Two exact white Ram ProMaster 1500 reference photographs seed uncertain component hypotheses. Appearance proposes; geometry and part-to-part compatibility negotiate; ISH preserves unresolved evidence.
Each region starts with two close, sometimes incorrectly ordered appearance priors plus ISH. Every proposed label predicts where the other labeled parts should lie. Compatible arrangements reinforce recursively; incompatible arrangements inhibit one another.
Select either held reference view. Boxes are normalized annotations; the displayed names are the strongest labels after convergence.
The same front-left photograph is reduced to a 96×64 field. Tiny gradient filters propose five local labels; eight compatibility passes let neighboring orientations reinforce one another. This is an inspectable first step—not object recognition yet.
Color is the strongest current label; brightness is confidence. Relaxation is deliberately small and fixed so the before/after comparison remains honest and immediate.
ISH exists as a genuine competitor rather than a forced Boolean answer.A component-separated 3D model would let us render color, depth, normals, segmentation masks, and exact part identities from thousands of cameras. This checkpoint implements the downstream relaxation field now, so a future CAD renderer or ordinary feature detector can replace the frozen proposals without changing the relational engine.
The open community CAD link says “feel free to use it,” but supplies no formal redistribution license. It is recorded as a future import candidate, not copied into this repository. Ram’s official dimensional drawing is likewise cited but not redistributed because it is marked All Rights Reserved.
Both displayed photographs are by MercurySable99 and licensed CC BY-SA 4.0: front-left source and rear-left source. Engineering dimensions were cross-checked against Ram’s Body Builder’s Guide. Ram and ProMaster are trademarks of their respective owner; this independent research project is not affiliated with or endorsed by Ram or Stellantis.
cd python python3 -m vision_grrle.selftest python3 -m vision_grrle.demo # machine-readable result cat vision_grrle/output/promaster_results.json
