Was curious how the API communicates with traied model data.
I have trained my own detection model and am able to use an OAK-D camera to detect the object it has been trained for.
But, how does my python script, using the sample code for the luxonis oak, use the API to reference the model?
Does it require a network connection? If not, which I suspect it doesn’t because my script works without internet connection, how does the API connect the python script to the model?
What are the pros and cons of using the API rather than downloading and training the model using one of the google colab guides?
The pros of using the python package we provide are that you don’t have to fuss with converting weights and maintaining code dependencies for OpenVINO updates, but instead just add the correct model and API keys, and you’re on your way to deploy. Updating a model is as simple as changing the version number.
The model weights are cached to the device for running inference on-device once you begin running your script. The comments in the code here show the response object format for the predictions: Luxonis OAK (On Device) - Roboflow
You’ll need a network connection just to access the weights with the python package, and then again only after the device is powered off.