What if building an industrial computer vision model no longer required switching between an online annotation service, Python notebooks, command-line training tools, conversion scripts and a separate deployment environment?
With the LabVIEW Annotation Tool, now available through the SOTA ecosystem, Graiphic brings the complete computer vision workflow into one coherent, local and LabVIEW-centered environment.
Developers can import a dataset, annotate images, prepare and augment the data, configure a model, train it locally, test the results and generate the LabVIEW projects required for integration into an industrial application.
One tool. One workflow. No mandatory cloud dependency. From raw images to an operational LabVIEW application.
The Missing Link Between a Dataset and an Industrial Machine
Training an object detection model is often presented as a simple sequence: collect images, draw bounding boxes and start training.
In practice, the workflow is usually fragmented across several technologies.
A typical team may use one platform to manage the dataset, another application to annotate images, a Python environment to train the model, additional scripts to export or convert it, and a custom software layer to integrate the final model into the target system.
Each transition introduces additional work:
- Data must be copied or uploaded between environments.
- Dataset formats must be converted.
- Dependencies and framework versions must remain compatible.
- Training configurations must be reproduced manually.
- The final model must be connected to the industrial application.
- Engineers must maintain several tools, languages and deployment processes.
The LabVIEW Annotation Tool was created to remove these unnecessary boundaries.
It is not simply an image-labeling application. It is the data and project-preparation layer of a wider end-to-end AI architecture built around the LabVIEW Deep Learning Toolkit, the LabVIEW Computer Vision Toolkit and the SOTA platform.
What Is the LabVIEW Annotation Tool?
The LabVIEW Annotation Tool, also presented as Annotator within SOTA, is a graphical environment for preparing computer vision datasets and transforming them into trainable and deployable LabVIEW projects.
Its role is to centralize the operations that normally happen before, during and immediately after model training.
The tool provides a workflow for:
- Importing existing image datasets.
- Reviewing and organizing images and classes.
- Creating and editing annotations.
- Preparing datasets for detection, segmentation or classification tasks.
- Applying preprocessing and data augmentation.
- Selecting a compatible model and training architecture.
- Generating the required LabVIEW training project.
- Training and testing the model locally.
- Preparing the trained model for native integration into a LabVIEW application.
This creates a direct path from the first image to the final industrial architecture.
A Complete End-to-End Workflow
| Stage | What the Annotation Tool Provides |
|---|---|
| 1. Import | Load existing image datasets and convert supported dataset structures into a coherent project. |
| 2. Review | Inspect images, labels, classes and existing annotations before training. |
| 3. Annotate | Create or edit the visual annotations required by the selected computer vision task. |
| 4. Prepare | Apply preprocessing, augmentation and dataset organization operations. |
| 5. Configure | Select the model, architecture and training parameters appropriate for the application. |
| 6. Train | Execute model training locally through the LabVIEW Deep Learning Toolkit. |
| 7. Test | Evaluate the model on representative images and inspect its detections. |
| 8. Generate | Create the LabVIEW projects required for training, inference and integration. |
| 9. Deploy | Connect the trained model to the camera, test bench, machine or production application. |
This is what end-to-end AI means in practice: the dataset, model and deployment architecture remain connected throughout the complete lifecycle.
Import More Than 40 Dataset Formats
Computer vision datasets are distributed through many different formats and ecosystems.
Depending on the original tool or model family, annotations may use different directory structures, coordinate conventions, class definitions and metadata files.
The LabVIEW Annotation Tool supports more than 40 dataset formats covering computer vision tasks such as:
- Object detection.
- Image segmentation.
- Image classification.
This enables users to reuse existing datasets instead of rebuilding them manually.
Datasets sourced from public ecosystems such as Roboflow can also become part of the workflow, providing access to a broad range of existing images and annotations for experimentation, transfer learning and application prototyping.
The environment is designed to work with projects ranging from relatively small collections to very large datasets containing hundreds of thousands or potentially millions of images, depending on the storage, memory and computing resources of the target system.
Annotation Directly Inside the LabVIEW Ecosystem
Annotations define what the model must learn.
For an object detection project, the user identifies the objects and associates each region with a class. For segmentation, the annotation describes the relevant visual region more precisely. For classification, images are organized according to the categories that the model must distinguish.
The Annotation Tool provides the visual environment required to inspect, create, modify and validate this information.
Keeping this step close to the rest of the LabVIEW workflow offers several advantages:
- The dataset remains connected to the final engineering project.
- Classes and annotations can be reviewed by domain specialists.
- Errors can be corrected without returning to a separate cloud platform.
- The same team can manage data preparation, training and integration.
- The annotation workflow can remain inside the organization’s controlled infrastructure.
For industrial AI, annotation is not merely a drawing task. It is the process through which engineering knowledge is transferred into the dataset.
An expert knows the difference between an acceptable surface variation and a real defect. The Annotation Tool provides the interface required to capture that expertise and transform it into training data.
Preprocessing and Data Augmentation
Model performance depends not only on the architecture but also on the quality and diversity of the training data.
