Skip to content

Reports

What is generated

scripts/build_vtracer_report.py generates output/vtracer/experiments/report.md by reading:

  • output/vtracer/experiments/experiment_summary.csv
  • generated SVG artifacts under output/vtracer/experiments/

The report is meant for fast review and contains embedded SVG snippets for each candidate.

Published Sample Results

These repo-managed SVGs are representative outputs from the current VTracer baseline. Because output/ is ignored, the examples below are copied into docs/public/results/ for GitHub Pages publishing.

kiyoka_1 / cleankiyoka_1 / poster
kiyoka_1 cleankiyoka_1 poster
129.7 KB / 93 paths / 1.30 s61.1 KB / 36 paths / 0.44 s
kiyoka_2 / cleankiyoka_2 / poster
kiyoka_2 cleankiyoka_2 poster
210.0 KB / 123 paths / 0.84 s108.2 KB / 46 paths / 0.51 s

These published samples are intentionally compact. Use the locally generated report.md to inspect the full preset matrix.

Full Published Preset Matrix

Every currently published preset variant is listed below.

SourcePresetSVGSizePathsTime
kiyoka_1defaultlink734.2 KB8071.26 s
kiyoka_1cleanlink129.7 KB931.30 s
kiyoka_1cutoutlink231.8 KB1771.91 s
kiyoka_1polygonlink62.7 KB2961.12 s
kiyoka_1posterlink61.1 KB360.44 s
kiyoka_1detaillink1672.0 KB326811.29 s
kiyoka_1binary_inklink219.8 KB80.20 s
kiyoka_2defaultlink938.8 KB8320.87 s
kiyoka_2cleanlink210.0 KB1230.84 s
kiyoka_2cutoutlink386.8 KB2121.49 s
kiyoka_2polygonlink88.1 KB3390.91 s
kiyoka_2posterlink108.2 KB460.51 s
kiyoka_2detaillink2121.9 KB388413.65 s
kiyoka_2binary_inklink40.2 KB90.18 s

Folder layout

Typical output layout:

text
output/
  preprocessed/
    image1.nobg.png
  vtracer/
    experiments/
      image1.nobg__default.svg
      image1.nobg__clean.svg
      experiment_summary.csv
      report.md
  1. Run experiments after any preset change.
  2. Build report.
  3. Compare a baseline and candidate rows side by side.
  4. Store the report in local artifacts for that run.

Keeping reports auditable

Keep experiment_summary.csv and scripts versioned in git while avoiding large source images/artifacts. Keep only a curated set of public sample SVGs under docs/public/results/ so the docs stay lightweight.

Built for reproducible image-to-SVG benchmarking.