Web2Code: A Large-scale Webpage-to-Code Dataset and Evaluation Framework for Multimodal LLMs

TL;DR

Web2Code introduces a large-scale webpage-to-code dataset and evaluation framework to improve multimodal large language models' ability to understand webpage images and generate accurate HTML code, addressing a current gap in multimodal AI capabilities.

Contribution

The paper presents a new dataset and evaluation framework specifically designed for webpage understanding and code generation tasks in multimodal LLMs, leveraging pretrained LLMs for dataset enhancement.

Findings

The dataset improves MLLMs' webpage understanding and code generation.

Models show limited performance on webpage-to-code tasks without specialized training.

The dataset benefits general visual domain tasks as well.

Abstract

Multimodal large language models (MLLMs) have shown impressive success across modalities such as image, video, and audio in a variety of understanding and generation tasks. However, current MLLMs are surprisingly poor at understanding webpage screenshots and generating their corresponding HTML code. To address this problem, we propose $\texttt{Web2Code}$, a benchmark consisting of a new large-scale webpage-to-code dataset for instruction tuning and an evaluation framework for the webpage understanding and HTML code translation abilities of MLLMs. For dataset construction, we leverage pretrained LLMs to enhance existing webpage-to-code datasets as well as generate a diverse pool of new webpages rendered into images. Specifically, the inputs are webpage images and instructions, while the responses are the webpage's HTML code. We further include diverse natural language QA pairs about the webpage content in the responses to enable a more comprehensive understanding of the web content. To evaluate model performance in these tasks, we develop an evaluation framework for testing MLLMs' abilities in webpage understanding and web-to-code generation. Extensive experiments show that our proposed dataset is beneficial not only to our proposed tasks but also in the general visual domain. We hope our work will contribute to the development of general MLLMs suitable for web-based content generation and task automation. Our data and code are available at https://github.com/MBZUAI-LLM/web2code.