Papers of the day   All papers



Dec 03 2019 DeepMind

We introduce LOGAN, a game-theory motivated algorithm, which improves the state-of-the-art in GAN image generation by over 30% measured in FID: Here are samples showing higher diversity:
13 replies, 1089 likes

Dec 03 2019 roadrunner01

LOGAN: Latent Optimisation for Generative Adversarial Networks pdf: abs:
2 replies, 72 likes

Dec 05 2019 Christian Mio Loclair

Wow - thats quite a step
1 replies, 10 likes

Feb 18 2020 Sander Dieleman

@fhuszar Not sure if it's been mentioned, but LOGAN ( is a recent practical example.
0 replies, 6 likes

Dec 03 2019 Dave Gershgorn

0 replies, 3 likes

Dec 03 2019 Elliot Turner

Great quality and diversity-of-samples improvement from LOGAN (from Google) compared to BigGAN-deep (LOGAN is on the right in my pasted image) -
0 replies, 1 likes

Dec 03 2019 Alistair Young

Is that a pukeko?
0 replies, 1 likes

Dec 04 2019 akira Improve GAN performance by optimizing latent variables z and greatly update ImageNet SOTA. D and G are updated using z' that is optimized z by D value. Optimizing latent variables with natural gradient is more effective.
0 replies, 1 likes

Dec 04 2019 Brundage Bot

LOGAN: Latent Optimisation for Generative Adversarial Networks. Yan Wu, Jeff Donahue, David Balduzzi, Karen Simonyan, and Timothy Lillicrap
1 replies, 0 likes