3DZip: Spatial-Aware Feature Diversity-Guided Token Compression for 3D Question Answering | Changwoo Baek, Kyeongbo Kong | ResearchPod