Saved in:
Bibliographic Details
Main Authors: Zuo, Yiming, Kayan, Karhan, Wang, Maggie, Jeon, Kevin, Deng, Jia, Griffiths, Thomas L.
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2410.10799
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866929620387889152
author Zuo, Yiming
Kayan, Karhan
Wang, Maggie
Jeon, Kevin
Deng, Jia
Griffiths, Thomas L.
author_facet Zuo, Yiming
Kayan, Karhan
Wang, Maggie
Jeon, Kevin
Deng, Jia
Griffiths, Thomas L.
contents Building a foundation model for 3D vision is a complex challenge that remains unsolved. Towards that goal, it is important to understand the 3D reasoning capabilities of current models as well as identify the gaps between these models and humans. Therefore, we construct a new 3D visual understanding benchmark named UniQA-3D. UniQA-3D covers fundamental 3D vision tasks in the Visual Question Answering (VQA) format. We evaluate state-of-the-art Vision-Language Models (VLMs), specialized models, and human subjects on it. Our results show that VLMs generally perform poorly, while the specialized models are accurate but not robust, failing under geometric perturbations. In contrast, human vision continues to be the most reliable 3D visual system. We further demonstrate that neural networks align more closely with human 3D vision mechanisms compared to classical computer vision methods, and Transformer-based networks such as ViT align more closely with human 3D vision mechanisms than CNNs. We hope our study will benefit the future development of foundation models for 3D vision. Code is available at https://github.com/princeton-vl/UniQA-3D .
format Preprint
id arxiv_https___arxiv_org_abs_2410_10799
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Towards Foundation Models for 3D Vision: How Close Are We?
Zuo, Yiming
Kayan, Karhan
Wang, Maggie
Jeon, Kevin
Deng, Jia
Griffiths, Thomas L.
Computer Vision and Pattern Recognition
Building a foundation model for 3D vision is a complex challenge that remains unsolved. Towards that goal, it is important to understand the 3D reasoning capabilities of current models as well as identify the gaps between these models and humans. Therefore, we construct a new 3D visual understanding benchmark named UniQA-3D. UniQA-3D covers fundamental 3D vision tasks in the Visual Question Answering (VQA) format. We evaluate state-of-the-art Vision-Language Models (VLMs), specialized models, and human subjects on it. Our results show that VLMs generally perform poorly, while the specialized models are accurate but not robust, failing under geometric perturbations. In contrast, human vision continues to be the most reliable 3D visual system. We further demonstrate that neural networks align more closely with human 3D vision mechanisms compared to classical computer vision methods, and Transformer-based networks such as ViT align more closely with human 3D vision mechanisms than CNNs. We hope our study will benefit the future development of foundation models for 3D vision. Code is available at https://github.com/princeton-vl/UniQA-3D .
title Towards Foundation Models for 3D Vision: How Close Are We?
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2410.10799