Cookie Consent by PrivacyPolicies.com
Quality Automation with AI and Relimetrics

Trained Model Prediction

Learn how to test previously trained models

A previously trained model can be tested/verified by applying the model on a different image set in order to verify how well the model performs. 

✓ Step 1:  Click on Start prediction button
Figure 18 - Training Screen - Prediction Tab

In the training screen, click on the Start prediction button in order to start the prediction configuration process. A pop-up window will be shown.

✓ Step 2:  Select Backend
Figure 19 - Training Screen - Prediction Tab

User can select the backend to run the prediction. The task will be added to the queue and if user selects the cloud backend which has a queued tasks. 

✓ Step 2:  Select Model Use

This step, the user can choose the type of the model to perform the evaluation on. It can be suitable for Detection, Classification, Semantic Segmentation or Instance Segmentation then click on the next button to move to the next step. For this version, only the instance segmentation type is available.

✓ Step 3:  Set Base Model

A model can be chosen among the existing ones.

✓ Step 4:  Select Image Set 

An image set can be selected for the prediction process.

✓ Step 5: Prediction Summary
Figure 20 - Prediction Popup - Summary

At the end of the configuration, the user can visualize a summary of the prediction (Fig.20). If something is not correctly set up, there is the option to click on the previous button to go back to the previous steps and apply changes. Once everything is correctly set, click on the Start Prediction button to start the prediction process. 

Prediction may take a while and the user can stop the process by clicking the Stop Prediction button. (Fig.21)

Figure 21 - Prediction In Progress

✓ Step 6:  Save the predictions as Annotations

Predictions are saved as annotations automatically to use them in future model trainings. User can change the Prediction Threshold value and filter the predictions. (Fig.22) 

Figure 22- Training Screen - Prediction is completed

✓ Step 7:  Filter Predictions

When the user reaches the intended value by changing the Prediction Threshold, press Delete Filtered Predictions and delete the rest of them. Remaining predictions will be saved to Annotations automatically.  (Fig.23)

Figure 23- Training Screen - Delete Filtered Predictions

✓ Step 8:  Predictions as ROI List 

The user can check out and control their predictions on the Annotation Screen. They organize them as they are doing annotations. Predictions are separated from the rest of the annotations with the Model Label.

(Fig.24)

Figure 24- Annotation Screen - Predictions as Annotations

Request Pilot