aleversn/WebRenderBench

WebRenderBench 是一个大型基准数据集，旨在通过在真实网页上进行评估来推进多模态大型语言模型（MLLMs）的 WebUI-to-Code 研究。该数据集具有高度多样性和复杂性，包含来自各个行业和设计风格的 45,100 个真实网页。它引入了新颖的评估指标，重点关注布局和样式一致性，并辅以 ALISA 强化学习框架来优化生成质量。数据集通过一个系统的过程构建，确保高质量和多样性，并被均匀地分为训练集和测试集。README 还提供了评估框架，包括 RDA、GDA 和 SDA 等指标，并详细介绍了基准工作流程，包括数据准备、模型推理、评估和可选的 ALISA 训练。该数据集在 Apache License 2.0 许可下发布，仅供研究使用。

WebRenderBench is a large-scale benchmark dataset designed to advance WebUI-to-Code research for multimodal large language models (MLLMs) through evaluation on real-world webpages. The dataset features high diversity and complexity, with 45,100 real webpages collected from various industries and design styles. It introduces novel evaluation metrics focusing on layout and style consistency, and is accompanied by the ALISA reinforcement learning framework for optimizing generation quality. The dataset is constructed through a systematic process ensuring high quality and diversity, and is split evenly into training and test sets. The README also provides an evaluation framework with metrics like RDA, GDA, and SDA, and details the benchmark workflow, including data preparation, model inference, evaluation, and optional ALISA training. The dataset is released under the Apache License 2.0 for research purposes only.