agentleFS
JiamingShi-JsPhysJiamingShi-JsPhys/3d-fabric/skills/source-part-segmentation/SKILL.md

source-part-segmentation

Skill2 starsChanged 3 days agoMITView on GitHub

Segment overlapping visual parts from source images, wireframes, texture atlases, and decals before mesh reconstruction. Use when a mascot/logo/template contains touching or overlapping components and exact structural part masks are needed before contour-to-mesh, UV fitting, or landmark repair.

1 other project uses a version of this, 1 of them edited.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

11 rules. About 0 are common, 9 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Distinctive:

    Use when: Overlapping part extraction, component masks, marker/watershed segmentation, contour hierarchy analysis, atlas part isolation, …

    line 10
  2. Distinctive:

    Before the first Blender call, read ../text-to-blender/references/mcp-backends.md.

    line 12
  3. Very specific:

    Use this before contour-to-mesh when a source image contains overlapping or touching designed parts.

    line 15
  4. Distinctive:

    If components touch, run distance-transform marker watershed first.

    line 28
  5. Distinctive:

    If watershed over/under-splits, switch to seeded segmentation:

    line 29
  6. Very specific:

    create named part seeds (bbox, polygon, or seed_point + optional flood/HSV tolerance).

    line 30
  7. Distinctive:

    Do not infer repeated parts from symmetry.

    line 38
  8. Distinctive:

    Do not merge overlapping components if the manifest expects separate structural meshes.

    line 39
  9. Distinctive:

    Do not proceed to final modeling when part count differs between source images.

    line 40
  10. Distinctive:

    If automatic segmentation is ambiguous, write an ambiguity report and require or create manual seed rectangles/points.

    line 41
  11. Distinctive:

    Keep stroke/line masks separate from filled-part masks.

    line 42

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

MCP backend

Use when: Overlapping part extraction, component masks, marker/watershed segmentation, contour hierarchy analysis, atlas part isolation, manual seed fallback for source-locked Blender reconstruction.

Before the first Blender call, read ../text-to-blender/references/mcp-backends.md. Prefer official MCP for shared capabilities; preserve the selected backend for the whole task. Tool names in recipes are client/legacy examples: map them through that reference and its agent-host guide.

Source Part Segmentation

Use this before contour-to-mesh when a source image contains overlapping or touching designed parts. The output is not “nice masks”; it is a source-of-truth part inventory that downstream geometry must obey.

Inputs

  • source image, wireframe, decal, or texture atlas;
  • optional manual seed manifest with named parts, polygons, seed points, rough rectangles, or HSV/color ranges;
  • source manifest with structural/decorative/context classification and expected part count.

Workflow

  1. Choose the cleanest modality: alpha, edge, dark-line, bright-on-dark, color-band, or atlas region.
  2. Extract contours and hierarchy to identify candidate objects, holes, nested details, and strokes.
  3. If components touch, run distance-transform marker watershed first.
  4. If watershed over/under-splits, switch to seeded segmentation:
    • create named part seeds (bbox, polygon, or seed_point + optional flood/HSV tolerance);
    • save one mask per named structural part;
    • mark ambiguous overlaps explicitly instead of merging them.
  5. Classify masks as structural, decorative, face_feature, aura_context, or validation_only.
  6. Pass structural masks to contour-to-mesh; pass feature masks/landmarks to landmark-fit-repair; pass atlas regions to atlas-uv-fitting.

Hard rules

  • Do not infer repeated parts from symmetry; segment what the source shows.
  • Do not merge overlapping components if the manifest expects separate structural meshes.
  • Do not proceed to final modeling when part count differs between source images; write a conflict report or canonical policy.
  • If automatic segmentation is ambiguous, write an ambiguity report and require or create manual seed rectangles/points.
  • Keep stroke/line masks separate from filled-part masks; wireframe strokes are guides unless explicitly used as the contour boundary.

Seed manifest schema

{
  "schema": "source_part_seed_manifest.v1",
  "image": "path/to/source.png",
  "parts": [
    {"name":"leaf_top", "class":"structural", "bbox":[x,y,w,h], "mode":"non_background"},
    {"name":"face_shell", "class":"structural", "polygon":[[x,y],[x,y],...], "mode":"polygon"}
  ]
}

Allowed mode values: polygon, bbox, non_background, dark_lines, bright_on_dark, hsv_range.

Scripts

  • scripts/segment_source_parts.py produces component masks and a JSON report from an image, with optional watershed.
  • scripts/seeded_part_masks.py converts a named seed manifest into deterministic named masks and a part inventory.

Sources distilled

  • OpenCV contours/hierarchy/moments are the base measurement layer.
  • OpenCV distance transform + marker watershed is the first automated split method for touching components.
  • Active contour refinement can improve a rough mask boundary after segmentation.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.