GCP Vertex AI · Model registry

Model performance

Compare, configure and deploy vision models to production

Production active

YOLO-V8.4

Optimized computer vision model for high-speed detection of surface, edge and dimensional defects across machined components.

FrameworkUltralytics / PyTorch

Dataset186,420 images

Last trainedAug 12, 2026

99.21%

Weighted F1 score

Inference latency18 ms

Object detection

Active

General defect detector tuned for mixed production lines.

F1 score99.21%

Latency18 ms

Image classification

Validated

High-precision surface texture and coating classifier.

F1 score98.84%

Latency12 ms

Anomaly detection

Testing

Unsupervised detector for new and uncommon failure patterns.

F1 score97.96%

Latency24 ms

Decision threshold · Line A-07

82%

Higher recallBalancedHigher precision