--- pretty_name: Pattern2Code license: odc-by task_categories: - image-to-text - text-generation language: - en tags: - screenshot-to-code - multimodal - html - css - visual-grounding - benchmark size_categories: - n<1K configs: - config_name: default default: true data_files: - split: train path: data/train-*.parquet dataset_info: features: - name: standard_image dtype: image - name: noise_image dtype: image - name: example_id dtype: string - name: source_id dtype: string - name: pattern_family dtype: string - name: pattern_name dtype: string - name: attribute_name dtype: string - name: attribute_token dtype: string - name: position dtype: string - name: target_value dtype: string - name: target_value_from_filename dtype: string - name: bias_value dtype: string - name: pattern_description dtype: string - name: element_description dtype: string - name: code_snippet_line dtype: string - name: code_snippet_context dtype: string - name: html_path dtype: string - name: standard_image_path dtype: string - name: noise_image_path dtype: string - name: modified_index dtype: int32 splits: - name: train num_examples: 720 --- # Pattern2Code **Benchmark for the ASE 2026 paper *"Pattern Over Pixels: Measuring Pattern Completion Bias in Multimodal Code Generation."*** Project page: https://pattern2code.github.io/ · Artifact (Zenodo): https://doi.org/10.5281/zenodo.19341952 Pattern2Code measures *visual pattern-completion bias* in screenshot-to-code generation: when a webpage contains a repeated UI pattern and exactly **one** element deviates from it, does a multimodal LLM write the code that matches the **pixels**, or the code that matches the **pattern**? Each example pairs a rendered screenshot with the element's HTML snippet in which the deviating CSS value is masked as `__`. The model must recover the masked value from the image. ## Key results (5 frontier MLLMs) | Family | Mean accuracy | Mean bias rate | |---|---|---| | Card patterns (`width`) | 21.17% | **69.78%** | | Text patterns (`font-size`) | 7.89% | **80.22%** | *Bias rate* = fraction of predictions equal to the repeated baseline `100%`. ## Contents - **720** perturbed instances = 30 webpages × 2 pattern families × 3 positions × 4 magnitudes - **360** structural-card (`width`) + **360** text-style (`font-size`) examples - **1,440** screenshots — every instance rendered under 2 conditions (`standard_image`, `noise_image`) - Perturbation magnitudes: `80%`, `90%`, `110%`, `120%`; unperturbed baseline is `100%` ## Layout ```text data/train-*.parquet # the dataset: one row per instance, screenshots embedded data/examples.jsonl # the same rows as raw JSON (image columns are paths only) metadata/dataset_info.json # summary statistics metadata/validation_report.json html/{cards,texts}/ # perturbed HTML images/{cards,texts}/{standard,noise}/ # the screenshots as loose PNG files ``` The Parquet files carry the screenshots inline, so both image columns render directly in the dataset viewer and decode to `PIL.Image` with no extra setup. The loose PNGs under `images/` are the same files, kept for anyone who wants to work with them outside the `datasets` library. ## Row schema | Field | Description | |---|---| | `standard_image` | **Image** — the rendered screenshot | | `noise_image` | **Image** — the same page with eight opaque rectangles overlaid | | `example_id` | Unique instance id | | `source_id` | Seed webpage id (Design2Code) | | `pattern_family` | `structural_card` or `text_style` | | `pattern_name`, `pattern_description`, `element_description` | The repeated pattern being probed | | `attribute_name` | Masked CSS property (`width` or `font_size`) | | `attribute_token` | The mask token (`__`) | | `position` | Which element was perturbed: `first`, `mid`, `last` | | `modified_index` | Index of the perturbed element within the pattern | | `target_value` | **Ground truth** — the perturbed value (e.g. `120%`) | | `bias_value` | The pattern-consistent baseline (always `100%`) | | `code_snippet_line` | The masked element's HTML | | `code_snippet_context` | Masked element plus its sibling elements (model input) | | `html_path` | Relative path to the full perturbed HTML | | `standard_image_path` / `noise_image_path` | Relative paths to the two screenshot files | ## Loading ```python from datasets import load_dataset ds = load_dataset("knguyennguyen/pattern2code", split="train") ex = ds[0] ex["standard_image"] # PIL.Image — already decoded, no cast_column needed ex["noise_image"] # PIL.Image print(ex["attribute_name"], ex["target_value"], "vs baseline", ex["bias_value"]) print(ex["code_snippet_context"]) ``` ## Evaluation protocol Prompt the model with `standard_image` (or `noise_image`) plus `code_snippet_context`: ``` You are doing a visual code fill-in-the-blank task. Given the webpage screenshot and HTML snippet below, fill the blank token __ with ONLY the missing CSS/HTML value. The goal is to predict the masked value to recreate the {pattern} in the webpage design. Return the final answer in JSON format: {"answer":""}. The blank must be a percentage value divisible by 10. ``` Score each prediction into three mutually exclusive outcomes: - **Correct** — equals `target_value` - **Bias-aligned** — equals `bias_value` (`100%`) - **Other error** — neither ## Citation ```bibtex @inproceedings{nguyen2026pattern, title = {Pattern Over Pixels: Measuring Pattern Completion Bias in Multimodal Code Generation}, author = {Nguyen, Khai-Nguyen and Chaparro, Oscar and Mastropaolo, Antonio}, booktitle = {Proceedings of the 41st IEEE/ACM International Conference on Automated Software Engineering (ASE)}, year = {2026} } ``` ## License & provenance The seed webpages come from the [Design2Code](https://github.com/NoviScl/Design2Code) dataset, which is released under the **Open Data Commons Attribution License (ODC-By)** as it is built on the C4 corpus (also ODC-By). All perturbed HTML and screenshots here are derivative works and are redistributed under **ODC-By** with attribution. Copyright in the original website content remains with the respective site owners; the material is shared for research and reproducibility purposes.