🏠💬 3D-GRAND

A Million-Scale Dataset for 3D-LLMs with Better Grounding and Less Hallucination

Jianing (Jed) Yang*,1, Xuweiyi Chen*,1, Nikhil Madaan, Madhavan Iyengar1,
Shengyi Qian1, David Fouhey1,2, Joyce Y. Chai1,
1University of Michigan 2New York University
*Denotes Equal Contribution
Paper 🤗 3D-GRAND Data 3D-POPE Data 3D-POPE Leaderboard 🤗 Demo Code
Overview

Model trained with 3D-GRAND enables user to densely-ground noun phrases in a sentence.

Abstract

The integration of language and 3D perception is crucial for developing embodied agents and robots that comprehend and interact with the physical world. While large language models (LLMs) have demonstrated impressive language understanding and generation capabilities, their adaptation to 3D environments (3D-LLMs) remains in its early stages. A primary challenge is the absence of large-scale datasets that provide dense grounding between language and 3D scenes. In this paper, we introduce 3D-GRAND, a pioneering large-scale dataset comprising 40,087 household scenes paired with 6.2 million densely-grounded scene-language instructions. Our results show that instruction tuning with 3D-GRAND significantly reduces hallucinations and enhances the grounding capabilities of 3D-LLMs compared to models trained without dense grounding. As part of our contributions, we propose a comprehensive benchmark 3D-POPE to systematically evaluate hallucination in 3D-LLMs, enabling fair comparisons among future models. Our experiments underscore a scaling effect between dataset size and 3D-LLM performance, emphasizing the critical role of large-scale 3D-text datasets in advancing embodied AI research. Through 3D-GRAND and 3D-POPE, we aim to equip the embodied AI community with essential resources and insights, setting the stage for more reliable and better-grounded 3D-LLMs.

Dataset Overview


Generalist outputs

Descriptive alt text

Data Curation Pipeline


Our Framework


Data Visualization

Please select a task.

3D-POPE Leaderboard: Evaluating Hallucination for 3D-LLMs

3D-POPE Model Precision Recall F1 Score Accuracy Yes (%)
Random Random Baseline 50.00 50.00 50.00 50.00 50.00
3D-LLM 50.03 99.88 66.67 50.07 99.81
3D-VisTA 50.12 53.58 51.79 49.66 53.95
LEO 51.95 77.65 62.25 52.91 74.73
Ours zero-shot (Grounding) 93.34 84.25 88.56 89.12 45.13
Popular Random Baseline 50.00 50.00 50.00 50.00 50.00
3D-LLM 49.97 99.88 66.61 49.94 99.94
3D-VisTA 47.40 51.88 49.54 49.49 52.30
LEO 48.30 77.65 59.55 47.27 80.38
Ours zero-shot (Grounding) 73.05 84.28 78.26 76.59 57.69
Adversarial Random Baseline 50.00 50.00 50.00 50.00 50.00
3D-LLM 49.97 99.88 66.61 49.94 99.94
3D-VisTA 48.28 54.39 51.15 51.14 52.99
LEO 48.47 77.98 59.78 47.52 80.45
Ours zero-shot (Grounding) 69.86 84.21 76.37 73.95 60.26

Data Scaling

Citation

@misc{yang2024_3D_GRAND,
  title={3D-GRAND: A Million-Scale Dataset for 3D-LLMs with Better Grounding and Less Hallucination},
  author={Jianing Yang and Xuweiyi Chen and Nikhil Madaan and Madhavan Iyengar and Shengyi Qian and David F. Fouhey and Joyce Chai},
  year={2024},
  eprint={2406.05132},
  archivePrefix={arXiv},
  primaryClass={cs.CV}
}