3D / Point Cloud
Objaverse
800K+ annotated 3D models for training 3D and multimodal models.
100K–1M·CC-BY·Point Cloud (LAS/PLY)
GMT ADT trajectory cache Preprocessed per-sequence trajectory cache used by GMT (3DV 2026) when training and evaluating on the Aria Digital Twin (ADT) dataset. Each .pkl file holds the extracted 6-DOF object trajectories, multimodal conditioning features, and indexing metadata for a single ADT sequence, so the GMT training/eval loop does not have to re-parse the raw ADT data on every run. Paper arXiv 2603.17993 Code Pretrained… See the full description on the dataset page:
Source: Hugging Face Hub (huajian-zeng/gmt-adt-cache). Metadata imported from the dataset’s Hub tags.