Validated models
After a model is trained, the status of the model changes to Trained. The model manager can validate the model version and either accept or reject the trained model. The model manager validates the model by using validation image sets. After a model is validated, it can be used and deployed.
You can validate a model version that has a status of Trained. A trained or retrained model version can trigger the validation process. When you validate a report, you must manage the image sets. Every image group must have at least one validation image set to validate the model. You must use different image sets and training image sets to validate the model version. To begin the validation process, select Validate.
- Classification model report
- In the classification model report, one image has one defect at most. The confusion matrix is used to generate the report where each column represents one real image group type in the validation data sets. Each row represents the predicted image group type. The last row in the chart represents the aggregate results.
- Object detection model report
- In the object detection model report, one image has multiple defects. In this report, the mean average precision and the recall are calculated to indicate the model accuracy.
To view the report, select Show Report. From the Validation Report window, you can revalidate, reject, or accept and deploy the model.
If the first model version is rejected, you can change the training parameters and then retrain the model. If a model version that is not the first version is rejected, you can retrain a new model version by using new image sets.