Python wrapper for CurveLab's 2D and 3D curvelet transforms. It uses the PyLops design framework to provide the forward and inverse curvelet transforms as matrix-free linear operations. If you are still confused, check out some examples below or the PyLops website!

Installing curvelops requires the following components:

• FFTW 2.1.5
• CurveLab >= 2.0.2

Both of these packages must be installed manually. See more information below. After these are installed, you may install curvelops with:

export FFTW=/path/to/fftw
export FDCT=/path/to/CurveLab
python3 -m pip install git+https://github.com/cako/curvelops@main

as long as you are using a pip>=10.0. To check, run python3 -m pip --version.

For a 2D transform, you can get started with:

import numpy as np
import curvelops as cl

x = np.random.normal(0., 1., (100, 50))
FDCT = cl.FDCT2D(dims=(100, 50))
c = FDCT * x.ravel()
xinv = FDCT.H * c
assert np.allclose(x, xinv.reshape(100, 50))

An excellent place to see how to use the library is the examples/ folder. Demo_Single_Curvelet for example contains a curvelops version of the CurveLab Matlab demo.

For FFTW 2.1.5, you must compile with position-independent code support. Do that with

./configure --with-pic --prefix=/home/user/opt/fftw-2.1.5 --with-gcc=/usr/bin/gcc

The --prefix and --with-gcc are optional and determine where it will install FFTW and where to find the GCC compiler, respectively. We recommend using the same compile for FFTW, CurveLab and curvelops.

In the file makefile.opt set FFTW_DIR, CC and CXX variables as required in the instructions. To keep things consistent, set FFTW_DIR=/home/user/opt/fftw-2.1.5 (or whatever directory was used in the --prefix option). For the others, use the same compiler which was used to compile FFTW.

The FFTW variable is the same as FFTW_DIR as provided in the CurveLab installation. The FDCT variable points to the root of the CurveLab installation. It will be something like /path/to/CurveLab-2.1.3 for the latest version.

• Paul Goyes has kindly contributed a rundown of how to install curvelops: link to YouTube video (in Spanish).

This package contains no CurveLab code apart from function calls. It is provided to simplify the use of CurveLab in a Python environment. Please ensure you own a CurveLab license as per required by the authors. See the CurveLab website for more information. All CurveLab rights are reserved to Emmanuel Candes, Laurent Demanet, David Donoho and Lexing Ying.