Graph-based Isometry Invariant Representation Learning

Renata Khasanova, Pascal Frossard
Proceedings of the 34th International Conference on Machine Learning, PMLR 70:1847-1856, 2017.

Abstract

Learning transformation invariant representations of visual data is an important problem in computer vision. Deep convolutional networks have demonstrated remarkable results for image and video classification tasks. However, they have achieved only limited success in the classification of images that undergo geometric transformations. In this work we present a novel Transformation Invariant Graph-based Network (TIGraNet), which learns graph-based features that are inherently invariant to isometric transformations such as rotation and translation of input images. In particular, images are represented as signals on graphs, which permits to replace classical convolution and pooling layers in deep networks with graph spectral convolution and dynamic graph pooling layers that together contribute to invariance to isometric transformation. Our experiments show high performance on rotated and translated images from the test set compared to classical architectures that are very sensitive to transformations in the data. The inherent invariance properties of our framework provide key advantages, such as increased resiliency to data variability and sustained performance with limited training sets.

Cite this Paper

BibTeX
@InProceedings{pmlr-v70-khasanova17a, title = {Graph-based Isometry Invariant Representation Learning}, author = {Renata Khasanova and Pascal Frossard}, booktitle = {Proceedings of the 34th International Conference on Machine Learning}, pages = {1847--1856}, year = {2017}, editor = {Precup, Doina and Teh, Yee Whye}, volume = {70}, series = {Proceedings of Machine Learning Research}, month = {06--11 Aug}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v70/khasanova17a/khasanova17a.pdf}, url = {https://proceedings.mlr.press/v70/khasanova17a.html}, abstract = {Learning transformation invariant representations of visual data is an important problem in computer vision. Deep convolutional networks have demonstrated remarkable results for image and video classification tasks. However, they have achieved only limited success in the classification of images that undergo geometric transformations. In this work we present a novel Transformation Invariant Graph-based Network (TIGraNet), which learns graph-based features that are inherently invariant to isometric transformations such as rotation and translation of input images. In particular, images are represented as signals on graphs, which permits to replace classical convolution and pooling layers in deep networks with graph spectral convolution and dynamic graph pooling layers that together contribute to invariance to isometric transformation. Our experiments show high performance on rotated and translated images from the test set compared to classical architectures that are very sensitive to transformations in the data. The inherent invariance properties of our framework provide key advantages, such as increased resiliency to data variability and sustained performance with limited training sets.} }
Endnote
%0 Conference Paper %T Graph-based Isometry Invariant Representation Learning %A Renata Khasanova %A Pascal Frossard %B Proceedings of the 34th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2017 %E Doina Precup %E Yee Whye Teh %F pmlr-v70-khasanova17a %I PMLR %P 1847--1856 %U https://proceedings.mlr.press/v70/khasanova17a.html %V 70 %X Learning transformation invariant representations of visual data is an important problem in computer vision. Deep convolutional networks have demonstrated remarkable results for image and video classification tasks. However, they have achieved only limited success in the classification of images that undergo geometric transformations. In this work we present a novel Transformation Invariant Graph-based Network (TIGraNet), which learns graph-based features that are inherently invariant to isometric transformations such as rotation and translation of input images. In particular, images are represented as signals on graphs, which permits to replace classical convolution and pooling layers in deep networks with graph spectral convolution and dynamic graph pooling layers that together contribute to invariance to isometric transformation. Our experiments show high performance on rotated and translated images from the test set compared to classical architectures that are very sensitive to transformations in the data. The inherent invariance properties of our framework provide key advantages, such as increased resiliency to data variability and sustained performance with limited training sets.
APA
Khasanova, R. & Frossard, P.. (2017). Graph-based Isometry Invariant Representation Learning. Proceedings of the 34th International Conference on Machine Learning, in Proceedings of Machine Learning Research 70:1847-1856 Available from https://proceedings.mlr.press/v70/khasanova17a.html.

Related Material