Why do cats' eyes glow in the dark? That eerie glow is actually a pair of built-in night-vision goggles. Cat eyes have even inspired some life-saving tech. Breakthroughs, discoveries, and DIY tips sent six days a week. One foggy night in 1933, a businessman named Percy Shaw was driving home from the pub in Yorkshire, England.

The scrapes and rubs the mammals leave behind shine under UV light humans can't see. Breakthroughs, discoveries, and DIY tips sent six days a week. Animals see the world around them in ways that we humans can only imagine. Arctic reindeer's eyes change color with the season to help them find food, while giant squid have eyes the size of dinner plates. Many species take advantage of seeing ultraviolet (UV) light that's invisible to humans--including deer .

Flow-based generative models 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 invertible 1x1 convolution. Using our method we demonstrate a significant improvement in log-likelihood and qualitative sample quality. Perhaps most strikingly, we demonstrate that a generative model optimized towards the plain log-likelihood objective is capable of efficient synthesis of large and subjectively realistic-looking images.

Flow-based generative models 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 invertible 1x1 convolution. Using our method we demonstrate a significant improvement in log-likelihood and qualitative sample quality. Perhaps most strikingly, we demonstrate that a generative model optimized towards the plain log-likelihood objective is capable of efficient synthesis of large and subjectively realistic-looking images.

We thank all reviewers for their insightful comments and useful suggestions. Reviewers 1 & 3 both suggested reporting FID scores. Our model achieves 46.16 on We also tested the official Glow model which got a FID score of 46.90, slightly worse than ours. We have additional experiments on 256x256 images where we achieve 1.00 bits/dim (Glow Our samples are similar to Glow's non-temperature We plan on releasing the weights for trained models to allow easier adoption in future works. We did not provide more downstream experiments (e.g.

Neural Information Processing Systems http://nips.cc/

We thank for the valuable feedback. We address the questions below and will revise our paper accordingly. On CIFAR-10, MaCow is 7.3 times slower than Glow, much faster than Emerging Convolution and MAF, whose factors are 360 and 600 respectively. We see that the time of generation increases linearly with the the image resolution. Convolutional Flow [Hoogeboom et al., 2019] is basically a linear transformation with masked convolutional kernels, Emerging Convolution [Hoogeboom et al., 2019] obtained 0.02 improvement on bits/dim by MaCow adopts additive coupling layers.

We will incorporate the answers/other feedback into the revised manuscript. The general quality of samples seems to be negatively impacted. We agree other wavelets could be potentially interesting. SR is not claimed as our primary goal/contribution. Rather, it is a fortuitous byproduct of the conditional structure that WF enables. A more thorough exploration of WF for SR is a promising direction for future work.