Examples and Application Notes

This section summarizes AIVT-OV All the sample programs included in the toolkit cover traditional image processing, target detection, semantic segmentation, text recognition, large language models and other directions to help users quickly understand the use of each module, quickly run and modify examples, and complete their own project prototypes.

Sample overview structure

The toolkit sample is located at the following path:

  • Path mode: LabVIEW Installation Directory

  • Menu mode: LabVIEW → Help → Find Examples → Directory Structure → VIRobotics

The following sample sets are currently included:

  1. AI Vision → CVTraditional Image Processing

  2. AI Vision → License Management: License Activation Management

  3. AI Vision → Object Detection: YOLO series target detection (including pose, seg, OBB)

  4. AI Vision → Segment: Semantic segmentation (Deeplab, SAM, Yiku-Seg)

  5. AI Vision → PaddleOCRppocrv3/v4/v5 Series Text Detection and Recognition

  6. GenAI → LLM: large language model, calling the locally deployed large language model (e. g. Qwen/Qianwen/DeepSeek, etc.) for natural language interaction or text generation.

✅(Reserved)GenAI → VLM: Multimodal Visual Language Model (future version)

1. CV (Legacy Vision Example)

Path:VIRobotics → AI Vision → CV

SubdirectoryFunction Description
Image Reading and CapturingImage reading, image saving, camera capture, video playback and saving
Image ProcessingEdge detection, filtering, corrosion dilation, threshold processing, etc.
Template MatchingTemplate matching (grayscale / Edge / multiple target matching)
Feature Detection and DescriptionFeature point extraction (Harris, ORB) and descriptor matching
Machine LearningSimple SVM Classification, etc.
Mat OperationsImage and matrix basic operation support (addition, subtraction, multiplication and division, flip, etc.)

🧠 Scenario: Image pre-processing and traditional vision tasks that do not rely on deep learning models, can be used for image pre-processing or combined with AI inference

2. License Management (activation example)

Path:VIRobotics → AI Vision → License Management

VI NameFunction Description
Activate License Local.viLoad Local .lic file to activate
Activate License through Network.viUse License ID / Key Online activation
Get License State.viGet the current activation status and authorization information
Open License Manager.viOpen the full license management tool UI

🧠 Applicable scenario: This module can be packaged to the deployment side EXE Activate.

3. Object Detection (Target Detection Example)

Path:VIRobotics → AI Vision → Object Detection

SubdirectoryModel Example
YOLOv5 ~ YOLOv9Mainstream Detection Tasks
YOLOv8_OBB / YOLO11_OBBRotating frame inspection (e. g. PCB/device)
YOLOv8_pose / YOLO11_posekey point detection
YOLOv8_seg / YOLO11_seginstance segmentation
YOLO11 / YOLO12 / YOLO13A new generation of lightweight high-precision model
RT_DETRtransformer structure target detection model

Operating instructions:

  1. Selectable device: CPU / GPU(Intel iGPU Such Iris Xe, A770)

  2. Custom model way:

    • Replace model_file Control Path

    • Replace class_names_file Control Path

  3. The resulting output includes detection frame coordinates, confidence levels, category labels, and support for image display.

🧠 Applicable scenarios: parts identification, defect detection, safety helmet identification, traffic identification, rotating parts, label direction identification, direction calibration of circuit board components, action analysis, behavior detection, worker action judgment and other scenarios.

4. PaddleOCR (Text Recognition Example)

Path:VIRobotics → AI Vision → PaddleOCR

VI NameFunction Description
paddleOcr_Openvino.viStandard Edition OCR, including OCRv3/v4/v5, full coverage of detection and recognition process
paddleOcr_Openvino_easy.viSimplified version, suitable for fixed scene quick test

🧠 Application scenario: suitable for various text recognition tasks, such as industrial control panel reading, two-dimensional code area recognition, label/nameplate text extraction, bill recognition, etc. It is suitable for text detection in Chinese and English mixed and natural scenes. The input can be camera and image folder, and horizontal Chinese text recognition is supported.

5. Segmentation (Semantic Segmentation Example)

Path:VIRobotics → AI Vision → Segment

SubdirectoryContent
Deeplabv3DeepLabV3/V3+ Real-time and batch segmentation
SamSegment Anything Model, supports mouse click segmentation
Yiku_SegSelf-developed lightweight segmentation model for industrial deployment

🧠 Applicable scenarios: real-time human-vehicle segmentation, medical image segmentation, scientific research image segmentation, pipeline defect area positioning, appearance scratch extraction, automatic driving and other scenarios

6. LLM (Large Language Model Example)

Path:VIRobotics → GenAI → LLM

VI NameFunction Description
LLM_Basic.viSend Plain Text Prompt → Get Reply
LLM_Stream.viSupport token level stream generation (output while input)

🧠 Applicable scenarios: applicable to natural language interface development, such Chat System, question and answer system, industrial equipment manual question and answer.

🔮 Preview: VLM (Visual Language Model Example)

📌 This catalogue will be used for future introduction VLM(Vision-Language Model) example, such:

  • Graphic Questions and Answers (VQA)

  • multimodal retrieval

  • Image Generation Description (Captioning)

  • text control image editing (InstructPix2Pix etc.)

Please look forward GenAI → VLM Subdirectory updates.

Example usage recommendations

Purpose of userecommend module
Familiar with image processing and cameraCV
Learning Object Detection Model inferenceObject Detection
Rapid deployment of split modelsSegment
Perform text recognitionPaddleOCR
Language Understanding/Q & A InterfaceLLM
Activate the packerLicense Management

⚙ Example Naming Specification

Named SuffixMeaning
_openvino.viUse OpenVINO Accelerated real-time camera detection
_openvino_imgs.viUse OpenVINO Detect the picture set one by one
nivisionUse NI VISION image acquisition
No SuffixDefault Use opencv_yiku image acquisition module

Operating instructions:

  • DeepLab Series supports batch image detection and camera real-time streaming

  • SAM Support user to click on the target area on the image and split

  • You can set the model path, category file, and device name before running.

Custom model loading method

The sample programs all support users to replace their own models:

  1. Open the sample program VI

  2. Settings:

    • model_file:ONNX Model Path

    • class_names_file: one category name per line .txt File

    • device_name:CPU,GPUWait

✅ Note: - Please avoid Chinese and spaces in the path - The model structure shall conform to LabVIEW inference requirements (recommended Netron Check input dimensions, channel sequence)

Personalized development recommendations

For secondary development on the basis of examples, please refer:

  • Control Settings and Model Replacement Methods

  • Input image type consistency (OpenCV Mat vs NI Image)

  • Use opencv_yiku The drawing, text, window and other tools provided to further beautify the output results

Technical Support

If you encounter problems during operation, please refer: