widget2code-bench 0.1.0

Benchmark evaluation for widget code generation — 12 quality metrics across layout, legibility, perceptual, style, and geometry.

widget2code-bench

Benchmark evaluation for widget code generation — 12 quality metrics across layout, legibility, perceptual, style, and geometry.

Installation

pip install widget2code-bench

Usage

widget2code-bench \
  --gt_dir /path/to/GT \
  --pred_dir /path/to/predictions \
  --pred_name output.png

Directory Structure

• GT dir: flat image files with 4-digit IDs in filenames (e.g. gt_0001.png)
• Pred dir: subfolders with 4-digit IDs in names, each containing --pred_name file

gt_dir/                     pred_dir/
  gt_0001.png                 image_0001/
  gt_0002.png                   output.png
  ...                         image_0002/
                                output.png

Options

Flag	Default	Description
--gt_dir	(required)	GT directory (flat image files)
--pred_dir	(required)	Prediction directory (subfolders)
--pred_name	output.png	Prediction filename inside each subfolder
--output_dir	{pred_dir}/.analysis	Statistics output directory
--workers	4	Parallel threads
--cuda	off	Enable GPU
--skip_eval	off	Skip evaluation, only generate statistics

Pipeline

1. Evaluation — Computes 12 quality metrics for each GT-prediction pair
2. Statistics — Aggregates results into metrics_stats.json and metrics.xlsx

License

Apache-2.0