widget-eval 0.1.0

Evaluation pipeline for widget generation — runs quality metrics and generates statistics reports.

Widget Eval

Evaluation pipeline for widget generation — runs widget-quality metrics and generates statistics reports.

Installation

pip install widget-eval

Usage

# Basic usage (CPU)
widget-eval --gt_dir /path/to/GT --pred_dir /path/to/results

# Use GPU
widget-eval --gt_dir /path/to/GT --pred_dir /path/to/results --cuda

# Skip evaluation, only generate statistics
widget-eval --gt_dir /path/to/GT --pred_dir /path/to/results --skip_eval

# Custom output directory and workers
widget-eval --gt_dir /path/to/GT --pred_dir /path/to/results --output_dir /path/to/stats --workers 8

Pipeline

1. Evaluation — Computes 12 quality metrics (layout, legibility, style, perceptual, geometry) for each GT-prediction pair
2. Statistics — Aggregates results into metrics_stats.json and metrics.xlsx

License

Apache-2.0