/kymatio

Wavelet scattering transforms in Python with GPU acceleration

Primary LanguagePythonBSD 3-Clause "New" or "Revised" LicenseBSD-3-Clause

Kymatio: Wavelet scattering in Python

Kymatio is an implementation of the wavelet scattering transform in the Python programming language, suitable for large-scale numerical experiments in signal processing and machine learning. Scattering transforms are translation-invariant signal representations implemented as convolutional networks whose filters are not learned, but fixed (as wavelet filters).

PyPI License Build Status Downloads codecov

Use Kymatio if you need a library that:

  • supports 1-D, 2-D, and 3-D wavelets,
  • integrates wavelet scattering in a deep learning architecture, and
  • runs seamlessly on CPU and GPU hardware, with major deep learning APIs, such as PyTorch and TensorFlow.

The Kymatio environment

Flexibility

The Kymatio organization associates the developers of several pre-existing packages for wavelet scattering, including ScatNet, scattering.m, PyScatWave, WaveletScattering.jl, and PyScatHarm.

The resort to PyTorch tensors as inputs to Kymatio allows the programmer to backpropagate the gradient of wavelet scattering coefficients, thus integrating them within an end-to-end trainable pipeline, such as a deep neural network.

Portability

Each of these algorithms is written in a high-level imperative paradigm, making it portable to any Python library for array operations as long as it enables complex-valued linear algebra and a fast Fourier transform (FFT).

Each algorithm comes packaged with a frontend and backend. The frontend takes care of interfacing with the user. The backend defines functions necessary for computation of the scattering transform.

Currently, there are six available frontend–backend pairs, NumPy (CPU), scikit-learn (CPU), pure PyTorch (CPU and GPU), PyTorch+scikit-cuda (GPU), TensorFlow (CPU and GPU), and Keras (CPU and GPU).

Scalability

Kymatio integrates the construction of wavelet filter banks in 1D, 2D, and 3D, as well as memory-efficient algorithms for extracting wavelet scattering coefficients, under a common application programming interface.

Running Kymatio on a graphics processing unit (GPU) rather than a multi-core conventional central processing unit (CPU) allows for significant speedups in computing the scattering transform. The current speedup with respect to CPU-based MATLAB code is of the order of 10 in 1D and 3D and of the order of 100 in 2D.

We refer to our official benchmarks for further details.

How to cite

If you use this package, please cite the following paper:

Andreux M., Angles T., Exarchakis G., Leonarduzzi R., Rochette G., Thiry L., Zarka J., Mallat S., Andén J., Belilovsky E., Bruna J., Lostanlen V., Hirn M. J., Oyallon E., Zhang S., Cella C., Eickenberg M. (2019). Kymatio: Scattering Transforms in Python. arXiv preprint arXiv:1812.11214. (paper)

Installation

Dependencies

Kymatio requires:

  • Python (>= 3.5)
  • SciPy (>= 0.13)

Standard installation (on CPU hardware)

We strongly recommend running Kymatio in an Anaconda environment, because this simplifies the installation of other dependencies. You may install the latest version of Kymatio using the package manager pip, which will automatically download Kymatio from the Python Package Index (PyPI):

pip install kymatio

Linux and macOS are the two officially supported operating systems.

Frontend

NumPy

To explicitly call the numpy frontend, run:

from kymatio.numpy import Scattering2D
scattering = Scattering2D(J=2, shape=(32, 32))

Scikit-learn

After installing the latest version of scikit-learn, you can call Scattering2D as a Transformer using:

from kymatio.sklearn import Scattering2D

scattering_transformer = Scattering2D(2, (32, 32))

PyTorch

After installing the latest version of PyTorch, you can call Scattering2D as a torch.nn.Module using:

from kymatio.torch import Scattering2D

scattering = Scattering2D(J=2, shape=(32, 32))

TensorFlow

After installing the latest version of TensorFlow, you can call Scattering2D as a tf.Module using:

from kymatio.tensorflow import Scattering2D

scattering = Scattering2D(J=2, shape=(32, 32))

Keras

Alternatively, with TensorFlow installed, you can call Scattering2D as a Keras Layer using:

from tensorflow.keras.layers import Input
from kymatio.keras import Scattering2D

inputs = Input(shape=(32, 32))
scattering = Scattering2D(J=2)(inputs)

Installation from source

Assuming the Kymatio source has been downloaded, you may install it by running

pip install -r requirements.txt
python setup.py install

Developers can also install Kymatio via:

pip install -r requirements.txt
python setup.py develop

GPU acceleration

Certain frontends, numpy and sklearn, only allow processing on the CPU and are therefore slower. The torch, tensorflow, and keras frontends, however, also support GPU processing, which can significantly accelerate computations. Additionally, the torch backend supports an optimized skcuda backend which currently provides the fastest performance in computing scattering transforms. In 2D, it may be instantiated using:

from kymatio.torch import Scattering2D

scattering = Scattering2D(J=2, shape=(32, 32), backend='torch_skcuda')

This is particularly useful when working with large images, such as those in ImageNet, which are of size 224×224.

PyTorch and scikit-cuda

To run Kymatio on a graphics processing unit (GPU), you can either use the PyTorch-style cuda() method to move your object to GPU. Kymatio is designed to operate on a variety of backends for tensor operations. For extra speed, install the CUDA library and the skcuda dependency by running the following pip command:

pip install scikit-cuda cupy

The user may control the choice of backend at runtime via for instance:

from kymatio.torch import Scattering2D
scattering = Scattering2D(J=2, shape=(32, 32)), backend='torch_skcuda')

Documentation

The documentation of Kymatio is officially hosted on the kymat.io website.

Online resources

Building the documentation from source

The documentation can also be found in the doc/ subfolder of the GitHub repository. To build the documentation locally, please clone this repository and run

pip install -r requirements_optional.txt
cd doc; make clean; make html

Support

We wish to thank the Scientific Computing Core at the Flatiron Institute for the use of their computing resources for testing.

We would also like to thank École Normale Supérieure for their support.

ENS

Kymatio

Kyma (κύμα) means wave in Greek. By the same token, Kymatio (κυμάτιο) means wavelet.

Note that the organization and the library are capitalized (Kymatio) whereas the corresponding Python module is written in lowercase (import kymatio).

The recommended pronunciation for Kymatio is kim-ah-tio. In other words, it rhymes with patio, not with ratio.