OpenMOSS-Team/GameQA-140K

GameQA是一个大规模、多样化和具有挑战性的多模态推理数据集，旨在增强视觉语言模型（VLMs）的通用推理能力。该数据集通过创新的Code2Logic框架生成，利用游戏代码合成高质量的视觉语言链式思维（CoT）数据。数据集解决了多模态推理数据稀缺的问题，这对推进VLMs中的复杂多步推理至关重要。每个样本包括视觉游戏状态、针对性问题、原始分析、增强的逐步推理（`refinement`）和从游戏代码逻辑结构中得出的最终答案。数据集包含约140,000个问答对（126,760个训练样本，15,047个测试样本），涵盖30个独特游戏和158个不同任务，涉及多种认知技能。游戏类别包括3D空间感知与理解、模式识别与匹配、多步推理和战略规划。数据集格式为视觉问答（VQA），包括游戏状态图像、针对性问题、逐步推理和最终答案。问题类型包括多选题（通常7-8个选项）和填空题（如数字、坐标）。数据集对当前最先进的VLMs具有挑战性（测试集准确率<50%）。Code2Logic框架在初始设置后能以最小成本实现大规模生成。难度等级分为图像复杂度（简单、中等、困难）和任务复杂度（简单、中等、困难）。

GameQA is a large-scale, diverse, and challenging multimodal reasoning dataset designed to enhance the general reasoning capabilities of Vision Language Models (VLMs). Generated using the innovative Code2Logic framework, it leverages game code to synthesize high-quality visual-language Chain-of-Thought (CoT) data. The dataset addresses the scarcity of multimodal reasoning data, critical for advancing complex multi-step reasoning in VLMs. Each sample includes visual game state, targeted question, original analysis, augmented step-by-step reasoning (`refinement`) and final answer, derived from the logical structures inherent in game code. The dataset contains ~140,000 question-answer pairs (126,760 training, 15,047 testing), covering 30 unique games and 158 distinct tasks involving various cognitive skills. Game categories include 3D Spatial Perception and Understanding, Pattern Recognition and Matching, Multi-step Reasoning, and Strategic Planning. The format is Visual Question Answering (VQA), including game state image, targeted question, step-by-step reasoning, and final answer. Question types include multiple-choice (typically 7-8 options) and fill-in-the-blank (e.g., numbers, coordinates). The dataset is challenging for SOTA VLMs (<50% accuracy on test set). Code2Logic enables massive-scale generation with minimal cost after initial setup. Difficulty levels are divided into Plot Level (Image Complexity): Easy, Medium, Hard and QA Level (Task Complexity): Easy, Medium, Hard.