5743912da6
Trim image codec features in the CLI and Python crates
...
vtracer only decodes input, but image's default features pulled a full AV1
encoder (ravif/rav1e) and OpenEXR into the CLI binary and the Python wheel.
Restrict to decode-only input formats (png, jpeg, gif, bmp, webp, tiff, ico,
pnm, tga, qoi). The release binary drops from ~3.23 MB to ~2.46 MB and builds
faster; supported inputs are unchanged in practice (avif decode was never in
image's defaults anyway).
2026-07-24 15:56:28 +01:00
e46c971845
Rewrite into a vectorization framework (pillars 1–4)
...
Replace the 0.6.x single-pipeline crate with a stage-based framework, per
docs/design/. Implements Motivation pillars 1–4 (frontend, curve fitting,
color fitting, optimizer); mosaic (5) and bindings are deferred.
Workspace:
- crates/vtracer — the framework library (wasm-safe, no I/O)
- crates/vtracer-cli — thin CLI wrapper (clap 4 + image I/O)
- cmdapp/ and webapp/ excluded from the workspace (git-preserved)
Stages behind object-safe traits, composed by a Pipeline driver:
- Frontend: ColorClusterFrontend (+ transparency keying), BinaryFrontend
- CurveFitter: Pixel / Polygon / Spline (region tracing via visioncortex)
- ColorFitter: Identity, FixedPalette (OKLab-nearest), AutoQuantize
(area-weighted median cut), MergeAdjacent
- OptimizerPass: QuantizePass, SimplifyPass
- SvgWriter: relative/absolute shortest encoding, H/V/S shorthands,
compact number formatting, <g fill> grouping
visioncortex is a path dependency on the local 0.9.0 checkout.
Verified: 14 unit/integration tests pass; framework builds for
wasm32-unknown-unknown; CLI output renders faithfully via rsvg.
2026-07-23 22:27:24 +01:00
Chris Tsang