The paper presents a novel tracking method called Tracker, which leverages the bounding box regression capabilities of an object detector to perform multi-object tracking without specialized training or optimization on tracking data. The Tracker converts an object detector into a tracker by predicting the position of objects in the next frame based on the regression output of the detector. The authors demonstrate that Tracker achieves state-of-the-art performance on three multi-object tracking benchmarks by extending it with a re-identification Siamese network and a motion model. They also conduct an analysis of failure cases and challenging tracking scenarios, showing that no dedicated tracking methods outperform Tracker in complex scenarios such as small and occluded objects or missing detections. The paper concludes by proposing Tracker as a new tracking paradigm and highlighting future research directions, emphasizing the need to focus on remaining complex tracking challenges.The paper presents a novel tracking method called Tracker, which leverages the bounding box regression capabilities of an object detector to perform multi-object tracking without specialized training or optimization on tracking data. The Tracker converts an object detector into a tracker by predicting the position of objects in the next frame based on the regression output of the detector. The authors demonstrate that Tracker achieves state-of-the-art performance on three multi-object tracking benchmarks by extending it with a re-identification Siamese network and a motion model. They also conduct an analysis of failure cases and challenging tracking scenarios, showing that no dedicated tracking methods outperform Tracker in complex scenarios such as small and occluded objects or missing detections. The paper concludes by proposing Tracker as a new tracking paradigm and highlighting future research directions, emphasizing the need to focus on remaining complex tracking challenges.