MLL-Lab/MindCube

---
language:
- en
license: mit
size_categories:
- 10K<n<100K
task_categories:
- question-answering
- image-text-to-text
pretty_name: MindCube
tags:
- spatial-reasoning
- visual-reasoning
- multimodal
- vision-language-model
- benchmark
---

# MindCube: Spatial Mental Modeling from Limited Views

MindCube is a novel benchmark designed to evaluate how well Vision Language Models (VLMs) can form robust spatial mental models from limited visual views. It comprises 21,154 questions across 3,268 images, assessing capabilities such as cognitive mapping (representing positions), perspective-taking (orientations), and mental simulation (dynamics for "what-if" movements). The dataset aims to expose critical gaps in existing VLMs' spatial understanding and facilitate research into improving their performance through structured spatial representations.

* **Paper:** [Spatial Mental Modeling from Limited Views](https://huggingface.co/papers/2506.21458)
* **Project Page:** [https://mll-lab-nu.github.io/mind-cube](https://mll-lab-nu.github.io/mind-cube)
* **GitHub Repository:** [https://github.com/mll-lab-nu/MindCube](https://github.com/mll-lab-nu/MindCube)

We release our MindCube dataset at this repo in the zip file `data.zip`. To use it, please follow the instructions in our [github repo](https://github.com/mll-lab-nu/MindCube).

## Sample Usage

To download the MindCube dataset:

```bash
bash scripts/bash_scripts/download_data.bash
```

Refer to the [GitHub repository](https://github.com/mll-lab-nu/MindCube) for further instructions on data generation, model inference, and evaluation.

## The File Structure of `data.zip`
```
data/
├── other_all_image/
│ ├── around/
│ │ └── <image_group_id>/
│ │ ├── image1.png
│ │ ├── image2.jpg
│ │ └── ...
│ ├── among/
│ │ └── <image_group_id>/
│ │ ├── image1.png
│ │ ├── image2.jpg
│ │ └── ...
│ ├── rotation/
│ │ └── <image_group_id>/
│ │ ├── image1.png
│ │ ├── image2.jpg
│ │ └── ...
│ └── translation/
│ └── <image_group_id>/
│ ├── image1.png
│ ├── image2.jpg
│ └── ...
└── raw/
├── MindCube.jsonl
├── MindCube_train.jsonl
└── MindCube_tinybench.jsonl

```

---
语言：
- 英语
许可证：MIT许可证
规模类别：
- 1万 < 样本量 < 10万
任务类别：
- 问答
- 图像-文本转文本
美观名称：MindCube
标签：
- 空间推理
- 视觉推理
- 多模态
- 视觉语言模型
- 基准测试
---

# MindCube：基于有限视角的空间心理建模
MindCube是一款全新的基准测试集，用于评估视觉语言模型（Vision Language Models, VLMs）从有限视觉视角构建稳健空间心理模型的能力。该数据集包含3268张图像对应的21154个问题，用于测评认知映射（表征物体位置）、视角转换（空间朝向）以及心理模拟（“假设”场景运动的动态推演）等多项核心能力。本数据集旨在揭示现有视觉语言模型在空间理解方面的关键短板，并推动基于结构化空间表征的模型性能优化相关研究。

* **论文**：[《基于有限视角的空间心理建模》](https://huggingface.co/papers/2506.21458)
* **项目主页**：[https://mll-lab-nu.github.io/mind-cube](https://mll-lab-nu.github.io/mind-cube)
* **GitHub代码仓库**：[https://github.com/mll-lab-nu/MindCube](https://github.com/mll-lab-nu/MindCube)

我们已将MindCube数据集打包为`data.zip`文件发布于本仓库。使用前请遵循我们[GitHub代码仓库](https://github.com/mll-lab-nu/MindCube)中的说明进行操作。

## 示例用法
### 下载MindCube数据集
使用以下命令下载数据集：
bash
bash scripts/bash_scripts/download_data.bash

有关数据生成、模型推理与模型评估的更多操作说明，请参阅[GitHub代码仓库](https://github.com/mll-lab-nu/MindCube)。

## `data.zip` 的文件结构

data/
├── other_all_image/
│ ├── around/
│ │ └── <image_group_id>/
│ │ ├── image1.png
│ │ ├── image2.jpg
│ │ └── ...
│ ├── among/
│ │ └── <image_group_id>/
│ │ ├── image1.png
│ │ ├── image2.jpg
│ │ └── ...
│ ├── rotation/
│ │ └── <image_group_id>/
│ │ ├── image1.png
│ │ ├── image2.jpg
│ │ └── ...
│ └── translation/
│ └── <image_group_id>/
│ ├── image1.png
│ ├── image2.jpg
│ └── ...
└── raw/
├── MindCube.jsonl
├── MindCube_train.jsonl
└── MindCube_tinybench.jsonl