Alan Lukežič presents Tracking-by-segmentation: state-of-the-art and beyond

On 2021-07-15 11:00:00 at
zoom link:

Visual object tracking remains one of the fundamental problems in computer
vision, which progressed significantly in the last decade. The progress has been
facilitated by the advent of deep learning as well as by emergence of moderated
datasets and challenges. Traditional trackers reported target location in form
of bounding boxes. However, in the last two years, tracking has started to
converge with a closely related field of video object segmentation. We are now
witnessing emergence of trackers that segment the target, while by far
outperforming their bounding-box-oriented predecessors. In this talk, I will
overview two major directions in segmentation-based tracking and an emerging
methodology we have detected in the most recent VOT2021 challenge.
Responsible person: Petr Pošík