Counterfactual Co-occurring Learning for Bias Mitigation in Weakly-supervised Object Localization

arXiv:2305.15354v2 Announce Type: replace
Abstract: Contemporary weakly-supervised object localization (WSOL) methods have primarily focused on addressing the challenge of localizing the most discriminative region while largely overlooking the relatively less explored issue of biased activation -- incorrectly spotlighting co-occurring background with the foreground feature. In this paper, we conduct a thorough causal analysis to investigate the origins of biased activation. Based on our analysis, we attribute this phenomenon to the presence of co-occurring background confounders. Building upon this profound insight, …

abstract arxiv bias challenge counterfactual cs.cv feature issue localization object paper type weakly-supervised