Hi! Prodigy itself doesn't ship with a built-in recipe for training a computer vision model. There are many different implementations and frameworks, so you'd have to pick the type of model you want to use. How the training works in detail depends on the framework (PyTorch, TensorFlow etc.) and model architecture/implementation.
You can see an example of the data format produced by Prodigy's image_manual recipe here: https://prodi.gy/docs/api-interfaces#image_manual This should make it easy to update the model with what it needs: for each annotated bounding box, you'll have the coordinates, width, height and label. You can use the db-out command to export your dataset to JSON, and then use the data to train your model.