The Unsupervised Visual Anomaly Detection Addon is designed for LabVIEW users who need to detect visual defects, outliers, or inconsistencies in images without annotated datasets.
Built on DeepLTK, this addon offers a streamlined workflow for integrating anomaly detection into industrial and research systems with minimal setup and configuration. Applications range from manufacturing quality control to biomedical screening and remote monitoring.
Key Features​
-
No need for labeled anomalies: Trains only on good samples
-
Data Efficient: Requires as few as 60-100 images for training
-
Fast Training: Less than 1 minute (10 seconds for 60 images)
-
Fast Inference: Around 10 ms per image
(3x faster than official PyTorch GPU implementation) -
High Throughput: Up to 3000 images/sec (Batch Size = 42)
-
High Accuracy: Proven on 80+ datasets with over 95% detection accuracy
-
Anomaly localization with heatmap visualization
-
Supports deployment on CPUs, GPUs and NI Real-Time targets
Integration & Requirements
​The addon integrates seamlessly with DeepLTK and can be combined with CuLab for GPU acceleration. It supports deployment on both Windows and NI Real-Time targets, making it suitable for development, testing, and production environments.
Installation
Provided as a VIPM (VI Package Manager) package, including the addon, documentation, and ready-to-run examples.
Requirements
-
DeepLTK v8.0.2 or later
-
CuLab v4.1.1 or later (optional for GPU acceleration)
-
LabVIEW 2020 or newer (32-bit or 64-bit; 64-bit recommended)
-
Windows 10 or 11 (64-bit)
Workflow
1. Collect Normal Samples
Provide a set of "good" or normal images for training
2. Train the Model
​Learn baseline visual patterns from normal data
3. Evaluate Performance
​Test accuracy and sensitivity on a validation or test dataset
4. Detect Anomalies
​Use the trained model to identify defects in new images
5. Visualize & Report
​Generate anomaly scores and heatmaps for the full dataset and export the results
UVAD is Available as a DeepLTK Add-On
UVAD integrates natively with DeepLTK and can utilize CuLab for high-speed GPU processing.
DeepLTK
Deep Learning Toolkit for
LabVIEW
A comprehensive, general-purpose toolkit for integrating deep learning into LabVIEW. Fully developed in LabVIEW, it requires no third-party frameworks such as PyTorch or TensorFlow. DeepLTK supports training and deployment of deep neural networks for a variety of tasks, including image classification, object detection, signal pattern recognition, and anomaly detection.​
CuLAB
GPU Toolkit for LabVIEW
A high-performance GPU acceleration toolkit for LabVIEW that offloads intensive tasks to NVIDIA GPUs. CuLab offers comprehensive APIs for vector math, matrix operations, filtering, and advanced signal processing. It allows custom parallel algorithms development within LabVIEW without external CUDA or C++ code. Ideal for high-throughput, low-latency applications like RF signal processing, computer vision, real-time analytics, and scientific computing.