Industrial images may vary because of lighting, camera position, production tolerances, reflections, focus, noise or changes in the surrounding environment.
The LabVIEW Annotation Tool includes preprocessing and data augmentation capabilities designed to prepare more robust datasets.
These operations can help teams:
- Normalize heterogeneous image collections.
- Increase the diversity of the training data.
- Reduce sensitivity to controlled visual variations.
- Prepare images for the selected model input requirements.
- Improve the balance between classes and use cases.
Augmentation does not replace representative real-world data, but it can significantly improve the value extracted from the available dataset when applied carefully.
Train the Model Locally
Once the dataset is prepared, the workflow continues with the LabVIEW Deep Learning Toolkit.
The training project can be generated and executed locally, without requiring the dataset to be uploaded to an external cloud training service.
Where compatible NVIDIA hardware and the appropriate runtime configuration are available, training can use CUDA acceleration to reduce execution time.
This local architecture is especially important for organizations working with:
- Confidential production images.
- Unreleased products and prototypes.
- Defense or aerospace systems.
- Medical or scientific research data.
- Customer-specific manufacturing processes.
- Intellectual property that cannot leave the organization.
No mandatory cloud dependency means that the organization retains control over where the dataset is stored, where the model is trained and how the resulting artifacts are managed.
Local execution does not remove the need for good cybersecurity practices, but it eliminates the mandatory transfer of sensitive images to an external annotation or training service.
Data Sovereignty, Confidentiality and Traceability
For many industrial projects, confidentiality is not an optional benefit. It is a deployment requirement.
A dataset may reveal far more than individual images. It can expose:
- Product geometry.
- Manufacturing processes.
- Known failure modes.
- Inspection limits.
- Customer components.
- Future product designs.
- Research results.
By keeping annotation and training within a controlled local environment, companies can define their own storage, access, backup and security policies.
This also improves traceability. A project can retain the exact dataset version, class definitions, model configuration, training parameters and deployment artifact associated with a production release.
For quality systems, regulated environments and long-lifecycle industrial machines, this reproducibility is often as important as the model’s raw accuracy.
No Mandatory Python Glue Code
Python remains a powerful and important language for AI research and development.
However, requiring every LabVIEW engineer to maintain Python environments, package versions, command-line tools, notebooks and custom integration services creates additional complexity.
The standard Annotation Tool workflow does not require the user to write Python scripts to move from annotation to training and LabVIEW integration.
This provides several practical gains:
- Fewer external dependencies to install and maintain.
- Less framework-specific glue code.
- A more visual and inspectable engineering workflow.
- Easier transfer of the project between LabVIEW teams.
- Reduced risk of environment and package-version conflicts.
- A more direct path from prototype to production architecture.
The objective is not to reject the Python ecosystem. It is to give LabVIEW teams a complete alternative when the final system is already being developed and maintained in LabVIEW.
From the Trained Model to the Industrial Application
Training a model is only one step. The real industrial value appears when the model is integrated into the machine, test bench or inspection system.
The Annotation Tool prepares LabVIEW projects that can be connected to the rest of the SOTA ecosystem.
The complete architecture may combine:
- The LabVIEW Computer Vision Toolkit for image acquisition, preprocessing, visualization, regions of interest and post-processing.
- The LabVIEW Deep Learning Toolkit for model training and execution.
- The Annotation Tool for dataset preparation and project generation.
- SOTA runtime components for installation, updates and hardware configuration.
- The existing LabVIEW application for instrumentation, measurements, test sequences and industrial communication.
The trained model can then become a native component of a larger LabVIEW dataflow.
A camera image can be combined with electrical measurements, pressure, temperature, vibration, dimensional data or production context before the application produces its final decision.
The project can also be packaged into a LabVIEW deployment architecture, including a single executable when the target design and dependencies permit it.
PCB Defect Detection in Under Two Minutes
The first demonstration shows an end-to-end object detection workflow for PCB defect detection.
In approximately one minute, the video illustrates the complete value chain:
- Loading the image dataset.
- Annotating the relevant defects.
- Preparing the training project.
- Training the model locally.
- Testing the resulting detector.
- Preparing native LabVIEW integration.
This type of application can support the identification of visible PCB anomalies such as missing elements, incorrect placement, assembly defects or other classes defined by the engineering team.
Screw Detection: A Simple Example with Immediate Industrial Value
The second demonstration uses screw detection to show how the same workflow can be applied to assembly verification.
A detection model can be trained to identify the presence, absence or position of screws and other fastening elements.
The same principle can be extended to:
- Connector presence and engagement.
- Component positioning.
- Assembly completeness.
- Label and marking verification.
- Packaging inspection.
- Tool or accessory detection.
- Foreign-object detection.
Watch the Original Annotation Tool Demonstration
The initial presentation introduced the vision behind the product: transform LabVIEW into a complete local environment for dataset preparation, annotation, training and deployment.
That early demonstration showed the foundations of a tool designed to remove fragmentation from industrial computer vision projects.
The Annotation Tool has now moved beyond that development-preview phase and is available as part of the SOTA ecosystem.
