Text this: Towards label-free defect detection in additive manufacturing via dual-classifier semi-supervised learning for vision-language models.