- CycleGAN-based image-to-image translation trained on 10,000+ artwork and painting images with TensorFlow and PyTorch.
- Reduced Fréchet Inception Distance from 58.4 to 49.6 (~15%) through preprocessing, normalisation and hyperparameter tuning.
- Evaluated with FID and visual assessment for texture transfer and style preservation.
Painting Generation with CycleGAN
Unpaired image-to-image translation that turns real photographs into landscape paintings.
