Implements the cascaded regressor for face alignment described in the paper: Cao, Xudong, et al. "Face Alignment by Explicit Shape Regression." Computer Vision and Pattern Recognition (CVPR), 2012 IEEE Conference on. IEEE, 2012.
The training images are assumed to all come in same size. These images should be the output of a face detector. Specify parameters and dataset information in the config.m file such as number of fiducial points, index of left eye, right eye etc.
Store the training and test images along with their ground truth fiducial location in a mat file in the format:
- Is: nRow x nCol x nImg matrix
- Sgts: 2 x nPoints x nImg matrix
Then run
do_esr('config')