Mohamed-Abbas/ChartQA-difficulty-tier

ChartQA难度等级数据集（基于规则）在原始的ChartQA基准数据集基础上，增加了一个难度等级标签，分为三个等级：简单（easy）、中等（medium）和困难（hard）。这些难度等级是通过一个专为ChartQA设计定制的规则基础分类器来确定的，考虑了视觉定位、推理操作、数值复杂性、极值线索等因素。此外，数据集遵循特定的安全规则，确保某些模式不会错误地分类为简单等级。数据集的原始字段保持不变，仅增加了一个difficulty_tier列来表示难度等级。The ChartQA difficulty tiers dataset (rule-based) adds a difficulty level label to the original ChartQA benchmark dataset, divided into three levels: easy, medium, and hard. These difficulty levels are determined by a rule-based classifier tailored specifically for ChartQA, considering factors such as visual grounding, reasoning operations, numeric complexity, and extremum cues. In addition, the dataset follows a safety rule to prevent certain patterns from being misclassified as easy. The original fields of the dataset remain unchanged, with only the addition of a difficulty_tier column to indicate the level of difficulty.