Ground3D Dataset

Ground3D Dataset A large-scale 3D vision-language question-answering dataset for point-grounded, metric-aware 3D scene understanding. Built on ScanNet and ScanNet++ with dense object and part annotations, the dataset spans eight downstream reasoning tasks at both object and part granularity, plus multi-turn dialogue that composes them. Answers are both: point-grounded: explicitly tied to the referred 3D region via label markup, and metric: physical quantities (size… See the full description on the dataset page: Dataset.

Source: Hugging Face Hub (amolharsh/Ground3D_Dataset). Metadata imported from the dataset’s Hub tags.