Evaluate Streamline open/free sets as corpus sources

[swernerx]

Goal

Evaluate Streamline open/free icon and illustration sets as a source for Morphēa corpus tests, especially two-color icons and flat multi-color illustrations.

Why this matters

Streamline assets are professionally designed, visually consistent, and broader than the current small real-image baseline. They could stress exactly the areas Morphēa should improve next:

• dual-tone icons;
• flat multi-color illustration regions;
• consistent stroke/fill systems;
• repeated visual language across many assets;
• realistic designer-authored SVG structure that can be rendered to PNG and reconstructed.

This could become a useful middle ground between generated toy fixtures and fully local real-image cases.

Candidate sources

Initial source review:

• Streamline homepage: https://www.streamlinehq.com/
• Streamline free license: https://help.streamlinehq.com/en/articles/5354376-streamline-free-license
• Streamline open vector repository: https://github.com/webalys-hq/streamline-vectors

Important licensing distinction:

• Streamline says some sets are open-source under Creative Commons Attribution 4.0 and require attribution.
• Other free sets are free to use but not open-source; the free license restricts using them as user-available assets inside asset-builder / Canva-style applications.
• The webalys-hq/streamline-vectors repository states CC BY 4.0 and requires attribution plus a link to Streamline.

Proposed first step

Use only explicitly open-source / CC BY 4.0 Streamline vectors until licensing is reviewed.

Start by auditing the public streamline-vectors repository:

• identify which folders contain icons vs illustrations;
• classify asset styles: line, block, plump, multi-color, emoji/illustration, logo-like, etc.;
• sample a small subset, for example 12-24 assets;
• render SVG -> PNG locally;
• run Morphēa PNG -> SVG/manifest;
• compare recovered structure against the original SVG structure.

Acceptance criteria

• Add a documented source-audit file naming the Streamline set, source URL, license, attribution requirement, and allowed repository usage.
• Do not commit Streamline assets unless attribution and redistribution terms are satisfied.
• If assets are committed, include appropriate attribution text and source links.
• Prefer a config-driven smoke that writes generated render/vectorize artifacts to /tmp.
• Evaluation separates visual similarity from editable structure: primitive class, group coherence, layer depth, node/parameter budget, palette recovery, and cut-out handling.
• The first test subset includes at least one dual-tone icon and one flat multi-color illustration-like asset if available in the open-source set.

Risks / constraints

• Free-but-not-open-source Streamline sets should not be treated as corpus assets without explicit review.
• Public gallery examples may need stronger attribution and licensing checks than local tests.
• Some Streamline assets may already be highly optimized SVGs; the useful test is whether Morphēa can recover comparable editable structure from a rasterized version, not whether it can copy the source SVG.

Relationship to other issues

• Complements Add synthetic curriculum generators for model training #9 synthetic curriculum generators by adding professional designer-authored fixtures.
• Complements Add curated multi-color vector illustration case #2 multi-color vector illustration case with a scalable source of flat-color examples.
• Could feed Add artifact-based vectorizer comparison examples #6 artifact-based vectorizer comparisons once licensing and attribution are handled.