DeMoLib – Home

Copyright © 2007-2011, Dr Jason Saragih, Dr Roland Goecke and contributors, Australian National University, University of Canberra (in alphabetical order). All rights reserved.

The Deformable Model Library DeMoLib provides efficient C++ implementations of a number of fitting methods under the Active Appearance Model (AAM) family. The AAM is a powerful method for modelling and segmenting deformable visual objects, such as for example faces. The utility of the AAM stems from two fronts: its compact representation as a linear object class and its rapid fitting procedure, which utilizes fixed linear updates. The AAM framework was first introduced by Edwards, Taylor and Cootes at FG’98, but later extended by researchers around the world who focussed on different fitting methods within this framework. For more information, please see the websites by Prof Tim Cootes, the Face Group at the Robotics Institute at CMU, and this homepage of Dr Roland Goecke.

DeMoLib contains implementations of the following AAM fitting methods:

  • Fixed Jacobian: Original fitting method by Cootes et al. (gzipped PS)
  • Inverse Compositional fitting methods by Baker, Mathews, Gross et al.
    • Project-Out Inverse Compositional (LINK)
    • Project-Out Inverse Compositional with 3D shape constraint
    • Simultaneous Inverse Compositional (LINK)
    • Robust Inverse Compositional (LINK)
    • Normalised Inverse Compositional (LINK)
  • Discriminative-Iterative fitting methods by Saragih and Goecke
    • Linear Discriminative-Iterative (original IEBM method (PDF), DI method (PDF))
    • Non-linear (Haar-like feature based) Discriminative-Iterative (PDF)
  • 2D+3D fitting method (LINK)

In addition, DeMoLib also contains an implementation of the Active Shape Model (ASM) algorithm, introduced by Prof Tim Cootes, which forms the basis for the Active Appearance Model. As general further background information, it is also recommended to read the series of papers “Lucas-Kanade 20 Years On: A Unifying Framework” by Simon Baker and Iain Matthews.

This is a cross-platform implementation that has been shown to successfully build on Linux/Unix OS, Mac OS X and Windows OS (XP, Vista, 7). DeMoLib was originally written by Dr Jason Saragih as part of his PhD project under the supervision of Dr Roland Goecke, but later extended by student contributors from the ANU and the University of Canberra.

The DeMoLib software library is available in source format for academic research purposes under an academic, non‐commercial, not‐for‐profit licence. If you are interested in obtaining a copy of the software, please contact Dr Roland Goecke.

Software Requirements

DeMoLib uses certain image processing and mathematical functions from OpenCV and VxL. You will therefore have to install these first, before installing DeMoLib. OpenCV and VxL source and installation files are available from their respective SourceForge websites (follow the links above). To facilitate building DeMoLib on different operating system platforms, the software is built from the sources via CMake.

  • OpenCV: Preferably v2.0 and higher, but also works for v1.1 (however not for OpenCV 1.0)
  • VxL: Preferably v1.14.0, but also works for v1.11.0 and higher
  • CMake: Preferably v2.8 or higher, but also works for v2.6 and higher