Image-Guided Point Cloud Shape Completion Based on CTA Mechanism

With the rapid development of sensors, the technique of image-guided point cloud shape completion has achieved increasing attention. Compared to single-modal point cloud completion techniques, cross-modal feature fusion methods can fully utilize texture feature of images to enhance point cloud shape analysis. This paper employs the images of point clouds to boost the completion of incomplete point clouds. Based on the recent methods XMFnet, we propose a convolutional triplet attention (CTA) module to improve the feature extraction of images and enhance cross-modal feature fusion, thereby increasing point cloud completion performance. Experimental results demonstrates that our method outperforms all current multi-modal point cloud completion methods on the widely used ShapeNet-ViPC dataset.