Kymatio: Scattering Transforms in Python
Mathieu Andreux, Tomás Angles(École Normale Supérieure - PSL), Georgios Exarchakis(École Normale Supérieure - PSL), Roberto Leonarduzzi(École Normale Supérieure - PSL), Gaspar Rochette(École Normale Supérieure - PSL), Louis Thiry(École Normale Supérieure - PSL), Zarka, John(École Normale Supérieure - PSL), Stéphane Mallat(Collège de France), Joakim Andén, Eugene Belilovsky, Joan Bruna(New York University), Vincent Lostanlen(New York University), Chaudhary, Muawiz(Great Lakes Environmental Center), Matthew Hirn, Edouard Oyallon(Peking University), Sixin Zhang(Creative Technologies (United States)), Carmine-Emanuele Cella(Flatiron Health (United States)), Michael Eickenberg
Cited by 77Open Access
Abstract
The wavelet scattering transform is an invariant signal representation\nsuitable for many signal processing and machine learning applications. We\npresent the Kymatio software package, an easy-to-use, high-performance Python\nimplementation of the scattering transform in 1D, 2D, and 3D that is compatible\nwith modern deep learning frameworks. All transforms may be executed on a GPU\n(in addition to CPU), offering a considerable speed up over CPU\nimplementations. The package also has a small memory footprint, resulting\ninefficient memory usage. The source code, documentation, and examples are\navailable undera BSD license at https://www.kymat.io/\n
Related Papers
Invariant Scattering Convolution Networks
Joan Bruna, Stéphane Mallat|IEEE Transactions on Pattern Analysis and Machine Intelligence|2013|1.6k