Beta model, might have issues.
Demo v0.2:
Demo v0.1:
If you like what I'm doing you can:
- Follow me on twitter
- Check my other projects GitHub
- You can sponsor me to support my open source work 💖 sponsor
Changelog
-
2022-09-010 Tested it on 64×64 -> 512×512 on FFHQ-CelebaHQ - Example Images
-
2022-08-026 Use this weights to train it further.
-
2022-08-026 Colab Demo v0.2:
-
2022-08-026 beta-v0.2
-
2022-08-023 beta-v0.1
-
2022-08-022 Colab demo v0.1
-
2022-08-022 Fast.ai model training url
64×64 -> 512×512
64×64 -> 512×512
Dataset was generated using the below models.
-GPEN
-FFHQ can be used, but generating a dataset would help in dealing with real image problems.
-Generated dataset was cropped to facial features and then trained.
Use this weights to train it further.
- works well with color photos.
- Model architecture: fastai v1 u-net
- Paired image2image training: fastai v1 superres notebook
- I call this a beta as it needs lot of improvement, But with minimum effort the results are good.