44 Downloads (Pure)

Abstract

This paper is the first work to propose a network to predict a structured uncertainty distribution for a synthesized image.
Previous approaches have been mostly limited to predicting diagonal covariance matrices.
Our novel model learns to predict a full Gaussian covariance matrix for each reconstruction, which permits efficient sampling and likelihood evaluation.

We demonstrate that our model can accurately reconstruct ground truth correlated residual distributions for synthetic datasets and generate plausible high frequency samples for real face images. We also illustrate the use of these predicted covariances for structure preserving image denoising.
Original languageEnglish
Number of pages8
Publication statusAccepted/In press - 20 Mar 2018
EventIEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2018 -
Duration: 18 Jun 201822 Jun 2018

Conference

ConferenceIEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2018
Period18/06/1822/06/18

Fingerprint

Covariance matrix
Image denoising
Sampling
Uncertainty

Keywords

  • Deep Neural Networks
  • Generative Models

Cite this

Dorta, G., Vicente, S., Agapito, L., Campbell, N., & Simpson, I. (Accepted/In press). Structured Uncertainty Prediction Networks. Paper presented at IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2018, .

Structured Uncertainty Prediction Networks. / Dorta, Garoe; Vicente, Sara; Agapito, Lourdes; Campbell, Neill; Simpson, Ivor.

2018. Paper presented at IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2018, .

Research output: Contribution to conferencePaper

Dorta, G, Vicente, S, Agapito, L, Campbell, N & Simpson, I 2018, 'Structured Uncertainty Prediction Networks' Paper presented at IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2018, 18/06/18 - 22/06/18, .
Dorta G, Vicente S, Agapito L, Campbell N, Simpson I. Structured Uncertainty Prediction Networks. 2018. Paper presented at IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2018, .
Dorta, Garoe ; Vicente, Sara ; Agapito, Lourdes ; Campbell, Neill ; Simpson, Ivor. / Structured Uncertainty Prediction Networks. Paper presented at IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2018, .8 p.
@conference{06498402a6994bf5b0e26349b7d21807,
title = "Structured Uncertainty Prediction Networks",
abstract = "This paper is the first work to propose a network to predict a structured uncertainty distribution for a synthesized image.Previous approaches have been mostly limited to predicting diagonal covariance matrices.Our novel model learns to predict a full Gaussian covariance matrix for each reconstruction, which permits efficient sampling and likelihood evaluation.We demonstrate that our model can accurately reconstruct ground truth correlated residual distributions for synthetic datasets and generate plausible high frequency samples for real face images. We also illustrate the use of these predicted covariances for structure preserving image denoising.",
keywords = "Deep Neural Networks, Generative Models",
author = "Garoe Dorta and Sara Vicente and Lourdes Agapito and Neill Campbell and Ivor Simpson",
year = "2018",
month = "3",
day = "20",
language = "English",
note = "IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2018 ; Conference date: 18-06-2018 Through 22-06-2018",

}

TY - CONF

T1 - Structured Uncertainty Prediction Networks

AU - Dorta, Garoe

AU - Vicente, Sara

AU - Agapito, Lourdes

AU - Campbell, Neill

AU - Simpson, Ivor

PY - 2018/3/20

Y1 - 2018/3/20

N2 - This paper is the first work to propose a network to predict a structured uncertainty distribution for a synthesized image.Previous approaches have been mostly limited to predicting diagonal covariance matrices.Our novel model learns to predict a full Gaussian covariance matrix for each reconstruction, which permits efficient sampling and likelihood evaluation.We demonstrate that our model can accurately reconstruct ground truth correlated residual distributions for synthetic datasets and generate plausible high frequency samples for real face images. We also illustrate the use of these predicted covariances for structure preserving image denoising.

AB - This paper is the first work to propose a network to predict a structured uncertainty distribution for a synthesized image.Previous approaches have been mostly limited to predicting diagonal covariance matrices.Our novel model learns to predict a full Gaussian covariance matrix for each reconstruction, which permits efficient sampling and likelihood evaluation.We demonstrate that our model can accurately reconstruct ground truth correlated residual distributions for synthetic datasets and generate plausible high frequency samples for real face images. We also illustrate the use of these predicted covariances for structure preserving image denoising.

KW - Deep Neural Networks

KW - Generative Models

M3 - Paper

ER -