Designed for Test, Measurement and Industrial Inspection
The Annotation Tool is particularly relevant when computer vision must become part of a wider engineering system.
In a test bench, a camera can act as another measurement source.
The application can correlate the visual result with:
- Voltage and current measurements.
- Temperature and pressure.
- Vibration or acoustic data.
- Dimensional measurements.
- Test-sequence states.
- Product serial numbers.
- Manufacturing and traceability information.
This enables more complete anomaly detection.
For example, an electrical deviation can be associated with a visible assembly defect. A pressure-test failure can be correlated with an incorrectly positioned seal. A vibration anomaly can be linked to a missing or loose component.
LabVIEW remains responsible for the complete test architecture, while the trained computer vision model adds a new source of information to the decision process.
Who Is the Tool Designed For?
LabVIEW Engineers
Teams already developing machines, test benches and acquisition systems in LabVIEW can add computer vision AI without rebuilding the application around another software environment.
Machine Vision and Automation Integrators
Integrators can create customer-specific object detection and segmentation models while retaining control over the complete deployment architecture.
Industrial R&D Teams
Researchers and engineers can rapidly test whether a dataset contains enough information to solve a visual inspection problem and then move the successful experiment toward production.
Organizations with Confidential Data
Companies that cannot upload images to a third-party service can keep dataset preparation and training within their controlled infrastructure.
Academic and Research Laboratories
The graphical workflow provides a practical environment for teaching and studying annotation, data preparation, training, model evaluation and deployment without making cloud access a prerequisite.
What Normally Requires Several Tools Becomes One Workflow
A conventional computer vision project may involve three to five separate tools before the model reaches production.
The LabVIEW Annotation Tool consolidates those stages into a unified workflow.
This can produce measurable gains in:
- Integration time.
- Project setup time.
- Dataset conversion effort.
- Software maintenance.
- Training reproducibility.
- Deployment consistency.
- Engineering handover.
It can also reduce recurring costs associated with cloud annotation, hosted training environments and external pipeline services.
The largest economy is often not the removal of one subscription. It is the reduction of engineering time spent connecting, debugging and maintaining fragmented tools.
An Important Building Block of SOTA
The Annotation Tool illustrates the wider purpose of SOTA.
SOTA is not intended to be a collection of unrelated VIs or libraries. It is a complete AI ecosystem designed to cover the lifecycle from data preparation to deployment.
Within that ecosystem:
- Annotator prepares datasets and generates projects.
- The LabVIEW Computer Vision Toolkit manages images, cameras, preprocessing, display and post-processing.
- The LabVIEW Deep Learning Toolkit trains and executes models.
- Execution runtimes and hardware integrations provide optimized local processing.
- LabVIEW connects AI with measurements, control logic, hardware and industrial interfaces.
This is how Graiphic is transforming LabVIEW into a true end-to-end AI engineering environment.
A Foundation for More Automated Annotation
The current release establishes the essential foundation: dataset import, manual annotation, preparation, local training, testing and deployment.
This architecture also opens the path toward increasingly assisted annotation workflows.
Vision-language models and Generative AI can eventually help identify objects, propose annotations, classify images or accelerate the review of large datasets, while the engineer retains final control.
The SOTA GenAI and Computer Vision foundations make this a natural direction for future evolution.
Available Now Through SOTA
The LabVIEW Annotation Tool is now available through SOTA.
It provides a direct route from a collection of images to an operational computer vision model integrated into a LabVIEW application.
No mandatory cloud service. No mandatory Python scripting. No fragmented deployment chain.
Yes, the annotation workflow runs in LabVIEW.
Yes, the model can be trained locally.
Yes, the result can be integrated directly into an industrial LabVIEW architecture.
And yes, SOTA is currently available free of charge.
Frequently Asked Questions
Is the LabVIEW Annotation Tool released?
Yes. The tool is now available through the SOTA ecosystem and is no longer limited to the early development version shown in the original announcement.
Does it require a cloud service?
No mandatory cloud service is required for the standard workflow. Dataset preparation and model training can be performed locally on controlled hardware.
Can the model be trained directly from LabVIEW?
Yes. The generated training project uses the LabVIEW Deep Learning Toolkit to execute the model-training workflow locally.
Does the user need to write Python code?
No Python scripting is required for the standard annotation, project-generation, training and LabVIEW integration workflow.
Which computer vision tasks are supported?
The tool is designed for datasets and projects involving object detection, segmentation and classification. Exact model and format compatibility depends on the current SOTA release.
How many dataset formats are supported?
The tool supports more than 40 dataset formats across the supported computer vision tasks.
Can it use a GPU?
Training can use CUDA acceleration when compatible NVIDIA hardware, drivers and the appropriate SOTA runtime configuration are installed.
Can the result be deployed as a LabVIEW executable?
The generated project is designed for integration into a LabVIEW deployment architecture and can be packaged into an executable when the project design and runtime dependencies permit it.
Is SOTA free?
At the time of publication, SOTA and its available tools and toolkits can be downloaded free of charge from Graiphic.
