Computer vision, autonomous driving, 3D perception, multi-object tracking, and sensor fusion.
For a complete list, see my Google Scholar and ResearchGate.
A Unified 3D Object Perception Framework for Real-Time Outside-In Multi-Camera Systems
arXiv preprint
* Equal contribution
Vision meets mmWave Radar: 3D Object Perception Benchmark for Autonomous Driving
IEEE Intelligent Vehicles Symposium (IV)
Split and Connect: A Universal Tracklet Booster for Multi-Object Tracking
IEEE Transactions on Multimedia
Multi-Target Multi-Camera Tracking of Vehicles by Graph Auto-Encoder and Self-Supervised Camera Link Model
WACV Workshops
ROD2021 Challenge: A Summary for Radar Object Detection Challenge for Autonomous Driving Applications
ACM ICMR
A deep radar object detection network (RODNet) that detects objects purely from processed radar frequency data (range-azimuth heatmaps). Instead of burdensome human labels, RODNet is trained with annotations generated automatically via a camera-radar fusion (CRF) strategy, achieving strong detection without a camera at inference time. Released alongside the CRUW dataset and the ROD2021 Challenge.
A robust method for 3D localization of monocular visual objects in road scenes, jointly integrating depth estimation, adaptive ground-plane estimation, and multi-object tracking. Important for autonomous driving and advanced driver-assistance systems (ADAS) where 3D cues are hard to obtain from a single camera.
A super-resolution approach for GIFs that uses two high-resolution frames (first and last) together with the low-resolution sequence to reconstruct a high-resolution GIF. Paired with GIFSR, the first dataset devoted to GIF super-resolution. Significantly outperforms popular video SR baselines while achieving at least 80x speedup on CPU.
Annual large-scale benchmarks advancing computer vision for intelligent transportation and smart spaces — including multi-camera 3D perception, multi-target multi-camera tracking, and synthetic-to-real (Sim2Real) generalization.
Visit →
A public camera-radar dataset for autonomous driving, with FMCW radar in the form of radio-frequency (RF) tensors synchronized and calibrated with camera frames. It underpins the RODNet work and the ROD2021 Challenge.
Visit →