AI BENCHMARK PROFILE

COMEX

General AIMultimodal Perception

COMEX is a benchmark for explainable aesthetic image cropping containing 33,161 quadruples of expanded image, crop box, composition category, and composition-grounded explanation. It supports joint evaluation of crop localization, composition understanding, and explanation generation.

Released: 2026-08-04
Readiness: Paper only
Primary field: General AI

Why it matters

Provides a structured evaluation of crop-and-explain methods, enabling comparison across models on composition-grounded reasoning and explanation generation, addressing the gap of evaluating explainability in image cropping.

Motivation

Explainable aesthetic image cropping requires not only localizing a visually pleasing crop but also explaining why it is preferred.

Primary resources

Paper ↗Hugging Face ↗

Benchmark Radar records only publicly supported details and links back to primary sources for verification.