Technical articles · 2020.10.08
Estimating and evaluating object-detection uncertainty
Unsicherheitsabschätzung bei der Objekterkennung – eine erfolgsentscheidende Aufgabe
Overview
Reviews Bayesian and sampling-based uncertainty approaches and detectors that also predict localization variance. An EfficientDet-D0 example compares confidence with spatial uncertainty, and the article discusses PDQ as a way to consider label and spatial quality together.