waifuGAN

This is a waifuGAN, a DCGAN network trained to generate anime faces, implemented in PyTorch with data taken from zerochan.net, shuushuu, and cropped using OpenCV. Read the full blog posts here: August 2021, January 2022.

Example Images

Here are some handpicked examples of the newly trained and updated model (1/9/2022):

Changelog:

5/23/2022 - Reorganization of project files + some docs. Overall just tuning up some legacy code 😆

1/9/2022 - Uploaded new model and updated blog with new info.

9/27/2021 - Moved entire README into a blog post onto my GitHub Pages. Check it out above.

Old Notes:

1. Check out blog post for links to dataset.
2. Models are in the models folder.
3. Plans for a webtool are coming soon, going to use ONNX.js.
4. Additionally, I scraped a ton more data from e-shuushuu.net, another free repository of data. Going to train again, and with more epochs (whenever I can find time to let my laptop overheat for a bit, busy college student rn). Update (1/9/2022): Done!