To solve the above problems, we propose a novel zero-shot goal-directed scanpath prediction model named CLIPGaze. We use CLIP to extract pre-matched features for the target prompt and input image, ...