---
dataset_info:
  features:
  - name: id
    dtype: string
  - name: sketch
    dtype: image
  - name: source_html
    dtype: string
  - name: source_screenshot
    dtype: image
  splits:
  - name: train
    num_bytes: 178040827.0
    num_examples: 731
  download_size: 136077581
  dataset_size: 178040827.0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

The Sketch2Code dataset consists of 731 human-drawn sketches paired with 484 real-world webpages from the [Design2Code dataset](https://huggingface.co/datasets/SALT-NLP/Design2Code), serving to benchmark Vision-Language Models (VLMs) on converting rudimentary sketches into web design prototypes.

See the dataset in raw files [here](https://huggingface.co/datasets/SALT-NLP/Sketch2Code).

Note that all images in these webpages are replaced by a blue placeholder image (rick.jpg).

Please refer to our [Project Page](https://salt-nlp.github.io/Sketch2Code-Project-Page/) for more detailed information.