VT-3DAD: 3D anomaly detection via visual-text alignment

Arxiv CS.CV (computer vision)

Paper introduces VT-3DAD, a few-shot cross-category 3D anomaly detection method that aligns visual and text features in normal space. It requires only a few normal samples to detect anomalies in unknown point cloud categories.

Arxiv CS.CV (computer vision)·Zi WangKatsuya HottaYawen ZouKoichiro KamideYijin WeiChao ZhangJun Yu

7 days ago