VOSM — Vision Open Statistical Model

C++ library of 2D/3D statistical shape & appearance models (ASM / AAM variants).

VOSM (Vision Open Statistical Model) is a C++ library implementing variants of 2D and 3D Active Shape Models (ASM) and Active Appearance Models (AAM). It grew out of earlier face-modeling and statistical-shape research and remains the reference implementation behind much of Vision Open’s work on landmark fitting.

What it does

VOSM is organized around two stages:

• Model building — both 2D and 3D models are supported. By selecting different parameters you can build a Shape Model (SM), Texture Model (TM), Appearance Model (AM, classic AAM), Image-Alignment Appearance Model (IA, ICIA AAM), a generalized Feature Model (FM), a Shape Model with Local Texture Constraints (SMLTC, CLM-style), and SMNDPROFILE, which includes both the original 1D-profile ASM and the 2D-profile ASM proposed in my PhD work.
• Fitting — testing model fit against images (2D fitting only).

Stack & requirements

• Language: C++ (≈98% of the codebase)
• OpenCV ≥ 3.4.1
• Boost ≥ 1.65.1
• License: LGPL-3.0
• Latest release: 0.3.5 (2018)

Build

git clone https://github.com/jiapei100/VOSM
mkdir build && cd build
ccmake ../
make -j8
sudo make install

This installs two test commands (testsmbuilding, testsmfitting) and eight libvosm_* shared libraries.

Background

VOSM ties directly into the 2D-profile ASM described in chapter 3 of my PhD thesis, Audio-visual based HMI for an Intelligent Wheelchair (University of Essex, 2010), and the 2D Statistical Models technical report (Vision Open Working Group, 2010).

Repository: github.com/jiapei100/VOSM · 60★ · 33 forks · LGPL-3.0