CROP: Expert-Aligned Image Cropping via Compositional Reasoning and Optimizing Preference

CROP proposes a compositional reasoning framework for aesthetic image cropping that moves beyond saliency prediction and retrieval methods by learning to understand composition and aesthetics principles for adaptive spatial cropping decisions.