ServiceNow/BigDocs-Bench

BigDocs-Bench是一个全面的基准套件，旨在评估将视觉输入转换为结构化输出的下游任务，如GUI2UserIntent（细粒度推理）和Image2Flow（结构化输出）。该数据集包含多个配置，每个配置都有详细的特征描述和数据集分割信息。数据集的使用需要下载外部图像资源，并提供了相应的脚本进行数据加载和匹配。数据集的许可为CC-BY-4.0，并使用了多个外部资源，如arXiv文档、SeeClick的HTML和元数据、InternVL-8B模型、Llama 3.1 405B模型和Graphviz的PNG可视化工具。

BigDocs-Bench is a comprehensive benchmark suite designed to evaluate downstream tasks that transform visual inputs into structured outputs, such as GUI2UserIntent (fine-grained reasoning) and Image2Flow (structured output). The dataset includes multiple configurations, each corresponding to different task types and featuring specific characteristics and data splits. The image data for the dataset is not included directly in the repository and must be downloaded separately by the user, with a provided script for alignment. The dataset generation involves multiple data sources and tools, such as arXiv documents, SeeClicks HTML and metadata, the InternVL-8B model, the Llama 3.1 model, and Graphvizs PNG visualizer.