3D model autotagging based on semi-supervised weighted metric learning

The remarkable growth of 3D models on the Internet has posed a great challenge for model search. Many 3D search engines are based on tag matching, which is usually more accurate in identifying relevant models by alleviating the challenge arising from the semantic gap. However, the performance of tag matching is highly dependent on the availability and quality of 3D model tags. Recent studies have shown that tags that are specific to the visual content of 3D models are often noisy and unreliable in real-world environment, leading to a limited performance of autotagging. To address this challenge, we propose a 3D model autotagging method based on semi-supervised weighted metric learning. Extensive experiments show that the proposed method is significantly more effective than the state-of-the-art.