Benchmark Concepts

Core idea

This repository compares image-to-SVG workflows by keeping each step explicit:

1. input normalization
2. optional preprocessing
3. conversion preset execution
4. metric collection
5. report assembly

This removes “manual tuning drift” and makes side-by-side results auditable.

What counts as a benchmark run

One run is a combination of:

• source image
• preprocessing method (none or background-stripped)
• runner (VTracer)
• preset name

Why CSV + Markdown

CSV gives stable machine-readable rows for automation and charts. Markdown gives quick human review and visual references for SVG quality.

Extending the benchmark design

Add new runners by following the same contract:

• accept image inputs from CLI
• emit generated outputs to a deterministic folder
• write timing and complexity metrics to structured records
• include enough metadata to rerun the exact experiment

Keep this contract stable and existing scripts remain easy to compare across tools.