Abstract: We tackle the challenging task of unsupervised object localization in this work. Recently, transformers trained with self-supervised learning have been shown to exhibit object localization ...

Abstract: Multiple extended objects (EOs) tracking has attracted a lot of attention due to the fast development of high-resolution sensors. A remarkable feature of EOs, compared to the traditional ...

TL; DR. We propose a novel perspective to regard the multiple object tracking task as an in-context ID prediction problem. Given a set of trajectories carried with ID information, MOTIP directly ...