Skill workflow

You can use AVEVA Vision AI Assistant in a combination of workflows to achieve the best results. The following diagram represents the workflow and functionality available for AVEVA Vision AI Assistant.

A basic workflow is as follows

1. Select a skill type and create a skill:

   1. Select the skill type.

   2. Provide a name and description and create a skill.

   For more information, see AVEVA Vision AI Assistant skill types and Create a skill.

2. Provide training dataset(s) and train the skill:

   1. You can directly upload images or upload a video from which images can be extracted.

   2. You can snapshot a live camera and use those images as a training dataset.

   3. Use these images and train the skill.

   For more information, see Manage training datasets, Train an Anomaly Detection skill, and Train a Discrete State Detection skill.

3. Provide a monitoring source (camera).

   After the skill is trained, connect to a camera. You can choose between a network camera, webcam, dataset, or connecting to a camera programmatically. For more information, see Connect your camera.

4. Preview the skill and review the classified images:

   1. After providing a camera source, you can optionally preview the skill.

   2. In Preview, you can view the skill classification. A skill runs only if it is deployed or the preview mode is selected.

   3. On the Review page, you can review the skill and provide feedback for the classified images.

   4. You can compare the current (deployed) skill to the retrained skill using metrics like the confusion matrix and accuracy score. Depending on the metrics, you can choose to keep the current deployed skill or deploy the retrained skill. For more information, see Review a skill.

   You cannot compare the trained version of a skill to the retrained version. The retrained version replaces the trained version. You can compare the deployed version of a skill to a retrained version of the skill.

   Both deployed and trained skills can be retrained with new feedback in a loop until the desired results are obtained. A retrained skill can be deployed, where it replaces the existing deployed skill.

5. When you are ready to use the skill in a real-time production environment, deploy the skill. For more information, Deploy or undeploy a skill.