Bugra Tekin awarded a Qualcomm Innovation Fellowship 2017-18
© 2017 EPFL
Bugra Tekin, an IC doctoral student in the Computer Vision Laboratory, has been awarded the Qualcomm Innovation Fellowship for 2017-18. He was among the three winners selected from top European Universities and the only winner from a Swiss University this year.
Bugra’s research project “Real-time 3D Human Pose Estimation in the Wild” will focus on developing new, efficient machine learning algorithms to estimate human 3D poses, which may have applications in various areas such as gaming, augmented reality, human-computer interaction, security and health-care.
Human 3D pose data is obtained by annotating the joint positions of a person in 3D. However, annotating in 3D is a very challenging task; it is not possible simply to click on an image to get 3D pose data, as is the case in 2D. One way to obtain this data is to use special optical sensors in a controlled studio setting. Another way is to generate this data automatically by synthesizing artificial training images with known 3D poses.
The project aims to address the current problem of lack of training data captured in uncontrolled outdoor environments, as opposed to artificial laboratory settings such as motion capture studios, required for the deep neural networks to successfully estimate realistic human 3D poses.
To tackle this problem, Bugra aims to render large amounts of 3D synthetic data and apply domain adaptation techniques to account for the shift between real and synthetic data. He also plans to leverage 2D pose annotations, at the lack of 3D pose data, to develop weakly supervised machine learning algorithms, namely algorithms that will infer a 3D pose with only partial 2D information.
The Qualcomm Innovation Fellowship program was set up to recognize, reward, and mentor innovative PhD students and foster new ideas, which ultimately lead to future technology advancements and growth.