Keras implement of flow-based models

nice.py: NICE (Non-linear Independent Components Estimation, intro)

glow.py: Glow (Glow: Generative Flow with Invertible 1×1 Convolutions, intro)

32x32 celeba with level=3,depth=6

32x32 cifar10 with level=3,depth=6

f-VAEs.py: f-VAEs (f-VAEs: Improve VAEs with Conditional Flows , intro, arxiv)

环境

测试环境包括：

• Keras 2.1.5 + tensorflow 1.2
• Keras 2.2.0 + tensorflow 1.8
• Keras 2.2.4 + tensorflow 1.13

均在Python 2.7下测试。

链接

• https://kexue.fm/archives/5776
• https://kexue.fm/archives/5807

交流

QQ交流群：67729435，微信群请加机器人微信号spaces_ac_cn