Is Contrastive Distillation Enough for Learning Comprehensive 3D Representations?

ArXi:2412.08973v4 Announce Type: replace-cross Cross-modal contrastive distillation has recently been explored for learning effective 3D representations. However, existing methods focus primarily on modality-shared features, neglecting the modality-specific features during the pre-