pleaseconnectwifi / 3DQA

3D Question Answering

Paper (Early access) | Paper (ArXiv) | Project Page | Pre-trained Models | Fully Human-annotated Dataset

This is the official repository for our paper 3D Question Answering. Training code is coming soon.

Shuquan Ye¹, Dongdong Chen², Songfang Han³, Jing Liao^1
1City University of Hong Kong, ²Microsoft Cloud AI,³ University of California

Download

ScanQA dataset

The fully human-annotated ScanQA dataset with 9160 questions for train/validation and 902 questions for test can be downloaded in this link.

Checkpoints

• Pre-trained checkpoint on color-only questions

• Fine-tuned checkpoint on ScanQA

Acknowledgement

The data collection webpage is based on ScanRefer. We thank the authors for sharing their codebase!

BibTeX

@ARTICLE {ye2021tvcg3dqa,
author = {S. Ye and D. Chen and S. Han and J. Liao},
journal = {IEEE Transactions on Visualization & Computer Graphics},
title = {3D Question Answering},
issn = {1941-0506},
pages = {1-16},
doi = {10.1109/TVCG.2022.3225327}
}