Normalizing Flow as a Flexible Fidelity Objective for Photo-Realistic Super-resolution
METADATA ONLY
Loading...
Author / Producer
Date
2022
Publication Type
Conference Paper
ETH Bibliography
yes
Citations
Altmetric
METADATA ONLY
Data
Rights / License
Abstract
Super-resolution is an ill-posed problem, where a ground-truth high-resolution image represents only one possibility in the space of plausible solutions. Yet, the dominant paradigm is to employ pixel-wise losses, such as L-1, which drive the prediction towards a blurry average. This leads to fundamentally conflicting objectives when combined with adversarial losses, which degrades the final quality. We address this issue by revisiting the L-1 loss and show that it corresponds to a one-layer conditional flow. Inspired by this relation, we explore general flows as a fidelity-based alternative to the L-1 objective. We demonstrate that the flexibility of deeper flows leads to better visual quality and consistency when combined with adversarial losses. We conduct extensive user studies for three datasets and scale factors, where our approach is shown to outperform state-of-the-art methods for photo-realistic super-resolution. Code and trained models: git.io/AdFlow
Permanent link
Publication status
published
External links
Editor
Book title
2022 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)
Journal / series
Volume
Pages / Article No.
874 - 883
Publisher
IEEE
Event
22nd IEEE/CVF Winter Conference on Applications of Computer Vision (WACV 2022)
Edition / version
Methods
Software
Geographic location
Date collected
Date created
Subject
Computational photography; Image and video synthesis
Organisational unit
03514 - Van Gool, Luc (emeritus) / Van Gool, Luc (emeritus)
09688 - Yu, Fisher (ehemalig) / Yu, Fisher (former)