Home | Education | Publications | Projects | Experience | Writing

Vahid Kazemi

E-Mail: vahidk[@]kth.se
Address: Teknikringen 14/608, Stockholm 10044, SWEDEN
Phone: +46 (0)8 790 6905

Education

PhD, Computer Science
Kungliga Tekniska Hogskolan, Stockholm, 2011-Present

Master of Systems, Control, and Robotics
Kungliga Tekniska Hogskolan, Stockholm, 2009-2010

Bachelor of Computer Science
Shahid Beheshti University, Tehran, 2001-2007

Publications

Face Alignment with Part-Based Modeling [Oral]
V. Kazemi and J. Sullivan
In Proc. IEEE British Machine Vision Conference (BMVC 2011), Dundee, Scotland, Sept 2011.

Projects/Research

Articulated Human Pose Estimation
I'm currently working under supervision of Stefan Carlsson, and Josephine Sullivan at KTH on articulated human pose estimation for motion capture of football players. The final goal of the project is to create a complete reconstruction of a football game based on a set of 2D image sequences.

The left image shows how multiple weak part models can be combined to create a single strong detector.
Face/Object Detector
A face/object detector based on VJ algorithm (Boosting on HAAR features). We have improved the performance drastically by using mid cumulative probability distribution point as threshold for weak classifiers.
Download Matlab Code (45KB)

Autonomous Robot
In this project we (a group of 4) built an autonomous mobile robot from scratch. I as the technical lead of the group worked on different sections of project including localization, planning, mapping and object detection. Our robot managed to gain the maximum points in the class at final competition.
Download C++ Code and Binary (4.63MB)
View Technical Report

Face Alignment with Part-Based Modeling
Under supervision of Stefan Carlsson, and Josephine Sullivan at KTH, I'm investigating different methods of constructing a view independent statistical model of human face. As a direct application, the model can be used to extract or capture generic features (e.g. expression, age,...) of a target face. Furthermore the model can be useful for segmentation, validation, and recognition of faces.
View Project Page

Digital Makeup
In a research project with cooperation of Ivan Laptev and Josef Sivic we have investigated an automated system to apply makeup extracted from a single template image to an arbitrary face image. It was done in several steps, starting from detecting the face boundary (using Boosted Histograms), finding facial features, matching interest points to template image (using Thin-Plate Splines), decomposing the image to different layers, dense grid histogram transfer, and at the end pixel wise detail transfer. The results as you can see are quite comparable to supervised methods which involve human interaction.
Flight Simulator Visualizer
The main goal of this project was to design a rendering engine for use in flight simulator projects. I have implemented a deferred rendering engine which is capable of rendering thousands of dynamic lights in the scene.

For more 3D graphics related sample works please visit my website: http://www.gameprogrammer.org

Work Experience

R&D Programmer
OculusAI 2008-2009, Stockholm
Development and research on state of the art Computer Vision algorithms for image classification and object recognition. OculusAI is a new born company providing image validation services targeting social network and dating websites. In OcolusAI I have worked on serveral projects regarding facial feature recognition and celebrity detection.

Lead Programmer
Puya Arts Software Inc., 2004-2008, Tehran
I have been working as the lead programmer on Quest of Persia game series. I have worked on several different parts, including implementation of character control/movement, NPC behavior and AI, camera, particle systems and effects. In the latest project beside software architecture and design of the game I roughly coded about 20,000 lines of code of the game.