Small object matters in weakly supervised object localization

Dongjun Hwang, Seong Joon Oh, Junsuk Choe

Abstract: Highlights•We tackle the problem that WSOL methods are less effective for localizing small objects.•We propose new evaluation metrics and a dataset to properly measure the localization performances of small objects.•We propose a novel consistency learning framework to zoom-in on small objects so that the model can see the objects more clearly.•The proposed method significantly improves the small object localization on four different backbone networks and three different datasets.

External IDs:doi:10.1016/j.neucom.2025.130494