Towards Foundation Models for 3D Vision: How Close Are We?
This work addresses the problem of assessing 3D reasoning capabilities for researchers in computer vision, providing a benchmark to guide future foundation model development, though it is incremental as it focuses on evaluation rather than proposing a new model.
The paper tackles the challenge of building foundation models for 3D vision by constructing a new benchmark, UniQA-3D, to evaluate current models and compare them to human performance. Results show that vision-language models perform poorly, specialized models are accurate but not robust, and humans remain the most reliable, with neural networks aligning more closely with human mechanisms than classical methods.
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 .