2-ada-pytorch — the official NVIDIA implementation of StyleGAN2-ADA
• nhgowtham/StyleGAN2-ADA-pytorch-for-gray-images — an earlier grayscale adaptation

Why this repo?
The original NVlabs code only supports RGB (3-channel) inputs. Here we’ve modified it to work seamlessly on single-channel (grayscale) data by setting N_channel=1 throughout the network and data pipelines.

---

Acknowledgements

• Original Implementation:
  Karras, Tero et al. “Training Generative Adversarial Networks with Limited Data.” NeurIPS 2020.
  Code: github.com/NVlabs/stylegan2-ada-pytorch

Data & Checkpoints

All pretrained models, training checkpoints, and supporting data for this project are hosted on Google Drive. You can download them here:

• Checkpoint & Data Archive
  https://drive.google.com/drive/folders/10RQeaYxs6OtmvbKku3_Ljgusjtn4Ff7A?usp=drive_link

Note: After downloading, place the contents in ./checkpoints/ (or whatever path your training scripts expect), e.g.:

mkdir -p checkpoints
cp ~/Downloads/your_downloaded_files/* checkpoints/

Citation

If you use this dataset or code, please cite our Zenodo release:

@dataset{nimmal_haribabu_2025_15670394,
  author       = {Nimmal Haribabu, Gowtham},
  title        = {Exploring StyleGAN2-ADA for titanium alloy
                   microstructure generation
                  },
  month        = jun,
  year         = 2025,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.15670394},
  url          = {https://doi.org/10.5281/zenodo.15670394},
}