| --- |
| dataset_info: |
| features: |
| - name: svg |
| dtype: string |
| - name: html |
| dtype: string |
| splits: |
| - name: train |
| num_bytes: 18129 |
| num_examples: 4 |
| download_size: 14493 |
| dataset_size: 18129 |
| configs: |
| - config_name: default |
| data_files: |
| - split: train |
| path: data/train-* |
| license: apache-2.0 |
| --- |
| # Reverse Browser Small Validation Dataset |
|
|
| This dataset contains a very small number of paired HTML/CSS markup and vector-based SVG representations simple websites. Rendered at mobile resolution. It created by custom browser rendering pipeline. The dataset is designed for quick evaluation of models during training that translate visual web designs to markup. |
|
|
| This dataset is not meant to train capable models, but to be used create proof-of-concept and run performance tests when evaluating model architectures. |
|
|
| For larger datasets created from public web pages one of the following: |
|
|
| - https://huggingface.co/datasets/tcz/rb-large |
| - https://huggingface.co/datasets/tcz/rb-large-reduced |
| - https://huggingface.co/datasets/tcz/rb-large-chopped |
|
|
| ## Contents |
|
|
| Each entry in the dataset includes: |
|
|
| - **HTML**: A cleaned and compressed version of the page’s markup, stripped of `<script>`, `<iframe>`, `<video>`, and other non-visual elements, only retaining `<body>` outerHTML. CSS has been inlined. Images are not saved. |
| - `html`: The preprocessed HTML/CSS source |
| - **SVG**: SVG representation of the page for the following viewport sizes: 393×852 |
|
|
| https://arxiv.org/abs/2509.05394 |
