nupurkmr9/vision-aided-gan
GAN training method that ensembles pretrained computer vision models as discriminator components to improve image generation quality.
Not currently ranked — collecting fresh signals.
star history
This repository implements a technique for improving GAN training by leveraging knowledge from pretrained vision models. The approach selects the most accurate vision models by probing linear separability between real and fake samples in their embeddings, then progressively adds them to a discriminator ensemble. The method has been extended to work with BigGAN and StyleGAN3 architectures, achieving state-of-the-art results on LSUN benchmarks.
Frequently asked
- What is nupurkmr9/vision-aided-gan?
- GAN training method that ensembles pretrained computer vision models as discriminator components to improve image generation quality.
- Is vision-aided-gan open source?
- Yes — nupurkmr9/vision-aided-gan is open source, released under the MIT license.
- What language is vision-aided-gan written in?
- nupurkmr9/vision-aided-gan is primarily written in Python.
- How popular is vision-aided-gan?
- nupurkmr9/vision-aided-gan has 421 stars on GitHub.
- Where can I find vision-aided-gan?
- nupurkmr9/vision-aided-gan is on GitHub at https://github.com/nupurkmr9/vision-aided-gan.