The new VOT2020 short-term tracking performance evaluation protocol and measures

This document is a brief overview of the new performance evaluation protocol and the new measures introduced in Visual Object Tracking challenge VOT2020 for evaluation of short-term trackers. The new protocol avoids tracker-dependent resets and reduces the variance of the performance evaluation measures – the VOT basic measures, accuracy (A) and robustness (R) and the primary performance measure expected average overlap (EAO) are re-defined. The presented performance evaluation methodology should be considered as the new default VOT methodology for testing short-term trackers in VOT2020 and the challenges to follow. Errata: This is an updated version of the document originally published at the VOT page on 31.3.2020. In effort to keep the new measures as similar as possible to those calculated in the previous VOT challenges, the measures have been slightly modified since initial post. The VOT toolkit has been updated with the modification in the measures as well. In particular, the robustness measure did not change, the accuracy measure normalization is corrected and in the EAO measure, all sub-sequences are equally weighted.