CNF DEVELOPER HANDBOOK

1. Purpose & Philosophy

The Chaparrito Nalgón Framework (CNF) is a standardized, modular computer-vision pipeline for lower-body contour geometry extraction, denim fit analysis, cloth-body interaction modeling, SMPL-X mesh reconstruction, fit-grade scoring, and manufacturing pattern optimization.

CNF is designed to be:

• Deterministic — same input → same metrics
• Modular — each subsystem deploys independently
• Scalable — GPU/CPU separation
• Interoperable — JSON schemas, API bindings, microservice architecture
• Explainable — every metric is traceable to geometry

2. System Overview Guided Link: Technical Documentation →

CNF consists of six major layers:

3. Ingestion Layer Guided Link: Live Scanner →

• Accept user uploads or camera captures.
• Validate image resolution and format.
• Store raw image in object storage.
• Emit NEW_SCAN event to orchestrator.
• Assign cnfcaseid and set initial scan_status.

4. CV Core (GPU) Guided Link: cv-pipeline.js →

• 4.1 Preprocessing: resize, normalize, estimate camera intrinsics.
• 4.2 Segmentation: person mask, jeans mask, seat region mask.
• 4.3 Pose Estimation: 2D keypoints → 3D skeleton via single-view lifting.
• 4.4 SMPL-X Body Mesh: fit shape parameters β, fit pose parameters θ, align silhouette.
• 4.5 Cloth Mesh Reconstruction: canonical jeans mesh, offset field deformation, tension-aware refinement.
• 4.6 Depth/Normal Refinement: optional but recommended for curvature accuracy.

5. CNF Metrics Engine (CPU) Guided Link: 3D Viewer →

• 5.1 Seat Curvature Radius: cross-sections sampled at multiple vertical levels.
• 5.2 Projection Vector: distance from pelvic reference plane to posterior mesh.
• 5.3 Contour Stability: silhouette smoothness + left/right symmetry.
• 5.4 Cloth-Body Distance Fields: used for FitBand classification.
• 5.5 Tension Zones: wrinkle density + highlight intensity + seam displacement.
• 5.6 FitBand Classification: tight / regular / relaxed / Levi’s-compatible.

6. CNF Outputs (JSON Schemas) Guided Link: Technical Standard →

Every CNF case produces four standardized outputs:

• cnffitgrade_report — seat grade, tier, score breakdown.
• cnfmeshspec — mesh validity, vertex count, reconstruction accuracy.
• cnftensionheatmap — tension zone intensities and classification.
• cnfcertlabel — final label with pass/fail flags and seal ID.

7. API Gateway (v1.2) Guided Link: API Reference →

• POST /api/cnf/scan
• GET /api/cnf/scan/{cnfcaseid}
• GET /api/cnf/fit-grade/{cnfcaseid}
• GET /api/cnf/mesh/{cnfcaseid}
• GET /api/cnf/tension/{cnfcaseid}
• GET /api/cnf/cert/{cnfcaseid}

Version header: X-CNF-Version: 1.2

Developer-ready OpenAPI contract:

• OpenAPI 3.0.3 YAML

8. Database Schema

• cnf_users
• cnf_scans
• cnfmeshspecs
• cnffitgrade_reports
• cnftensionheatmaps
• cnfcertlabels

9. Microservice Architecture

• upload_api — ingestion and validation
• orchestrator — pipeline coordination
• cv_engine — GPU-based CV core
• metrics_engine — CNF metric computation
• api_gateway — REST/WebSocket exposure

10. Certification Workflow Guided Link: CNF Handbook →

1. Stage 1: Scan ingestion — assign cnfcaseid, store raw image, trigger orchestrator.
2. Stage 2: CV processing — segmentation, pose lifting, SMPL-X fitting, cloth mesh, optional depth refinement.
3. Stage 3: Metric extraction — curvature, projection vector, distance fields, tension zones, FitBand.
4. Stage 4: Fit-grade report — assign CNF grades A/B/C, evaluate Levi’s compatibility, set Mesh-Ready flag.
5. Stage 5: Certification label — seat grade, stability, tension, FitBand, pass/fail, Master Mold candidate flag.
6. Stage 6: Storage & API — store all artifacts, expose via /api/cnf/scan/{id}.
7. Stage 7: Master Mold integration — high-grade cases promoted to template updates.

11. Developer Training

• Concept exam — CNF metrics, terminology, compliance tiers.
• Pipeline exam — CV stages, mesh reconstruction, metric extraction.
• Practical case evaluation — run end-to-end scan, verify outputs.
• Reviewer guidelines — analyst approval, provenance checks, override rules.

12. Roadmap

• v1.3 — multi-view fusion
• v1.4 — cloth simulation integration
• v2.0 — full 3D scanning alignment