A “realistic” generator (Stylegan2-ada) modified to run locally in-browser. Current version: ada-04704
- Network: StyleGAN2 (https://github.com/NVlabs/stylegan2) and StyleGAN2-ada (https://github.com/NVlabs/stylegan2-ada)
- Input image source: https://www.wga.hu/ (Paintings Only, custom scraper @ 2019-01-24, 31k images, resized and cropped to 512×512)
- Output images: 512×512 RGB
- Training duration: 2000k images using StyleGAN2, converted and 4700k images in StyleGAN2-ada.
- Online size: ~112 MByte (Please be considerate with my bandwidth)
- GPU/CPU RAM Memory size: Unkown
- Web-backend: ONNX (https://github.com/microsoft/onnxruntime/tree/master/js/web)
- See the blogpost “Running StyleGAN2-ada in browser” to see how to run these type of networks in the browser.
NOTE: This model is very slow! Generating images can take as much as a minute. I just switched to ONNX in browser. ONNX WebGL isn’t working (yet) for this model. It has only been trained for 6700k images so it is a long way from being done.
If you press the “Load Model” button you accept the Terms and Conditions!