Transcription of Abstract - arXiv
{{id}} {{{paragraph}}}
Example: barber
Abstract - arXiv
Transformer Meets Tracker:Exploiting Temporal Context for Robust Visual TrackingNing Wang1 Wengang Zhou1,2 Jie Wang1,2 Houqiang Li1,21 CAS Key Laboratory of GIPAS, EEIS Department, University of Science and Technology of China (USTC)2 Institute of Artificial Intelligence, Hefei Comprehensive National Science video object tracking , there exist rich temporal con-texts among successive frames, which have been largelyoverlooked in existing trackers. In this work, we bridge theindividual video frames and explore the temporal contextsacross them via a transformer architecture for robust objecttracking. Different from classic usage of the transformer innatural language processing tasks, we separate its encoderand decoder into two parallel branches and carefully designthem within the Siamese-like tracking pipelines. The trans-former encoder promotes the target templates via attention-based feature reinforcement, which benefits the high-qualitytracking model generation. The transformer decoder prop-agates the tracking cues from previous templates to the cur-rent frame, which facilitates the object searching transformer-assisted tracking framework is neat andtrained in an end-to-end manner.
we briefly review the recent popular tracking approaches for the sake of completeness. As shown in Figure3, the mainstream tracking methods such as Siamese network [1] or discriminative correlation filter (DCF) [46,7,3] can be formulated into the Siamese-like pipeline, where the top branch learns the tracking model using templates and the
Loading..
Tags:
Information
Domain:
Source:
Link to this page:
Please notify us if you found a problem with this document: