PVCLN: Point-View Complementary Learning Network for 3D Shape Recognition

As an important topic in computer vision and multimedia analysis, 3D shape recognition has attracted much research attention recent years. For point cloud data multiview data, various approaches have been proposed with remarkable performance. However, few works simultaneously employ the to represent shapes, which is complementary beneficial our consideration. Moreover, existing multimodal mainly focus on fusion strategy or exploring relation between them. intra-modality characteristic information inter-modality are ignored these methods. In this paper, we tackle above limitations by introducing a novel Point-View Complementary Learning Network (PVCLN) explore potential of both for recognition. Inspired successful application graph neural networks capturing relations features, introduce strategy. Concretely, first separately extract visual feature from structural data. We then project into same space learn two modalities modeling affinities. The each modality also preserved considering affinities compensate lacking enhance learning process. Finally, updated features further combined achieve unified representation shape. conduct extensive experiments validate superiority overall network effectiveness component. method evaluated ModelNet40 dataset experimental results demonstrate that framework achieves competitive performance task.