3D / Point Cloud
Objaverse
800K+ annotated 3D models for training 3D and multimodal models.
100K–1M·CC-BY·Point Cloud (LAS/PLY)
Berkeley GRAFIQ — Graphics Inquiry Benchmark Synthetic VQA benchmark for evaluating vision-language models on 3D rendering properties (depth, surface roughness, normal orientation, light direction). The on-disk dataset is the contents of examples/showcase/ from the grafiq datagen repository, packed into a single tarball to fit free-tier HF upload limits. Layout File What it is showcase.tar Full showcase folder, including 1.5k render dirs (~3.5 GB).… See the full description on the dataset page:
Source: Hugging Face Hub (tsunghanwu/berkeley-grafiq-dataset). Metadata imported from the dataset’s Hub tags.