Metrics and Outputs

Current metric set

The default report currently contains:

• runtime (seconds)
• output SVG byte size
• SVG path count
• unique fill count
• output width and height
• generated artifact file names

These metrics are suitable for ranking throughput and structural complexity.

How to read the CSV

output/vtracer/experiments/experiment_summary.csv has one row per image, preset, and run condition.

Recommended reading pattern:

1. sort by size + runtime to find efficient settings
2. scan fill/path counts for topology complexity
3. use report previews to confirm visual fidelity

Current limits

• no structural image-level ground-truth fidelity score yet
• no editability quality score yet
• no cross-run hardware normalization by default

All limits are intentionally documented so future additions can target gaps explicitly.