SDM for facial landmark alignment, based on the work of Xiong & De La Torre.
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
lib
300w.mat
README.md
runDemo.m
test1.jpg
test2.jpg
test3.png

README.md

SDM for Face Alignment

This is my MATLAB implementation of the Supervised Descent Method (SDM) for face alignment proposed by Xiong and De La Torre (https://courses.cs.washington.edu/courses/cse590v/13au/intraface.pdf).

The included model is trained on the 300W dataset using 49 points (no jaw points).

Training code is included in the lib directory. Training should be a fairly straight-forward process. The functions for training are:

  • SDMInitModel - Allocates a new SDM model
  • SDMAddDataMemoryFrugal* - these functions gather the training and validation data in a memory efficient way for training and parameter selection.
  • SDMSolve - Solves for the regressor
  • SDMLayerXval - Performs parameter selection via a golden search (use for selecting regularizer)
  • SDMFinalizeLayer - Adds the final regressor to the SDM model.

Review these functions for details on model structure, and how to provide a shape model (mean shape).

Running the code

There are no external dependancies or mex functions. Running "runDemo.m" should produce an example fit.

License

This code is not to be used for commerical purposes. This code can be freely used for personal, academic and research purposes. However, we ask that any files retain our copyright notice when redistributed. This software is provided ``as is'', with no warranty or guarantee. We accept no responsibility for any damages incurred.

The original algorithm (described in: https://courses.cs.washington.edu/courses/cse590v/13au/intraface.pdf) is the work of Xiong and De La Torre.

Acknowledgements

I would like to acknowledge Chen-Hsuan Lin for providing his SIFT code and for many useful discussions when developing this package. I would also like to thank Prof. Simon Lucey for his guidance.