From object difficulty to image scoring: A strategy for active learning in object detection

doi:10.1016/j.knosys.2026.115946

From object difficulty to image scoring: A strategy for active learning in object detection

2026-04-06

0

PRE

AI

DOI：10.1016/j.knosys.2026.115946

Original

Original request for help

Save

Abstract

En 中文

• Present a feature-driven framework that reduces labeling costs in object detection. • Propose a unified measure capturing both classification and localization difficulty. • Achieve consistent accuracy gains of 0.27%–1.99% across standard benchmarks. • Deliver a scalable and efficient selection process, running in 0.06–0.63 s per image.

Keywords:

object detection

active learning

labeling cost

difficulty measurement

image scoring

From object difficulty to image scoring: A strategy for active learning in object detection

Abstract

Journal

Organization