Cross-Paradigm Leaderboard

9 methods evaluated across 4 learning paradigms on the Industrial3D test set (Areas 6 + 12, 84.9M points). Metric: mean IoU (mIoU) over 12 classes via vote-based evaluation.

Test Set
Areas 6 + 12
84.9M points, 2 areas from different facilities
Metric
mIoU (12-class)
Mean intersection over union across all classes
Evaluation
Vote-Based
test_smooth = 0.95, multi-pass inference
Split Protocol
Area-Based
S3DIS-style, no room-level leakage
Framework
PyTorch
Unified codebase for all methods; TF methods re-implemented
Val Set
Area 9
15.1M points, used for hyperparameter selection only

All Methods

+ Submit Result
sorted by miou ↓
# Method Paradigm Labels mIoU (%) OA (%) Year Code / Status
Loading results…

Click column headers to sort. Results verified by the Industrial3D team. OA = overall accuracy. — = not reported.

Per-Class IoU (12 Classes)

Vote-based per-class IoU for methods with full results available. Head classes dominate performance; tail classes (Reducer, Strainer, Pump, Elbow, Tee) remain near zero for most methods.

Loading…

Cyan = best per class among listed methods. 0.00 = zero IoU on that class. Only vote-based (smooth test) results shown.

The Domain Transfer Gap

39.95 pp

Best supervised method (Boundary-CB): 55.74% mIoU
Best zero-shot foundation model (Point-SAM One-vs-Rest): 15.79% mIoU

A 39.95 percentage-point gap quantifies the unresolved industrial domain transfer challenge. Even the oracle Point-SAM (21.08%) — given GT mask proposals — trails the best supervised method by 34.66 pp. This gap arises from the dual crisis: foundation models trained on architectural/outdoor data are ill-equipped to distinguish geometrically ambiguous MEP fittings under extreme class imbalance.

Remarkably, SQN with only 0.1% labels (44.29%) outperforms fully-supervised RandLA-Net (39.83%), suggesting that the inductive bias of sparse-label training may be better suited to this domain than dense supervision with vanilla architectures.

Have better results?

Submit your method's results to the leaderboard. We welcome results from all paradigms, including few-shot, semi-supervised, and multi-modal approaches.

Submit Results →