CLM-WILD Face Tracker

NEW! Our new tracker CHEHRA is now available for download.

Facial landmark detection under generic settings (i.e. the subject being tracked is not in the training set) is an extremely challenging problem that gets compounded in presence of external factors such as illumination changes, pose variation and occlusion. Therefore, the goal of this project is to build fully automatic facial landmark detection system capable of handling faces under uncontrolled natural setting (i.e. 'wild' faces).

For any further questions, please email me at

The following codes are made available strictly for non-commercial and academic purposes.

DRMF Matlab Fitting Code

About the code
  • Fully Automatic System.
  • Also contains a robust face detector, suitable for 'wild' faces, based on the Zhu and Ramanan (CVPR 2012) Face Detector.
  • Detects 66 Facial Landmark Points.
  • Detects Rough 3D Head-Pose.
  • Supports the functionality for incorporating your own preferred face detector.
  • Supports the functionality for directly using the pre-defined face detection bounding boxes.

If you use any part of this code, please cite the following papers:

[1]    Akshay Asthana, Stefanos Zafeiriou, Shiyang Cheng and Maja Pantic.
        Robust Discriminative Response Map Fitting with Constrained Local Models
        In Proc. of 2013 IEEE Conference on Computer Vision and Pattern Recognition (CVPR 2013)Portland, Oregon, USA, June 2013.

In the news

CLM-WILD Face Tracker Demo

Demo 1 - With Glasses

Demo 1 - No Glasses

Demo 1 - Occlusion

Demo 2

Demo 3 - J K Rowling