Anime-face-generator

Anime Face Generator

Task

The task is to train generative networks based on a diffusion probabilistic model for generating anime faces.

Dataset

The dataset of animes faces downloading from Crypko 71,287 images. Each image is 96x96 pixels in size. Below are some samples of anime faces from the dataset:

Results

Samples of generated anime faces after 1000 training iterations:

Samples of generated anime faces after 5000 training iterations:

Samples of generated anime faces after 10000 training iterations:

Samples of generated anime faces after 15000 training iterations:

Samples of generated anime faces after 20000 training iterations:

Acknowledgment

I an grateful to the Center for Information Services and High Performance Computing [Zentrum für Informationsdienste und Hochleistungsrechnen (ZIH)] at TU Dresden for providing its facilities for high throughput calculations.

Reference

1. J. Ho et al., “Denoising Diffusion Probabilistic Models,” arXiv:2006.11239, Dec. 2020, https://arxiv.org/abs/2006.11239
2. Github implementation: https://github.com/lucidrains/denoising-diffusion-pytorch

This site is open source. Improve this page.