video captioning without human labels or large teacher models, improving accuracy and efficiency by self-generating and refining captions automatically.