3D-CoT Benchmark: Chain-of-Thought Datasets for 3D Point Cloud-Language Models
3
22 commits
1 linked in READMEs
updated Mar 14, 2025
The 3D-CoT Benchmark is a structured reasoning dataset designed explicitly to facilitate the systematic study of Chain-of-Thought (CoT)'s impact on 3D vision-language alignment. By extending existing 3D datasets with carefully structured reasoning annotations, this benchmark enables rigorous exploration of multimodal reasoning capabilities, significantly enhancing interpretability and effectiveness in vision-language tasks.
Each annotated 3D shape instance includes:
Annotations are presented in two explicit reasoning formats:
<think> markers.A rigorous two-stage annotation process ensures high-quality annotations:
Our benchmark leverages the high-quality Objaverse dataset exclusively for the CoT-CAP3D dataset, and the GAPartNet dataset provided within this repository.
gapartnet_pcs.zip).Each subset includes distinct test sets of approximately 100 samples each, explicitly designed to allow controlled comparisons between Tagged CoT and Unmarked CoT reasoning styles, ensuring robust performance evaluation across reasoning paradigms.
Please cite the following paper if you use this benchmark:
@article{chen2025integrating,
title={Integrating Chain-of-Thought for Multimodal Alignment: A Study on 3D Vision-Language Learning},
author={Chen, Yanjun and Sun, Yirong and Chen, Xinghao and Wang, Jian and Shen, Xiaoyu and Li, Wenjie and Zhang, Wei},
journal={arXiv preprint arXiv:2503.06232},
year={2025}
}
22 commits
3D-CoT Benchmark: Chain-of-Thought Datasets for 3D Point Cloud-Language Models
3
22 commits
1 linked in READMEs
updated Mar 14, 2025
The 3D-CoT Benchmark is a structured reasoning dataset designed explicitly to facilitate the systematic study of Chain-of-Thought (CoT)'s impact on 3D vision-language alignment. By extending existing 3D datasets with carefully structured reasoning annotations, this benchmark enables rigorous exploration of multimodal reasoning capabilities, significantly enhancing interpretability and effectiveness in vision-language tasks.
Each annotated 3D shape instance includes:
Annotations are presented in two explicit reasoning formats:
<think> markers.A rigorous two-stage annotation process ensures high-quality annotations:
Our benchmark leverages the high-quality Objaverse dataset exclusively for the CoT-CAP3D dataset, and the GAPartNet dataset provided within this repository.
gapartnet_pcs.zip).Each subset includes distinct test sets of approximately 100 samples each, explicitly designed to allow controlled comparisons between Tagged CoT and Unmarked CoT reasoning styles, ensuring robust performance evaluation across reasoning paradigms.
Please cite the following paper if you use this benchmark:
@article{chen2025integrating,
title={Integrating Chain-of-Thought for Multimodal Alignment: A Study on 3D Vision-Language Learning},
author={Chen, Yanjun and Sun, Yirong and Chen, Xinghao and Wang, Jian and Shen, Xiaoyu and Li, Wenjie and Zhang, Wei},
journal={arXiv preprint arXiv:2503.06232},
year={2025}
}
22 commits