Distributed GNN training on contemporary massive and densely connected graphs requires information aggregation from all neighboring nodes, which leads to an explosion of inter-server communications. This paper proposes AdaGL, a highly scalable end-to-end framework for rapid distributed GNN training. AdaGL novelty lies upon our adaptive-learning based graph-allocation engine as well as utilizing multi-resolution coarse representation of dense graphs. As a result, AdaGL achieves an unprecedented level of balanced server computation while minimizing the communication overhead. Extensive proof-of-concept evaluations on billion-scale graphs show AdaGL attains ∼30−40% faster convergence compared with prior arts.

We present a new solution for real-time head pose estimation. The key to our method is a model-based approach based on the fusion of color and time-of-flight depth data. Our method has several advantages over existing head-pose estimation solutions. It requires no initial setup or knowledge of a pre-built model or training data. The use of additional depth data leads to a robust solution, while maintaining real-time performance. The method outperforms the state-of-the art in several experiments using extreme situations such as sudden changes in lighting, large rotations, and fast motion.

Animating in-game avatars using real time motion capture data is highly appealing and is becoming more widespread due to the accessibility of researchers and developers to consumer priced depth sensors [Microsoft Kinect 2010]. Depth sensors allow for a cheap and robust motion capture solution which can be naturally adapted to games. However, in spite of the many advantages of using real-time motion capture data for animating avatars in games, there are two major challenges. The first is due to the limitations of the current tracking techniques in producing smooth, noise free, accurate animation in real-time. The second, and more acute problem, stems from the fact that in most games, the movements of the animated avatar are expected to be more expressive than the player's actual movements. In such cases, one would like to visually enhance the player's motion to display exaggerated or even super-natural motions.

We present a robust framework for tracking skeleton joints in real-time by using a single time-of-flight depth sensor. The framework is able to remove the background noise inherent in time-of-flight cameras, detect multiple people, and track up to 30 joints of free motion for each person. The approach has several advantages over traditional motion capture, as it is a cheap alternative to magnetic and optical systems, and requires no markers whatsoever. Unlike markerless systems based on RGB cameras [Duetscher et al. 2000; Kehl and Van Gool 2006], our framework yields dependable results at an interactive rate using a single camera.