Debiasing Text-to-Image Evaluation via Implicit Cultural Alignment Reward Modeling

Bo-An Chang, Yu-Chih Chen

As Text-to-Image (T2I) systems rapidly advance, evaluating the cultural authenticity of synthesized content has become increasingly important for fair and trustworthy generative AI…