language-image

Definition

A compound signaling that both text and images are handled and aligned with each other, as in contrastive language-image pretraining. The payoff of aligning them in one space is cross-modal use: retrieve images by a description, score how well a caption matches a picture, or condition generation on text. It is near-synonymous with vision-language, though this word order is more common in contrastive-alignment contexts than in generative multimodal ones.