|
--- |
|
license: mit |
|
library_name: pytorch |
|
tags: |
|
- image-generation |
|
- gan |
|
- stylegan |
|
- stylegan3 |
|
- nvidia |
|
--- |
|
|
|
# Anime Faces Generator (StyleGAN3 by NVIDIA) |
|
|
|
<img width="679" alt="Generated Faces" src="https://user-images.githubusercontent.com/35907066/161809457-e6467724-5942-4a89-b379-85ddfd6ac86c.png"> |
|
|
|
This is a [StyleGAN3 PyTorch](https://github.com/NVlabs/stylegan3) model trained on this [Anime Face Dataset](https://github.com/bchao1/Anime-Face-Dataset). |
|
|
|
### Usage |
|
|
|
Demo on Spaces is not yet implemented. |
|
|
|
You can run the model pickle file locally using the instructions in this generator-script-only subset of the StyleGAN3 repo: |
|
- https://github.com/venture-anime/stylegan3-anime-faces-generator |
|
|
|
### Dataset & Model Details |
|
|
|
The [Anime Face Dataset](https://github.com/bchao1/Anime-Face-Dataset) was created by Mckinsey666. |
|
|
|
Training was done in [Paperspace Gradient](https://gradient.run/) on a free `RTX-5000` instance with the following parameters: |
|
|
|
- Configuration: `stylegan3-t` |
|
- GPUs: `1` |
|
- Batch Size: `8` |
|
- Gamma: `0.5` |
|
- Final tick: `102` |
|
- Final fid50k_full value (this pickle): `9.26043547642206`. |
|
|
|
# Train your own StyleGAN3 on PaperSpace or Colab |
|
|
|
You can use the notebooks here for a ready-to-use training pipeline |
|
https://github.com/akiyamasho/stylegan3-training-notebook |
