pfshawn / rcpr

C++ code for Sketch and RGB Image Face Alignment in the Wild

RCPR for RGB and Sketch Face Alignment.

C++ code for Robust Cascaded Pose Regression using Random Ferns, traine on both RGB and sketch images. If you use this code please cite my paper as well as the orignal RCPR paper.

*Heng Yang, C. Zou and Ioannis Patras, "Face sketch landmarks localization in the wild", IEEE Signal Processing Letters, 2014. **

If you have any problems running this code, please feel free to contact Heng Yang (yanghengnudt AT gmail.com)

The original RCPR code was written in matlab which limits its run-time speed performance. I transformed the testing code into C++ and using a python interface. I trained the model fern.dat as described in the above paper.

Requirements: --OpenCV 2.4 --boost 1.55 --Tested on Ubuntu 12.04

------------------Optional: for creating the model file-------------------------------------

!!SKIP this step if you do not want to create your own model but only to use the trained model I have created.!!

Dowload the matlab code from here: http://www.vision.caltech.edu/xpburgos/ICCV13/ First, I transform the trained model [regModel.mat] into many .txt file (might be not optimal but it works) and store them in a directory called ferns. A cpp program is used to serialize the txt files into a fern class object file, called fern.dat, which is the model file. You can directly use the fern.dat and ignore the following steps.

1. load regModel

2. mkdir ferns

3. run save_regmodel in matlab (it takes a few minutes)

4. make -f Makefile_Serialize

5. $./RCPRSerializer

---

---

For running the model.

$ make

RCPR will be created

Then run

$./RCRP Model/fern.dat yourfilelist.txt

e.g.: ./RCPR Model/fern.dat img.txt

it will save the result to result_img.txt when it finishes