Train a Discrete State Detection skill

The inputs required for training a Anomaly Detection and Discrete State Detection are slightly different. For a Discrete State Detection skill, provide good and bad image datasets.

To train a Discrete State Detection skill

1. From the Good card, select Import Data.

2. From the side sheet, select one of the following options:

   • Images – Select images from the local computer. For more information, see Upload images.

   • Videos – Select videos from the local computer. For more information, see Upload videos

   • Live Camera – Select images from a live camera or webcam. For more information, see Upload training images from a live camera or webcam.

3. Repeat steps 1–2 for the Bad card.

4. Change the Label Name on the Good or Bad card through the following steps:

   1. Select the card, and then enter a new name in the Label Name box.

   2. Select to save the new label.

      For example, Open-Shut.

   3. Select to revert to the original label.

      The Label Name box allows a maximum of 32 alphanumeric characters. The special characters allowed are + - _

5. Optional: To delete the existing images, see Remove images associated with a label.

   

6. Select Train AI.

   The skill starts to train and run through 20 epochs. The number of images in the datasets determines the total time of processing. A large dataset takes longer to process.

7. Optional: To cancel the training:

   1. Select Cancel training.

   2. On the Cancel Training dialog, select Confirm.

8. Select Show training progress details to view the training log.

9. After the training is complete, select Choose Camera to connect the camera that monitors in runtime.

   The Cameras side sheet appears.

   Note: For instructions on how to add a camera, see Connect your camera.