Project Suggestions

Faces in the Wild: Multi-Person Face Recognition Using Kernel SVM and OSS Kernel

Project Goal:
	Implement the face recognition algorithm described in http://www.cs.tau.ac.il/~wolf/papers/osskernel.pdf and test it on LFW data set. You should recreate the results reported in the paper (or to be close to them) Download the aligned faces here. The LBP implementation can be downloaded here. OSS kernel website. LIBSVM for SVM training and classification: download.

Exemplar Models for Object Detection

Project Goal:
	Implement exemplar LDA object detection system and test it on PASCAL 2007 In this project you can use the exemplar SVM system implementation as your basis and replace the SVM with LDA as described in here. ESVM project webpage (contains a link to the code and the relevant publications) See the instructor for details.

Discriminative Decorrelation for Clustering

Project Goal:

Implement and test clustering in WHO space and object detection using the models trained on the clusters. Use PASCAL 2007 for your tests. The implementation details can be found here. For the detection part you can use the exemplar SVM system implementation as your basis, use single exemplar mode, and replace the eSVM model with the classifier trained on a cluster. ESVM project webpage (contains a link to the code and the relevant publications) See the instructor for details.

Part-Based Random Face Synthesis

Project Goal:

Given a collection of faces, create new faces as a collage of facial parts obtained from the given collection. The patches should be blended together seamlessly and have a photographic quality. The goal is to create as much faces as possible, such that the faces could be classified (by a human) as different people. The data bases of annotated faces can be found here and here See the following link for ideas on stitching and blending between patches: http://cs.brown.edu/courses/csci1950-g/asgn/proj2/ http://cs.brown.edu/courses/csci1950-g/asgn/proj3/ http://cs.brown.edu/courses/csci1950-g/asgn/proj4/

Synthesis of Photographic Quality Faces using Statistical Shape and Appearance models

Project Goal:

Design and implement an algorithm that takes a collection of images annotated with facial landmarks, builds a statistical model of appearance and shape variation in this set and synthesizes novel faces using evolutionary algorithms (or a different kind of an algorithm). The basic steps are described here. Note that in this project we are not interested in creating faces similar to the target face (as in the link). The goal is to create many different random faces. The data bases of annotated faces can be found here and here See website for details on shape statistical model. See website for details on combined appearance models. See website describing how to build, display and use statistical appearance models.

Interactive Recognition

Project Goal:
	The goal of this project is interactive recognition of gestures using Kinect camera. You can get many ideas for projects that use the Kinect from the following sites: OpenKinect, or KinectHacks. Note: The camera will NOT be provided. You can choose this project only if you have your own Kinect camera.