Abstract
Multi-object tracking (MOT) aims at estimating bounding boxes and identitiesof objects in videos. Most methods obtain identities by associating detectionboxes whose scores are higher than a threshold. The objects with low detectionscores, e.g. occluded objects, are simply thrown away, which bringsnon-negligible true object missing and fragmented trajectories. To solve thisproblem, we present a simple, effective and generic association method, calledBYTE, tracking BY associaTing Every detection box instead of only the highscore ones. For the low score detection boxes, we utilize their similaritieswith tracklets to recover true objects and filter out the backgrounddetections. We apply BYTE to 9 different state-of-the-art trackers and achieveconsistent improvement on IDF1 score ranging from 1 to 10 points. To putforwards the state-of-the-art performance of MOT, we design a simple and strongtracker, named ByteTrack. For the first time, we achieve 80.3 MOTA, 77.3 IDF1and 63.1 HOTA on the test set of MOT17 with 30 FPS running speed on a singleV100 GPU. The source code, pre-trained models with deploy versions andtutorials of applying to other trackers are released at\url{https://github.com/ifzhang/ByteTrack}.