Track to Detect and Segment: An Online Multi-Object Tracker

CVPR 2021  ยท  Jialian Wu, Jiale Cao, Liangchen Song, Yu Wang, Ming Yang, Junsong Yuan ยท

Most online multi-object trackers perform object detection stand-alone in a neural net without any input from tracking. In this paper, we present a new online joint detection and tracking model, TraDeS (TRAck to DEtect and Segment), exploiting tracking clues to assist detection end-to-end. TraDeS infers object tracking offset by a cost volume, which is used to propagate previous object features for improving current object detection and segmentation. Effectiveness and superiority of TraDeS are shown on 4 datasets, including MOT (2D tracking), nuScenes (3D tracking), MOTS and Youtube-VIS (instance segmentation tracking). Project page: https://jialianwu.com/projects/TraDeS.html.

PDF Abstract CVPR 2021 PDF CVPR 2021 Abstract

Results from the Paper


Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Result Benchmark
Multi-Object Tracking DanceTrack TraDes HOTA 43.3 # 31
DetA 74.5 # 23
AssA 25.4 # 30
MOTA 86.2 # 26
IDF1 41.2 # 31
Multi-Object Tracking MOT15 Baseline+MFW MOTA 66.5 # 1
Multi-Object Tracking MOT16 TraDeS MOTA 70.1 # 9
IDF1 64.7 # 8
Online Multi-Object Tracking MOT16 TraDeS MOTA 67.7 # 2
Multi-Object Tracking MOT17 TraDeS MOTA 69.1 # 30
IDF1 63.9 # 34
Multi-Object Tracking MOTS20 TraDes sMOTSA 50.8 # 5
IDF1 58.7 # 3
Instance Segmentation nuScenes TraDeS MOTA 68.2 # 1
Video Instance Segmentation YouTube-VIS validation TraDeS mask AP 32.6 # 47
AP50 52.6 # 46
AP75 32.8 # 48

Methods