Detecting
Defects at Scale
From MVTec AD to weld X-rays, track the models powering Industry 4.0 quality control. One-class learning. Zero defects.
Anomaly Detection Stats
Metrics that Matter
AUROC (Area Under ROC Curve)
The gold standard for anomaly detection. Measures the trade-off between true positive rate and false positive rate across all thresholds.
- Image AUROC: Binary classification per image.
- Pixel AUROC: Per-pixel anomaly localization.
- 99%+: Production-ready performance.
PRO (Per-Region Overlap)
Measures segmentation quality by averaging IoU across connected anomaly regions. More sensitive to small defects than pixel AUROC.
Anomaly Detection Leaderboard
Comparing architectures on standard datasets. Higher AUROC is better.
| Rank | Model | MVTec AD | VisA | Architecture |
|---|---|---|---|---|
| #1 | SimpleNet Research | 99.6% | 95.5% | Feature Adapter + Discriminator |
| #2 | EfficientAD Research | 99.1% | 94.8% | Student-Teacher + Autoencoder |
| #3 | PatchCore Amazon | 99.1% | 92.1% | Memory Bank + k-NN |
| #4 | FastFlow Research | 99.4% | - | Normalizing Flow + CNN |
| #5 | Reverse Distillation Research | 98.5% | - | Student-Teacher Reverse |
| #6 | CFLOW-AD Research | 98.3% | - | Conditional Normalizing Flow |
| #7 | DRAEM Research | 98.0% | - | Reconstructive + Discriminative |
| #8 | PaDiM Research | 97.9% | - | Multivariate Gaussian + Pretrained CNN |
Three Dominant Approaches
Memory Bank
Store embeddings of normal samples. Detect anomalies via nearest-neighbor distance. No training required after feature extraction.
Normalizing Flows
Learn the distribution of normal features. Anomalies have low likelihood under the learned distribution. Precise localization.
Student-Teacher
Student network mimics teacher on normal data. Discrepancy indicates anomaly. Extremely fast inference (600+ FPS).
Weld & Steel Inspection
Critical for pipeline integrity and structural safety. AI detects defects invisible to the naked eye in X-ray radiographs and surface inspection.
Common Defect Types:
YOLOv8 fine-tuned. Real-time at 30+ FPS.
Kaggle competition winning solution.
6 defect types on hot-rolled steel.
The Datasets
MVTec AD
20195,354 high-resolution images across 15 object and texture categories. The gold standard for industrial anomaly detection with pixel-level annotations.
MVTec 3D-AD
20214,147 high-resolution 3D point cloud scans and RGB images across 10 categories. First comprehensive 3D anomaly detection benchmark.
VisA
202210,821 high-resolution images across 12 objects with complex structures. More challenging than MVTec with realistic industrial scenarios.
KolektorSDD2
20213,335 images of electrical commutators with surface defects. Real industrial dataset with challenging small defects.
Weld Defect X-Ray
2021Radiographic images of welds with annotations for porosity, slag inclusion, lack of fusion, cracks. Critical for pipeline and structural inspection.
Severstal Steel Defect
201912,568 steel sheet images with 4 defect classes from Kaggle competition. Real industrial data from major steel producer.
Run Your Own Benchmarks
We're looking for GPU compute partners to run comprehensive anomaly detection benchmarks on industrial datasets.