Deploying to OneVision

Once you have a satisfactory defect detection model, it's time to deploy it to your production environment.

Step 1: Compare Model Versions (If You Have Multiple)

If you've created multiple versions of your defect detection model:

1. Use the version selector to switch between versions
2. Compare their performance metrics:
   • Overall detection accuracy (mAP)
   • Per-defect detection rates
   • False positive rates
   • Inference speed (if available)
3. Test each version with the same set of images
4. Choose the version that best balances accuracy and performance for your needs

Comparing different versions of your defect detection model

Step 2: Export Your Defect Detection Model

Export your best-performing defect detection model:

1. Select your preferred version from the version dropdown
2. Click the Download button in the top toolbar
3. Choose the appropriate export format based on your deployment target:
   • OneVision Package: For deploying to your OneVision quality control system
   • ONNX: For cross-platform compatibility
   • TensorFlow SavedModel: For TensorFlow-based environments
   • PyTorch Model: For PyTorch-based applications
4. Configure any export-specific settings:
   • For defect detection, consider enabling quantization for faster inference if accuracy isn't compromised
   • Set confidence thresholds appropriate for your quality control requirements
5. Download the exported model package

Exporting your defect detection model

Step 3: Deploy to Your Quality Control System

Option 1: Deploy to OneVision

If you're using OneVision for quality control:

1. Select "Deploy to OneVision" instead of download
2. Choose your target OneVision instance
3. Configure how defects should be handled:
   • Alert thresholds for different defect types
   • Integration with your production line systems
   • Visualization settings for operators
4. Start the deployment
5. Verify that the model is working correctly in your OneVision environment

Option 2: Custom Integration

For other quality control systems:

1. Export the model in your preferred format
2. Integrate the model with your quality inspection software:
   • Set up proper image preprocessing to match training conditions
   • Configure detection thresholds for each defect type
   • Implement appropriate alert mechanisms for detected defects
3. Create visualization tools to help operators understand model detections
4. Set up logging to track detected defects for quality control metrics

Important for Defect Detection: Always implement appropriate confidence thresholds based on your quality control requirements. Lower thresholds catch more defects but may increase false positives.

Deploying your defect detection model to a production environment

Step 4: Validate in Production

Before fully relying on your automated defect detection:

1. Run a validation period where both the model and human inspectors check products
2. Track key metrics:
   • Detection rate: Percentage of actual defects caught by the model
   • False positive rate: Non-defects incorrectly flagged
   • Processing time: How quickly the model analyzes each product
3. Adjust confidence thresholds based on validation results
4. Create a feedback mechanism where inspectors can flag incorrect detections
5. Use this feedback to collect data for future model improvements

Validating model performance in the production environment