Brian Cheong
MASc student and computer vision researcher in autonomous vehicles
Summary
Computer vision researcher focused on LiDAR-based multi-object tracking and joint detection-and-tracking methods. ecva+1
Academic contributor with multiple peer-reviewed conference publications (ECCV, ICCV, WACV) and open-source code releases that accompany research. ecva+2
Student leader and systems integrator in autonomous vehicle competitions, having served as autonomy technical director and various lead roles on the aUToronto team that won the 2024 AutoDrive Challenge II. utoronto+1
Open-source practitioner who publishes code and reproducible research artifacts (repositories and project websites) to support academic work. github+2
Work
Education
Projects
Writing
SCATR: Mitigating New Instance Suppression in LiDAR-based Tracking-by-Attention via Second Chance Assignment and Track Query Dropout
January 1, 2026WACV 2026 paper describing SCATR, which introduces Second Chance Assignment and Track Query Dropout training strategies to improve LiDAR-based tracking-by-attention methods and close the gap with tracking-by-detection.
ForeSight: Multi-View Streaming Joint Object Detection and Trajectory Forecasting
January 1, 2025ICCV 2025 paper introducing ForeSight, a joint detection and forecasting framework for multi-view streaming vision-based 3D perception that shares query memory between detection and forecasting to improve temporal consistency and forecasting-aware detection.
JDT3D: Addressing the Gaps in LiDAR-Based Tracking-by-Attention
January 1, 2024ECCV 2024 paper presenting JDT3D, analyzing why LiDAR-based tracking-by-attention lags behind tracking-by-detection and proposing augmentations and propagation strategies to improve performance on nuScenes.