Image Generation Evaluation
Overview
Image generation evaluation asks whether generated images look realistic, match prompts and references, preserve identities, and satisfy constraints such as spatial layout or readable text. Common approaches combine distributional metrics such as Fréchet Inception Distance and precision–recall, vision-language and feature-based scores, human judgments, and benchmarks designed for specific tasks. Recent work expands evaluation to fine-grained capabilities—including multi-subject identity and attribute binding, editing consistency, world plausibility, artistic style, and medical-image utility—and uses multimodal models to provide scalable judgments or explanations. A central challenge is making automated scores reliable across datasets and domains: a single metric can miss important failures or favor one quality dimension, while human and expert assessments are costly and difficult to standardize.
Latest papers
113 papers18 papers in the last four weeks, up 80% on the four weeks before. 0.1% of all new papers.