Wavelet Flow: Fast Training of High Resolution Normalizing Flows
By Jason J. Yu, Konstantinos G. Derpanis, Marcus A. Brubaker. 2020
47 citations recorded by the source corpus.
Normalizing flows are a class of probabilistic generative models which allow for both fast density computation and efficient sampling and are effective at modelling complex distributions like images. A drawback among current methods is their significant training cost, sometimes requiring months of GPU training time to achieve state-of-the-art results. This paper introduces Wavelet Flow, a multi-scale, normalizing flow architecture based on wavelets. A Wavelet Flow has an explicit representation of signal scale that inherently includes models of lower resolution signals and conditional generation of higher resolution signals, i.e., super resolution. A major advantage of Wavelet Flow is the ability to construct generative models for high resolution data (e.g., 1024 x 1024 images) that are impractical with previous models. Furthermore, Wavelet Flow is competitive with previous normalizing flows in terms of bits per dimension on standard (low resolution) benchmarks while being up to 15x faster to train.
Subfields: Generative Models
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Generative Models
- Andreas Blattmann
- Björn Ommer
- Robin Rombach
- Michael Dorkenwald
- Timo Milbich
- Konstantinos G. Derpanis
- Ying Shan
- Chen Change Loy
Citations in the corpus
- Image Super-Resolution via Iterative Refinement (2021, neutral): “with a multi-scale approach [32], [60].”
- Stochastic Image-to-Video Synthesis using cINNs (2021, methodology): “In particular, INNs can be implemented as normalizing flows [60], a special class of likelihood-based generative models which have recently been applied to various tasks, such as image synthesis [38, 2, 57], domain transfer [62, 61, 23, 82], superresolution [47, 82], and video synthesis [40].”