Skip to main content

Labeling & Retraining

The predictions of your model become more accurate the more real, labeled production data it is trained on. navio VISION closes this loop: label the images collected via your live data connection and retrain the model with a single click.

Labeling Images​

  1. Open the use case and go to the Live Data Connection tab.
  2. Click Start Labeling. The labeling view opens, showing the collected images together with the model's prediction for each image.
  3. For each image, confirm or correct the label (Normal or Anomaly).

Your labels are stored automatically — there is no separate save or export step.

info

Labeling is powered by the navio platform's data labeling integration (an embedded Label Studio). Access to the labeling view uses your navio VISION login.

Retraining the Model​

Once you have labeled a meaningful set of images, click Retrain Model on the Live Data Connection tab. Retraining uses all labeled images and can take up to an hour — you can check back later; the improved model is deployed automatically when done.

note

Retraining is not available for demo use cases, and only one retraining can be in progress at a time.

After retraining, compare the results in the Monitoring view to see how the model performance developed.