Mind-the-Glitch: Visual Correspondence for Detecting Inconsistencies in Subject-Driven Generation

Abdelrahman Eldesokey, Aleksandar Cvejic, Bernard Ghanem, Peter Wonka

Abstract

Introduces a method for detecting visual inconsistencies in subject-driven image generation by leveraging visual correspondence, improving reliability for diffusion-based generative pipelines that must preserve a subject's identity across edits.