---
license: unknown
datasets:
- ylecun/mnist
pipeline_tag: text-to-image
tags:
- Stylegan
- Conditional
---
# Model Card for conditional StyleGAN3-t-MNIST

<!-- Provide a quick summary of what the model is/does. -->
This model allows to generate MNIST-like images. The model is based on [StyleGAN3-t](https://github.com/NVlabs/stylegan3?tab=License-1-ov-file#readme) and used the MNIST training split padded to 32x32. 
The prepared dataset can be found [here](https://github.com/oliverweissl/ConditionalStyleGANDatasets).

The model was trained using the following parameters:

- gamma: 0.125
- kimg: 5000
- cbase: 16384
- batch: 128
- metrics: fid50k (3.623 final)

This training took ~12h on 2 Nvidia A100 GPU

![fake numbers](./fakes.png "Generated Samples")