- Anomaly detection
Unsupervised image anomaly detection technology
Detects anomalies from just a few normal images.
- Calculates features using general-purpose models, so large-scale learning is not required.
- Compared to methods using reconstruction error, such as AE*1 and GAN*2, enables highly accurate anomaly detection from only a few normal images.
- Can be used both indoors and outdoors.
*1 Auto Encoder, *2 Generative Adversarial Network
Applications
- Visual inspections of parts and products (quality checks, inspections)
- Detects contaminations (recycling, manufacturing lines)
- Anomaly detection using security cameras (trespassers, infrastructure facility monitoring)
Benchmarks, strengths, and track record
- Learning requires only a small number of images.
- Can be used easily because stable accuracy can be achieved without adjusting hyper parameters.
- Completed trial using customer data. Currently proposing paid FS to customers.