Multi-Resolution Continuous Normalizing Flows

Vikram Voleti, Chris Finlay, Adam M Oberman, Christopher Pal

Keywords: Normalizing flows, generative models, neural ode, continuous normalizing flows, computer vision

Abstract: Recent work has shown that Neural Ordinary Differential Equations (ODEs) can serve as generative models of images using the perspective of Continuous Normalizing Flows (CNFs). Such models offer exact likelihood calculation, and invertible generation/density estimation. In this work we introduce a Multi-Resolution variant of such models (MRCNF), by characterizing the conditional distribution over the additional information required to generate a fine image that is consistent with the coarse image. We introduce a transformation between resolutions that allows for no change in the log likelihood. We show that this approach yields comparable likelihood values for various image datasets, using orders of magnitude fewer parameters than the prior methods, in significantly less training time, using only one GPU.

One-sentence Summary: We formulate a multi-resolution variant of continuous normalizing flows, and are able to train generative models of images to comparable performance with a fraction of the parameters in significantly less time using only 1 GPU.

Community Implementations: [ 1 code implementation](https://www.catalyzex.com/paper/arxiv:2106.08462/code)