Figure 13.

Two schematic situations in which perfect prediction would not preclude a tracking error. The true locations of a target (blue) and a distractor (gray) are shown, together with unlabeled samples (yellow). (Note that in MOT, the targets and nontargets share the same color). (a) Noisy observations lead to two potential correspondences (one correct and one incorrect) with equal probability. (b) Noisy observations lead to an incorrect correspondence appearing more likely.