Advanced search
1 file | 734.77 KB

Exploiting reflectional and rotational invariance in single image superresolution

Simon Donné (UGent) , Laurens Meeus (UGent) , Hiep Luong (UGent) , Bart Goossens (UGent) and Wilfried Philips (UGent)
Author
Organization
Abstract
Stationarity of reconstruction problems is the crux to enabling convolutional neural networks for many image processing tasks: the output estimate for a pixel is generally not dependent on its location within the image but only on its immediate neighbourhood. We expect other invariances, too. For most pixel-processing tasks, rigid transformations should commute with the processing: a rigid transformation of the input should result in that same transformation of the output. In existing literature this is taken into account indirectly by augmenting the training set: reflected and ro- tated versions of the inputs are also fed to the network when optimizing the network weights. In contrast, we enforce this invariance through the network design. Because of the encompassing nature of the proposed architecture, it can directly enhance existing CNN-based algorithms. We show how it can be applied to SRCNN and FSRCNN both, speed ing up convergence in the initial training phase, and improving performance both for pretrained weights and after finetuning.
Keywords
Single image superresolution, deep learning, rotational equivariance

Downloads

  • Donne Exploiting Reflectional and CVPR 2017 paper.pdf
    • full text
    • |
    • open access
    • |
    • PDF
    • |
    • 734.77 KB

Citation

Please use this url to cite or link to this publication:

Chicago
Donné, Simon, Laurens Meeus, Hiep Luong, Bart Goossens, and Wilfried Philips. 2017. “Exploiting Reflectional and Rotational Invariance in Single Image Superresolution.” In 2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS (CVPRW) , 1043–1049. IEEE.
APA
Donné, S., Meeus, L., Luong, H., Goossens, B., & Philips, W. (2017). Exploiting reflectional and rotational invariance in single image superresolution. 2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS (CVPRW) (pp. 1043–1049). Presented at the 30th IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) , IEEE.
Vancouver
1.
Donné S, Meeus L, Luong H, Goossens B, Philips W. Exploiting reflectional and rotational invariance in single image superresolution. 2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS (CVPRW) . IEEE; 2017. p. 1043–9.
MLA
Donné, Simon, Laurens Meeus, Hiep Luong, et al. “Exploiting Reflectional and Rotational Invariance in Single Image Superresolution.” 2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS (CVPRW) . IEEE, 2017. 1043–1049. Print.
@inproceedings{8538773,
  abstract     = {Stationarity of reconstruction problems is the crux to enabling convolutional neural networks for many image processing tasks: the output estimate for a pixel is generally not dependent on its location within the image but only on
its immediate neighbourhood. We expect other invariances, too. For most pixel-processing tasks, rigid transformations
should commute with the processing: a rigid transformation of the input should result in that same transformation
of the output. In existing literature this is taken into account indirectly by augmenting the training set: reflected and ro-
tated versions of the inputs are also fed to the network when optimizing the network weights. In contrast, we enforce this invariance through the network design. Because of the encompassing nature of the proposed architecture, it can directly enhance existing CNN-based algorithms. We show how it can be applied to SRCNN and FSRCNN both, speed
ing up convergence in the initial training phase, and improving performance both for pretrained weights and after finetuning.},
  author       = {Donn{\'e}, Simon and Meeus, Laurens and Luong, Hiep and Goossens, Bart and Philips, Wilfried},
  booktitle    = {2017 IEEE CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION WORKSHOPS (CVPRW) },
  isbn         = {9781538607336},
  keyword      = {Single image superresolution,deep learning,rotational equivariance},
  language     = {eng},
  location     = {Honolulu, Hawaii},
  pages        = {1043--1049},
  publisher    = {IEEE},
  title        = {Exploiting reflectional and rotational invariance in single image superresolution},
  url          = {http://dx.doi.org/10.1109/cvprw.2017.141},
  year         = {2017},
}

Altmetric
View in Altmetric
Web of Science
Times cited: