説明

CLIP can predict the most relevant text snippet, given an image, without directly optimizing for the task (zero-shot capabilities).