Volltext-Downloads (blau) und Frontdoor-Views (grau)
  • search hit 22 of 70
Back to Result List

Watch your Up-Convolution: CNN Based Generative Deep Neural Networks are Failing to Reproduce Spectral Distributions

  • Generative convolutional deep neural networks, e.g. popular GAN architectures, are relying on convolution based up-sampling methods to produce non-scalar outputs like images or video sequences. In this paper, we show that common up-sampling methods, i.e. known as up-convolution or transposed convolution, are causing the inability of such models to reproduce spectral distributions of naturalGenerative convolutional deep neural networks, e.g. popular GAN architectures, are relying on convolution based up-sampling methods to produce non-scalar outputs like images or video sequences. In this paper, we show that common up-sampling methods, i.e. known as up-convolution or transposed convolution, are causing the inability of such models to reproduce spectral distributions of natural training data correctly. This effect is independent of the underlying architecture and we show that it can be used to easily detect generated data like deepfakes with up to 100% accuracy on public benchmarks. To overcome this drawback of current generative models, we propose to add a novel spectral regularization term to the training optimization objective. We show that this approach not only allows to train spectral consistent GANs that are avoiding high frequency errors. Also, we show that a correct approximation of the frequency spectrum has positive effects on the training stability and output quality of generative networks.show moreshow less

Export metadata

Statistics

frontdoor_oas
Metadaten
Author:Ricard Durall Lopez, Margret Keuper, Janis KeuperORCiDGND
Creating Corporation:IEEE
Year of Publication:2020
Pagenumber:8
ISBN:978-1-7281-7168-5 (digital)
ISBN:978-1-7281-7169-2 (Print on Demand)
Language:English
Parent Title (English):Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition
ISSN:2575-7075 (digital)
ISSN:1063-6919 (Print on Demand)
First Page:7887
Last Page:7896
Document Type:Conference Proceeding
Open Access:Frei zugänglich
Institutes:Bibliografie
Release Date:2021/01/07
Licence (German):License LogoEs gilt das UrhG
Note:
Konferenz: 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 13-19 June 2020, Seattle, WA, USA
URL:https://openaccess.thecvf.com/content_CVPR_2020/html/Durall_Watch_Your_Up-Convolution_CNN_Based_Generative_Deep_Neural_Networks_Are_CVPR_2020_paper.html
DOI:https://doi.org/10.1109/CVPR42600.2020.00791
ArXiv Id:http://arxiv.org/abs/2003.01826