Our Team Win the BMTT Challenge Workshop in CVPR 2020
June 11, 2020
Our team representing the University of Washington is the Winner of Track 3 (multi-object tracking and segmentation in KITTI-MOTS and MOTS20 dataset with public detection) and the Runner-up of Track 2 (multi-object detection, tracking and segmentation in KITTI-MOTS dataset) the 5th BMTT Challenge workshop in CVPR 2020.
IEEE/CVF CVPR 2020: 5th BMTT MOTChallenge Workshop
- Track 1 (MOTSChallenge): Multi-Object Tracking and Segmentation for crowded pedestrian scenes on the MOTS20 dataset (pedestrians only).
- Track 2 (KITTI-MOTS): Multi-Object Tracking and Segmentation for road traffic scenarios on the KITTI MOTS dataset (cars and pedestrians).
- Track 3 (Supplied masks): Multi-Object Tracking using the supplied segmentation masks on both the MOTS20 and the KITTI MOTS datasets (cars and pedestrians).
Track 1&3 Solution: IA-MOT
In this work, we propose “instance-aware MOT” (IA-MOT), that can track multiple objects in either static or moving cameras by jointly considering the instance-level features and object motions. Evaluated on the MOTS20 and KITTI-MOTS dataset, IA-MOT won the first place in Track 3 of the BMTT Challenge in CVPR2020 workshops.
Track 2 Solution: LIFTS
In this work, we propose “Lidar and monocular Image Fusion based multi-object Tracking and Segmentation (LIFTS)” for multi-object tracking and segmentation (MOTS). Evaluated on KITTI-MOTS dataset, LIFTS achieves a 79.6 sMOTSA for Car and 64.9 for Pedestrian, with the second place ranking in the competition.
