fbcotter / dtcwt_slim

Slim implementation of the dtcwt in tensorflow

Dual-Tree Complex Wavelet Transform library for Python

This library provides support for computing 2D dual-tree complex wavelet transforms and their inverse using python and tensorflow.

The implementation is designed to be used with batches of multichannel images. The code can handle 2-D, 3-D or 4-D inputs, but the last 2 dimensions must be the height and width. In tensorflow terms this equates to using the 'NCHW' format.

Installation

The easiest way to install dtcwt_slim is to clone the repo and pip install it. Later versions will be released on PyPi but the docs need to updated first:

$ git clone https://github.com/fbcotter/dtcwt_slim
$ cd dtcwt_slim
$ pip install .

$ python setup.py install

(Although the develop command may be more useful if you intend to perform any significant modification to the library.) A test suite is provided so that you may verify the code works on your system:

$ pip install -r tests/requirements.txt
$ py.test

This will also write test-coverage information to the cover/ directory.

Example Use

This repo was based off the one here. There is a tensorflow back-end in there but conforming to the standards set up in there meant not doing the most efficient operations in tensorflow (in terms of the shape of inputs). This means that the interface is almost identical. Below is an example of how to use dtcwt_slim

The key differences between dtcwt_slim and dtcwt are:

• dtcwt_slim doesn't use pyramids. Calling the forward method returns Yl and Yh
• Yh and Yscales are returned as lists not tuples, making it easier to manipulate individual scales without worrying about immutability
• inverse no longer has the gain_mask input. Applying gains can still be done before calling Transform2d.inverse
• dtcwt_slim can naturally handle batches of multi-channel signals. E.g. inputs of data type 'hw', 'nhw/chw' or 'nchw' are permitted.
• The highpass outputs have the 6 orientations as the third last dimension now, rather than the last. Before an output would be of shape [H, W, 6] but now they will be returned as [6, H, W]. If there are multiple channels, then the orientations dimension still comes in the third last position - i.e. the output will be [C, 6, H, W].

• dtcwt_slim can return real highpass outputs instead of complex ones. This saves overhead on having to constantly go back to real operations to do things that aren't supported in the complex domain (like convolutions). To return real highpasses, pass the complex=False parameter to the class initializer:

  import dtcwt_slim
  xfm = dtcwt_slim.Transform2d(complex=False)

  The highpass outputs then are a list of length 2.

Provenance

Based on the Dual-Tree Complex Wavelet Transform Pack for MATLAB by Nick Kingsbury, Cambridge University. The original README can be found in ORIGINAL_README.txt. This file outlines the conditions of use of the original MATLAB toolbox.