fjxmlzn / InfoGAN-CR

[ICML 2020] InfoGAN-CR and ModelCentrality: Self-supervised Model Training and Selection for Disentangling GANs
https://arxiv.org/abs/1906.06034
41 stars 8 forks source link

MIG mutual information gap? #3

Closed sharonzhou closed 4 years ago

sharonzhou commented 4 years ago

How well do you perform on dSprites using the mutual information gap (MIG)? I'd expect this to do well, so would love to see results.

fjxmlzn commented 4 years ago

Hi, I just uploaded our code for computing MIG, FactorVAE metric, BetaVAE metric, SAP, Explicitness, Modularity, DCI, and dHSIC. The instructions on how to use it are in README. The metric numbers we get will soon be available in ICML camera ready.

fjxmlzn commented 4 years ago

We have just updated the metric numbers (including MIG) in the arxiv version: https://arxiv.org/abs/1906.06034. It is also available in the readme.