ML-project

A repo for EECS 349 Machine Learning project for Northwestern University

In this project, we try to use AAM(Active Appearance Model) and ESR(Explicit Shape Regression)[1] to train the face alignment model and use this trained model to do the face alignment job.

Dataset

We use the following dataset which are commonly used by face alignment research:

1. IBUG[2,3]
2. COFW[4]

Preprocess

1. Convert all the raw image into grayscale image
2. Use the face detector provided by OpenCV to pre-compute the bouding box for alignment
3. Extract the coordinates of landmark from the dataset

Our Work

1. Implement ESR algorithm to train an ESR model
2. Use an existing AAM library to train an AAM model
3. Optimize our ESR implementation by tuning and using better pixel selection strategy
4. Study how the parameters influence the performance of ESR algorithm
5. Compare the results of ESR and AAM
6. Build up a demo using ESR on web page
7. Complete the final report and present out results

Collaborator

Yang Yang, Haomin Hu, Can Wang, Lijun Tang

Reference

[1] Cao, Xudong, et al. "Face alignment by explicit shape regression." International Journal of Computer Vision 107.2 (2014): 177-190.

[2] C. Sagonas, G. Tzimiropoulos, S. Zafeiriou, M. Pantic. A semi-automatic methodology for facial landmark annotation. Proceedings of IEEE Int’l Conf. Computer Vision and Pattern Recognition (CVPR-W’13), 5th Workshop on Analysis and Modeling of Faces and Gestures (AMFG '13). Oregon, USA, June 2013.

[3] C. Sagonas, G. Tzimiropoulos, S. Zafeiriou, M. Pantic. 300 Faces in-the-Wild Challenge: The first facial landmark localization Challenge. Proceedings of IEEE Int’l Conf. on Computer Vision (ICCV-W 2013), 300 Faces in-the-Wild Challenge (300-W). Sydney, Australia, December 2013.

[4] Burgos-Artizzu, Xavier P., Pietro Perona, and Piotr Dollár. "Robust face landmark estimation under occlusion." Computer Vision (ICCV), 2013 IEEE International Conference on. IEEE, 2013.