The third of its name...
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Failed to load latest commit information.


Build Status

What is it?

autotres is a collection of tools to analyze tongue surface contours in ultrasound images.

Where did it come from?

autotres is a direct descendant of AutoTrace (AutoTrace III). Read about the original in Jeff Berry's dissertation:

  title={Machine learning methods for articulatory data},
  author={Berry, Jeffrey James},
  school={The University of Arizona.}



We recommend running the project using a virtual environment. virtualenv can be installed with pip install virtualenv.

  1. virtualenv -p python3 venv
  2. source venv/bin/activate

Installation instructions:

System dependencies


The system dependencies can be installed via homebrew:

brew update;
brew install python3;
brew install gfortran;
brew tap homebrew/science;
brew install openblas;
brew install hdf5;


sudo apt-get update;
sudo apt-get install build-essential;
sudo apt-get install gcc;
sudo apt-get install python3-dev;
sudo apt-get install libhdf5-dev;
sudo apt-get install gfortran;
sudo apt-get build-dep libopenblas-dev;
sudo apt-get build-dep nvidia-cuda-toolkit;

Installing autotres and its Python dependencies:

pip install -e .

Using autotres

We provide network training and usage tutorials under examples.

Example data

You will need git-lfs to pull the example data.

Once you've installed git-lfs, simply run this command:

git-lfs fetch

Training networks

Networks can be trained using either a GPU or CPU. GPU training will save a great deal of time.

GPU-based training

The code for training deep networks uses Lasagne, a wrapper for Theano. One of the advantages of relying on these libraries is that networks can easily be trained on a CUDA-capable GPU, if present (with limited support for open-cl).

CPU-based training

If no GPU is present, Theano will use the CPU. For best results, a BLAS library with multithreading support is suggested, such as 'BLAS' or OpenBLAS.

What's missing?

Currently, the project lacks a graphical interface, and has only been tested on Ubuntu 14.04 and OSX 10.11. With luck, future versions will rectify these shortcomings.