| --- |
| license: mit |
| library_name: pytorch |
| tags: |
| - pointllm |
| - point-cloud |
| - 3d |
| - multimodal |
| - chain-of-thought |
| - reasoning |
| base_model: RunsenXu/PointLLM_7B_v1.2 |
| datasets: |
| - QileXu/PoCoTI-55K |
| language: |
| - en |
| --- |
| |
| # PointLLM-R: Enhancing 3D Point Cloud Reasoning via Chain-of-Thought |
|
|
| **Chaoqi Chen**¹\*, **Qile Xu**¹\*, **Wenjun Zhou**¹, **Hui Huang**¹† |
|
|
| ¹Shenzhen University \*Equal contribution †Corresponding author |
| |
| [Paper](https://arxiv.org/abs/2605.22013) | [Project Page](https://vcc.tech/research/2026/PointLLM-R) | [Code](https://github.com/Xqle/PointLLM-R) | [Collection](https://huggingface.co/collections/QileXu/pointllm-r) |
| |
| --- |
| |
| Official model weights for the paper **PointLLM-R: Enhancing 3D Point Cloud Reasoning via Chain-of-Thought** (SIGGRAPH 2026). |
| |
| PointLLM-R-7B is a 3D multimodal LLM fine-tuned from [PointLLM](https://github.com/OpenRobotLab/PointLLM) on the [PoCoTI-55K](https://huggingface.co/datasets/QileXu/PoCoTI-55K) dataset, which augments point-cloud QA pairs with structured 5-step chain-of-thought reasoning. |
| |
| ## Links |
| |
| - 📄 Paper: https://arxiv.org/abs/2605.22013 |
| - 🌐 Project page: https://vcc.tech/research/2026/PointLLM-R |
| - 💻 Code: https://github.com/Xqle/PointLLM-R |
| - 📦 Collection: https://huggingface.co/collections/QileXu/pointllm-r |
| - 🗂️ Training data: [QileXu/PoCoTI-55K](https://huggingface.co/datasets/QileXu/PoCoTI-55K) |
| - 📊 Eval GT: [QileXu/OmniObject3D_brief_description_val_GT](https://huggingface.co/datasets/QileXu/OmniObject3D_brief_description_val_GT) |
| |
| ## Quick Start |
| |
| See the [GitHub repository](https://github.com/Xqle/PointLLM-R) for installation, inference, and evaluation instructions. |
| |
| ## Citation |
| |
| ```bibtex |
| @inproceedings{chen2026pointllmr, |
| title = {PointLLM-R: Enhancing 3D Point Cloud Reasoning via Chain-of-Thought}, |
| author = {Chen, Chaoqi and Xu, Qile and Zhou, Wenjun and Huang, Hui}, |
| booktitle = {ACM SIGGRAPH}, |
| year = {2026}, |
| pages = {} |
| } |
| ``` |
| |
| ## License |
| |
| MIT. The base model and Objaverse-derived data retain their original licenses. |
| |