UrbanOmniDetect
Monocular 3D object detection with the UrbanOmniDetect system and UrbanOmniView dataset.
UrbanOmniDetect is a calibration-free, view-agnostic monocular 3D object detection framework for urban scenes. It targets a common deployment bottleneck in V2X and infrastructure sensing: Camera intrinsics may be unavailable, imprecise, or drifting.
The work was presented as an oral paper at the CVPR 2026 DriveX workshop and is paired with UrbanOmniView, a dataset combining real-world driving data, infrastructure-camera data, and high-fidelity synthetic data.
- Role: Lead author
- Recognition: Oral presentation at the CVPR 2026 DriveX workshop
- Keywords: Monocular 3D detection, ordered 3D box-vertex projection, heterogeneous camera viewpoints, synthetic data, infrastructure sensing, calibration-free perception
Links: GitHub, Hugging Face model, UrbanOmniView dataset, CVPR Open Access, DriveX workshop.
