Title: Re3: Real-Time Recurrent Regression Networks for Object Tracking
Advisors: Dieter Fox and Ali Farhadi
Abstract: Robust object tracking requires knowledge and understanding of the object being tracked: its appearance, its motion, how it changes over time. It also must be able to modify its underlying model and adapt to new observations. We present Re3, the first real-time deep object tracker capable of incorporating long-term temporal information into its model. In line with other recent deep learning techniques, we do not train an online tracker. Instead, we use a recurrent neural network to represent the appearance and motion of the object. We train the network to learn how object appearance and motion changes offline, letting it track with a single forward pass at test time. This lightweight model is capable of tracking objects at 150 FPS, while still attaining competitive results on challenging benchmarks. We also present experiments that directly target how occlusions affect tracker accuracy, and we show that our method handles occlusions better than other comparable trackers.