FrontierThe story, in brief
Glow: Better reversible generative models
OpenAI's Glow: Reversible generative models that generate high-res images with efficient sampling.
The KeyNews take
Why it matters
OpenAI introduces a simplified architecture for reversible generative models using invertible convolutions, advancing the state of generative modeling with practical improvements in sampling efficiency and interpretability.
The key facts
9 to knowGlow uses invertible 1x1 convolutions
Supports efficient sampling
Generates realistic high-resolution images
Discovers manipulable data attributes
Code and visualization tool released open-source
Published July 2018
Invertible 1x1 convolutions for reversible generation
High-resolution image generation capability
Open-source code and visualization tool released
Go to the source
OpenAI Blogopenai.com
Publisher excerpt: We introduce Glow, a reversible generative model which uses invertible 1x1 convolutions. It extends previous work on reversible generative models and simplifies the architecture. Our model can generate realistic high resolution images, supports efficient sampling, and discovers features that can be…