|
--- |
|
license: mit |
|
library_name: pytorch |
|
tags: |
|
- image-generation |
|
- gan |
|
- stylegan |
|
- stylegan3 |
|
- nvidia |
|
--- |
|
# Male Faces Generator (StyleGAN3 by NVIDIA) |
|
|
|
 |
|
|
|
This is a [StyleGAN3 PyTorch](https://github.com/NVlabs/stylegan3) model trained on 50k faces of men scraped from pinterest and its associated bias. |
|
|
|
### Usage |
|
|
|
Demo on Spaces is not yet implemented. |
|
|
|
If you want to generate your own images, follow the steps on [StyleGAN3 PyTorch](https://github.com/NVlabs/stylegan3) under "Getting started." Run gen_images.py and specify your seed and truncation. |
|
|
|
#### Sample Images |
|
Sample images were generated directly from model with no-postprocessing. Images were generated using a truncation of: '0.7' |
|
Recommend using [CodeFormer](https://github.com/sczhou/CodeFormer) to restore faces and upscale to desired resolution (I like upscaling by 2x to 512x512). |
|
|
|
### Dataset & Model Details |
|
|
|
Dataset was scraped from pinterest and cropped using dlib. Further symmetry and rotation filtering applied via U2Net and MTCNN. |
|
|
|
Training was done locally using StyleGAN3 with an RTX 4090 and cuda 11.8. Starting checkpoint was the ffhq 256x256 pre-trained model. Training took roughly 24 hours but hyperparameter tuning/restarting was modified about halfway through. Roughly at 400 ticks, dataset was filtered and reduced by half to images with greater symmetry. |
|
|
|
- Configuration: `stylegan3-r` |
|
- GPUs: `1` |
|
- Batch Size: `32` |
|
- Gamma: `6.6` |
|
- Final tick: `575` |
|
- Image Resolution: '256x256' |
|
- Final fid50k_full value (this pickle): `13.854398226827872` |
|
|
|
### Credits |
|
|
|
Please don't forget to give credit if you decide to share/distribute this model. Training these take a lot of time and effort :) |
