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

ICLR 2018 Tero Karras • Timo Aila • Samuli Laine • Jaakko Lehtinen

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 1024^2.

Full paper

Evaluation


Task Dataset Model Metric name Metric value Global rank Compare
Image Generation CelebA-HQ 1024x1024 PGGAN FID 7.30 # 2
Image Generation CIFAR-10 PGGAN Inception score 8.80 # 1
Image Generation FFHQ PGGAN FID 8.04 # 2
Image Generation LSUN Bedroom 256 x 256 PGGAN FID 8.34 # 1