Creating a Vision Model


If you want to create a vision-based solution for your use case, then training a computer vision model is what you need. Follow the below steps to begin training your own model that lives locally, on your device.

Computer Vision Model Training

Start by creating a new Canvas. Then open the Elements Drawer and drag either the Object Detection or Classification Trainer element onto the Canvas.

Open the Trainer Element settings and adjust the following settings:

  1. Training data path: Using the Select Directory button, choose the folder where your training data is located.

    For object detection training, all images must be in the same folder and formatted in the COCO JSON format.

  2. Output artifact path: Using the Select Directory button, choose the folder where you would like to save the training artifacts.

  3. Leave all other settings as the default.

  4. You can now hit Run on the Canvas.

The first time this flow runs, it may take a moment due to downloading dependencies.

Once the training is completed, you will have a selectable computer vision inference model.

Computer Vision Model Inference

To use your newly trained inference model, follow these steps:

  1. Create a new Canvas.

  2. Drag the inference element of the type of computer vision model you trained onto the Canvas. This will be either an Object Detection Inference Element or a Classifier Inference Element.

  3. Open the inference element's settings and select either your trained artifact from the dropdown or file selector, not both.

  4. Drag an input element onto the canvas, either Camera Element or Media Loader Element, and connect it to the input side of your inference element.

  5. Drag an Output Preview element onto the canvas and connect it to the output side of your inference element.

If you are using a Camera Element, you must first have an input device configured under the Devices tab.