A robust short-term tracker with redetectionDžubur, Benjamin (Avtor) Kristan, Matej (Mentor) Lukežič, Alan (Komentor) computer visionvisual trackingvideo segmentationlong-term trackingState-of-the-art long-term visual object tracking methods are limited to predicting target position as an axis-aligned bounding box. Segmentation-based trackers exist, however they do not address long-term disappearances of the target. Thus, by upgrading a short-term segmentation-based tracker with the capability of redetecting a lost target, we develop a new discriminative single shot segmentation tracker -- D3SLT, which is capable of long-term tracking in addition to recovering from short-term tracking failures.We upgrade the previously developed short-term D3S tracker with a global redetection module, based on an image-wide discriminative correlation filter response and Gaussian motion model. An online learned confidence estimation module robustly estimates target disappearance. An additional backtracking module enables recovery from tracking failures and further improves tracking performance. On the bounding box based VOT-LT2021 Challenge, D3SLT achieves F-score of 0.667, while on LaSOT it achieves success of 0.616 and normalized precision of 0.692. D3SLT achieves results close to those of state-of-the-art long-term trackers while additionally outputting segmentation masks.20222022-09-12 08:00:00Magistrsko delo/naloga140153sl