I suspect it'd be fun to play with a hybrid approach there - use the local on-device capability to detect "scenes of interest", then ship those out to the cloud

I suspect it'd be fun to play with a hybrid approach there - use the local on-device capability to detect "scenes of interest", then ship those out to the cloud service (with significantly more horsepower) to get more accurate results. Possibly, if it works for your use case, you could detect and store "interesting looking stuff" and ship it to the cloud later for analysis if your device only has intermittent internet connection.