Baseline for 2D Object Detection and Tracking Cascade R-CNN X152 and SORT
This paper explores efficient and reliable baseline for 2D tracking in autonomous driving: implementation of a tracking-by-detection framework only using combination of existing techniques: Cascade R-CNN X152 and Simple Online and Realtime Tracking. Our detector and tracker has been trained and tuned on Waymo open dataset, and finished in 3rd place of Waymo open dataset challenge 2D tracking track. The code is available at https://github.com/xuyuan/waymo_2d_tracking.
Waymo Open Dataset Challenge Reports, Workshop on Scalability in Autonomous Driving, CVPR20