ResearchBenchmarksCounterfactual Hallucination Detection on ROCO v2Follow74.9Pairwise AccuracySoft-MSD71.57272.43673.374.164May 7, 2026Evaluation ResultsMethodMethodLinksPairwise AccuracySoft-MSDBackbone=BiomedCLIP [44]Backbone=BiomedCLIP [44]2026.0574.9MSDBackbone=BiomedCLIP [44]Backbone=BiomedCLIP [44]2026.0573.3CLIPScoreBackbone=BiomedCLIP [44]Backbone=BiomedCLIP [44]2026.0571.7