progressive_growing_of_gans

 Progressive Growing of GANs for Improved Quality, Stability, and Variation.

http://research.nvidia.com/publication/2017-10_Progressive-Growing-of

Progressive Growing of GANs for Improved Quality, Stability, and Variation
— Official TensorFlow implementation of the ICLR 2018 paper

Tero Karras (NVIDIA), Timo Aila (NVIDIA), Samuli Laine (NVIDIA), Jaakko Lehtinen (NVIDIA and Aalto University)

• For business inquiries, please contact researchinquiries@nvidia.com
• For press and other inquiries, please contact Hector Marinez at hmarinez@nvidia.com

Picture: Two imaginary celebrities that were dreamed up by a random number generator.

Abstract:
We describe a new training methodology for generative adversarial networks. The key idea is to grow both the generator and discriminator progressively: starting from a low resolution, we add new layers that model increasingly fine details as training progresses. This both speeds the training up and greatly stabilizes it, allowing us to produce images of unprecedented quality, e.g., CelebA images at 10242. We also propose a simple way to increase the variation in generated images, and achieve a record inception score of 8.80 in unsupervised CIFAR10. Additionally, we describe several implementation details that are important for discouraging unhealthy competition between the generator and discriminator. Finally, we suggest a new metric for evaluating GAN results, both in terms of image quality and variation. As an additional contribution, we construct a higher-quality version of the CelebA dataset.

★★★ NEW: StyleGAN2-ADA-PyTorch is now available; see the full list of versions here ★★★

Resources

• Paper (NVIDIA research)
• Paper (arXiv)
• Result video (YouTube)
• Additional material (Google Drive)
  • ICLR 2018 poster (karras2018iclr-poster.pdf)
  • ICLR 2018 slides (karras2018iclr-slides.pptx)
  • Representative images (images/representative-images)
  • High-quality video clips (videos/high-quality-video-clips)
  • Huge collection of non-curated images for each dataset (images/100k-generated-images)
  • Extensive video of random interpolations for each dataset (videos/one-hour-of-random-interpolations)
  • Pre-trained networks (networks/tensorflow-version)
  • Minimal example script for importing the pre-trained networks (networks/tensorflow-version/example_import_script)
  • Data files needed to reconstruct the CelebA-HQ dataset (datasets/celeba-hq-deltas)
  • Example training logs and progress snapshots (networks/tensorflow-version/example_training_runs)

All the material, including source code, is made freely available for non-commercial use under the Creative Commons CC BY-NC 4.0 license. Feel free to use any of the material in your own work, as long as you give us appropriate credit by mentioning the title and author list of our paper.

from  https://github.com/tkarras/progressive_growing_of_gans