Vision Open Statistical Models
VOSM has moved from SourceForge to Github. Please refer to Github html for the online reference.
Local documentation is also provided both in html and pdf format.
Current VOSM implements all popular ASMs/AAMs, and one new fitting methods (namely, 2D profile ASM) both for static images and dynamic image sequences, including:
- Original 1D profile ASM
- Originated 2D profile ASM
- Direct Local texture constrained (LTC) ASM
- Basic AAM
- ICIA (Inverse compositional image alignment) AAM
- IAIA (Inverse additive image alignment) AAM
- OpenCV 3.0.0 or above
- boost-1.58 or above (boost-regex, boost-filesystem and boost-system are required)
- Before using VOSM, it’s REQUIRED to download respective ShapeInfo.txt file summarized in the following table:
- To build the 2D statistical model, use command test_smbuilding:
- To test the 2D statistical model, use command test_smfitting:
More details about these two commands can be found by respective help commands.
A wxWidget based GUI tool is provided to simplify the process of building an AAM model. OpenCV is the kernel for image processing, while BOOST is for file I/O. Since this tool was roughly finished in Toronto, Canada, I call it a Toronto version.
You can obtain some basic AAM conceptions from my technical report 2D Statistical Models.
aambuilding-1.0 is a stable version of AAM Building, which has been tested with Boost 1.36 and OpenCV-cvs, both in Windows Vista and Ubuntu 8.04.
Some demonstrative videos can be found on this page.