Skip to content

Dataset · amolharsh

3D scene question-answering with point grounding and metrics

amolharsh/Ground3D_Dataset

Large English corpus of 3D scenes and dialogues based on ScanNet for training models in spatial and object size understanding.

What's inside

It contains 1 million to 10 million English examples combining 3D data and text. Built on ScanNet and ScanNet++ with dense annotations of objects and parts, it covers eight reasoning tasks and multi-turn dialogues. Answers are grounded in specific 3D regions via specialized markup and include physical quantities (e.g., dimensions).

What you can do with it

  • Training multimodal 3D models (3D VLMs) to answer questions on spatial scenes
  • Training systems to estimate real physical dimensions of objects and their parts in 3D
  • Fine-tuning (SFT) assistants for spatial navigation and reasoning in robotics
Tasks
Visual question answering, Question answering
Languages
English
Size
1M–10M rows
Downloads this month
60.5K
Likes
1

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("amolharsh/Ground3D_Dataset")