Presentations · 2026.09.16
Estimating localization quality from object-detector outputs
Output Monitoring for Object Detector
Overview
The poster presents a monitor that estimates each detection’s 2D IoU from box scores, geometry and frame statistics. In an FCOS3D pedestrian-detection example using ZOD data, gradient-boosted trees estimate localization quality and a threshold identifies candidate false positives.
Publication and version information
One-page research poster in the materials for the 16 September 2026 IAA Transportation midterm event.