Glow: Generative Flow with Invertible 1x1 Convolutions

NeurIPS 2018 Diederik P. KingmaPrafulla Dhariwal

Flow-based generative models (Dinh et al., 2014) are conceptually attractive due to tractability of the exact log-likelihood, tractability of exact latent-variable inference, and parallelizability of both training and synthesis. In this paper we propose Glow, a simple type of generative flow using an invertible 1x1 convolution... (read more)

PDF Abstract

Evaluation Results from the Paper


TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK COMPARE
Image Generation CIFAR-10 GLOW NLL Test 3.35 # 13
Image Generation ImageNet 64x64 GLOW Bits per byte 3.81 # 8