widget2code-bench 0.2.3

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

# 1. Install PyTorch with CUDA support first (skip if CPU-only)
pip install torch torchvision --index-url https://download.pytorch.org/whl/cu126

# 2. Install widget2code-bench
pip install widget2code-bench

Note: PyPI only ships CPU-only PyTorch. To use --cuda, you must install PyTorch from the official index before installing this package.

Usage

Single image mode

Evaluate one GT-prediction pair. Prints JSON results to stdout, no files saved.

widget2code-bench \
  --gt_image /path/to/gt.png \
  --pred_image /path/to/pred.png \
  --cuda

Batch mode

Evaluate all matched pairs in directories.

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

Directory Structure (batch mode)

• 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_image	—	Single GT image path
--pred_image	—	Single prediction image path
--gt_dir	—	GT directory (flat image files)
--pred_dir	—	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
--no_fill	off	Disable fill-image evaluation for missing predictions (fill is on by default)

Output (batch mode)

1. Evaluation — Saves evaluation.json in each prediction subfolder + evaluation.xlsx in pred_dir
2. Statistics — Saves metrics_stats.json and metrics.xlsx to {pred_dir}/.analysis/

Handling missing predictions (fill mode, default)

When a GT image has no matching prediction, the evaluator also scores it against synthetic fill images, so the summary xlsx can show how different assumptions about missing samples affect the aggregate metrics. Each summary xlsx contains up to 4 rows:

Row	Description
<run>	Average over matched pairs only
<run> (+ black fill)	Missing preds scored against an all-black image
<run> (+ white fill)	Missing preds scored against an all-white image
<run> (+ zero fill)	Missing preds contribute the worst-case value (LPIPS = 1.0, others = 0)

Two extra columns are appended after Geometry:

• SuccessRate / ratio — matched pairs / total GT, as a percentage
• SuccessRate / count — e.g. 993/1000

Pass --no_fill to disable this behavior (only row 1 is produced and missing preds are skipped).

All metrics are higher-is-better except lp (LPIPS), which is a distance (lower-is-better).

License

Apache-2.0