Gallardot/AtomBlock-WebUI

AtomBlock-WebUI是一个合成的网页用户界面数据集，具有像素级精确的原子元素和结构块，通过LLM增强的HTML渲染和无头浏览器截图捕获生成。数据集包含约9,700张全页网页截图，带有14个UI元素类别的YOLO格式边界框标注，包括基本组件（按钮、输入框）和语义块级地标（导航栏、侧边栏、页脚）。与依赖人工标注或启发式DOM解析方法的数据集不同，AtomBlock-WebUI中的边界框是通过Playwright直接从渲染的DOM中提取的，确保与视觉输出严格几何对齐。真实世界的图像被注入到合成的HTML中，以弥合合成布局与真实网页环境之间的视觉分布差距。

AtomBlock-WebUI is a synthetic web UI dataset featuring pixel-perfect atomic elements and structural blocks, generated via LLM-augmented HTML rendering and headless browser screenshot capture. The dataset contains ~9,700 full-page web screenshots with YOLO-format bounding box annotations for 14 UI element categories, including both primitive components (buttons, inputs) and semantic block-level landmarks (navigation, sidebar, footer). Unlike datasets reliant on human annotation or heuristic DOM-parsing methods, the bounding boxes in AtomBlock-WebUI are directly extracted from the rendered DOM via Playwright, ensuring strict geometric alignment with the visual output. Real-world images are injected into the synthetic HTML to bridge the visual distribution gap between synthetic layouts and real web environments.