task_categories: | |
- image-text-to-text | |
language: | |
- zh | |
tags: | |
- benchmark | |
- vlm | |
- graphic-reasoning | |
- intelligence-test | |
This repository contains **ReasonBench**, a novel benchmark designed to evaluate Visual Language Models (VLMs) on complex graphic reasoning tasks. It comprises 1,613 questions derived from real-world intelligence tests, covering reasoning dimensions related to location, attribute, quantity, and multi-element tasks. ReasonBench provides a comprehensive framework for assessing VLMs' spatial, relational, and abstract reasoning capabilities. | |
Paper: [Oedipus and the Sphinx: Benchmarking and Improving Visual Language Models for Complex Graphic Reasoning](https://huggingface.co/papers/2508.00323) | |
Code and Data: https://github.com/ReasonBench/ReasonBench | |
--- | |
## ReasonBench:用于复杂图形推理的视觉语言模型评估基准 | |
ReasonBench 旨在评估视觉语言模型(VLMs)在复杂图形推理中的表现。数据集包含从真实智力测试中收集的 1,613 个问题,覆盖 11 个核心认知推理维度和 29 种任务类型,为评估 VLMs 的空间、关系和抽象推理能力提供综合框架。 | |
数据集类型:视觉语言推理、图形推理、基准测试 | |
|11个核心认知维度 | 29种任务类型 |数量 | | |
|--------------|----------------|-----| | |
| 位置规律 | 平移 |94| | |
| | 旋转 |56| | |
| | 组合 |30| | |
| 样式规律 | 穿越 |54| | |
| | 加减法 |67| | |
| | 黑白运算 |63| | |
| 属性规律 | 对称 |109| | |
| | 开闭状态 |19| | |
| | 组合 |6| | |
| 数量规律 | 线 |173| | |
| | 面 |137| | |
| | 点 |66| | |
| | 元素 |94| | |
| | 组合 |50| | |
| 空间规律 | 立方体 |109| | |
| | 3d |46| | |
| | 多面体 |17| | |
| | 三视图 |40| | |
| | 剖视图 |35| | |
| | 空间数量变换 |10| | |
| 特殊规律 | 2d 组合 |31| | |
| | 图形关系 |40| | |
| 字母数字 | 字母数字 |27| | |
| 黑白块 | 黑白块 |32| | |
| 其他规律 | 综合 |34| | |
| 门萨 |任务1 |35| | |
| |任务2 |39| | |
| 瑞文 |任务1 |40| | |
| |任务2 |60| | |
本基准考虑了输入图片的整体式和分隔式两种输入格式对模型的影响。 | |
|格式|介绍| | |
|-------------|-------| | |
|集成格式(Integrated format)|问题与选项呈现在单个图形中,便于模型整体处理| | |
|分离格式(Separated format)|问题与选项拆分为多个图形,依次输入模型,测试分步推理能力| | |
为未来研究者方便复现以及后续研究,我们将所有格式的图片url全部公开,分别对应题目,4-8个选项,题目➕选项。 | |
同时,我们公开了人类基准的准确率。 |