3D / Point Cloud
Objaverse
800K+ annotated 3D models for training 3D and multimodal models.
100K–1M·CC-BY·Point Cloud (LAS/PLY)
Taskonomy Subset (3DVLM) A small, fast-to-download slice of the Taskonomy real indoor-scan dataset, converted to a uniform posed-RGB-D format for quick model test-runs. This is a subset: 25 buildings (seeded pick, seed 0, from the Omnidata tiny split) × 100 frames each = 2,500 frames. These are real photographs of scanned buildings with sensor-grade ground-truth depth and exact camera poses — no reconstruction or pseudo-labelling involved. This subset is part of a family of… See the full description on the dataset page: subset.
Source: Hugging Face Hub (helioom/3dvlm-taskonomy_subset). Metadata imported from the dataset’s Hub tags.