Skip to main content

Classifier Not Running

This page helps diagnose and resolve issues where the OV80i’s classification model fails to activate or return results during inspection.

💡 Quick Tip: In classification mode, once a trigger is received (via hardware or continuous mode), the image is automatically captured, and the result is displayed on the HMI — if the classifier is working correctly.

Common Symptoms​

  • Image is captured, but no classification result appears on the HMI
  • Classifier tab shows “Not Trained” or “Needs Training”
  • UI stuck in “Training…” or classifier never finishes
  • Results area on HMI stays blank or shows “No classification”
  • Only images appear, but no labels, colors, or class scores

Step-by-Step Troubleshooting​

1. Confirm Recipe Is Active​

  • Navigate to All Recipes
  • Ensure your classification recipe is marked as 🟢 Active
  • Only one recipe can be active at a time

2. Verify Classifier Setup​

  • Open the Recipe Editor → Regions of Interest (ROIs)
  • Confirm that:
    • At least one ROI is defined
    • The ROI has a classifier block added
    • No errors or warnings appear in the ROI config

3. Check Training Data​

  • Go to the Classifier tab for the ROI
  • Confirm:
    • Images have been uploaded and labeled for at least two classes
    • The Train button has been clicked and completed
    • Status reads “Trained”, not “Needs Training” or “Not Ready”
  • If trained in Fast Mode, accuracy may be low; retrain in Accurate Mode for production

4. Run a Manual Test Capture​

  1. Use continuous mode or send a hardware trigger
  2. Confirm:
    • Image is captured
    • Classifier result appears (pass/fail, good/bad, class name)

If there's no output after capture → the model may not be trained or failed to load.

5. Restart the Camera​

If everything seems configured correctly but the classifier still doesn’t respond:

  1. Power cycle the camera via switch or disconnect/reconnect the power
  2. Wait 20–30 seconds for boot to complete
  3. Open the HMI and test with a new trigger

Restarting the camera reloads the model and classifier runtime from scratch.

6. Review Logs​

  • Visit logs.overview.ai
  • Upload the log bundle for internal analysis
  • Use this to verify classifier startup status and runtime behavior

Classifier Architecture Reference​

Classification Models: “Cats vs. Dogs” Approach​

  • Best when each ROI/image has a single, discrete class
  • Use Fast Mode for testing, Accurate Mode for production
  • Common use cases:
    • Loose bolt detection
    • Shaft alignment
    • Grease presence
    • Radiator pin condition

Use classification when you're answering: “Which of these options does this look like?”

Segmentation Models: “Where’s Waldo?” Approach​

  • Use when:
    • Defects vary in size and shape
    • You need to locate exact defect regions
    • Speed is critical (faster cycle time)
  • Common use cases:
    • Surface scratches
    • Foam coverage
    • Gap detection

Use segmentation when you’re answering: “Where is the problem?”

Final Checklist​

ItemRequired for Classifier to Run
Active recipe selected✅
At least one ROI defined✅
ROI includes classifier block✅
Trained model status: “Trained”✅
Capture triggers working✅

🔗 See Also​

  • [How-To: Create a Classification Recipe]
  • [How-To: Train AI Models]
  • [Troubleshooting: No Trigger or Output]
  • [Digital I/O Logic: NPN vs PNP]
  • [Concepts: Classification vs Segmentation]
  • [Logs & USB Debug Access]