MeshCLIP: Efficient cross-modal information processing for 3D mesh data in zero/few-shot learning

Yupeng Song, Naifu Liang, Qing Guo, Jicheng Dai, Junwei Bai, Fazhi He

Abstract: Highlights•The first 3D mesh processing approach by leveraging across-modal information in a zero/few-shot manner.•A novel cross-modal learning strategy combining the modality vision data and 3D graphics data.•A novel self-attention adapter for few-shot learning to explore key information and enhance discrimination.•The proposed MeshCLIP can achieve state-of-the-art results on multiple challenging 3D mesh datasets.

External IDs:dblp:journals/ipm/SongLGDBH23