At this seminar, we will discuss a paper that utilizes reinforcement learning techniques to train convolutional and recurrent networks in the task of visual object tracking.
The proposed model can directly use both spatial and temporal link between video frames and optimize its accuracy to predict an object's long-term trajectory.
Thanks to its simplicity, this algorithm can run with high framerate keeping excellent tracking accuracy.
Speaker: Beliaev Vladislav
Presentation language: Russian.
Date and Time: October 8th, 18:30-20:00.
Place: Times, room 204.