PhaseHarmonics: Wavelet phase harmonic transform in PyTorch

This is an implementation of the wavelet phase harmonic transform based on Kymatio (in the Python programming language). It is suitable for audio and image analysis and modeling.

Related Publications

• [2019] Stéphane Mallat, Sixin Zhang, Gaspar Rochette. Phase Harmonic Correlations and Convolutional Neural Networks. (paper).
• [2020] E. Allys, T. Marchand, et al. New Interpretable Statistics for Large Scale Structure Analysis and Generation. (paper).(code).
• [2021] Sixin Zhang, Stéphane Mallat. Maximum Entropy Models from Phase Harmonic Covariances. (paper).(code).
• [2021] Bruno Regaldo-Saint Blancard, Erwan Allys, François Boulanger, François Levrier, Niall Jeffrey. A new approach for the statistical denoising of Planck interstellar dust polarization data. (paper). (code).
• [2022] Brochard, Zhang, Mallat. Generalized Rectifier Wavelet Covariance Model For texture Synthesis. (paper). (code).
• [2022] Antoine Brochard, Bartłomiej Błaszczyszyn, Sixin Zhang, Stéphane Mallat. Particle gradient descent model for point process generation. (paper). (code)

Installation

For general installation, please follow the instructions at (kymatio). You may also use the script ./install_cuda92.sh to setup GPU-supported anaconda env.

Reproducing 2d reconstructions in paper [2019].

The code is tested on Ubuntu 16 + two TITAN Xp GPU + cuda 9.2 + Nvidia Driver Version: 410.66. Please follow the README in the folder code_rec2d. Matlab is needed to generate 2d wavelet filters.