CAD + process conditional-diffusion representations

Weak-label diagnostic only. 1 NG run(s), registration and defect extents are unvalidated, and local heater projection is invalid. Type is an observed diagnostic label, never a validated classifier.

Prepared representation and condition inputs

image_2d prepared input · voxel_3d prepared input · point_6000 prepared input · graph_6000 prepared input

Denoising loss

Same 14h06 CAD view and colour scale

representationnodes/runbest loss14h06 MAE mm14h06 RMSE mmCAD-correspondent Chamfer mmregistered-scan Chamfer mm14h06 weak AP
image_2d60000.59755.7387.5505.46932.2250.1428600761500438
voxel_3d25920.54776.1037.9205.85532.3630.23467333127356058
point_600060000.57645.6457.4025.17433.9310.057655521830424085
graph_600061550.42985.9147.6495.11132.7280.17699639616153692

Feature attribution diagnostics

image_2d attribution (JSON) · voxel_3d attribution (JSON) · point_6000 attribution (JSON) · graph_6000 attribution (JSON)

full report

TAG Graph U-Net candidate

Paper-inspired three-level kd-tree Graph U-Net inside the conditional DDPM. It is a weak-label, 1-run fit and is not a validation result.

14h06MAE mm ↓CAD Chamfer mm ↓weak AP ↑IoU@0.5 ↑uncertainty mm ↓
graph_60005.9145.1110.1770.03929.21
graph_unet_60009.6357.6530.0060.001317.09

same-view graph baseline vs TAG U-Net · metric comparison · kd-tree hierarchy · hierarchy JSON · attribution · full run report