The styleGAN that nobody asked for! I had a lot of fun learning about GANs and decided to train one of my own.
- The model was designed by NVIDIA: https://github.com/NVlabs/stylegan2-ada-pytorch
- I used flicker to scrape images of poodles and barely filtered the data, mostly for my own curiosity.
- The final model took 14 hours to train using my RTX 2060 NVIDIA gpu.
- Make sure you are using Python version 3.9.13 (I did this using a pyenv shell)
- Create a new pip enviroment
- git clone https://github.com/NVlabs/stylegan2-ada-pytorch to replace the empty folder named "stylegan2-ada-pytorch" in the repository
- Activate the enviroment and pip install requirements.txt
- Decompress the .pkl file holding the model. I had to split it into 11 zip files to upload to Github, but you can use 7zip file manager to combine and unzip the file.
- Open 7zip
- Select all parts of the .zip.001 to .zip.011
- Click extract
- Run demo.ipynb and have fun! You can change the seed values to generate new images of poodles.
