ML Kit vs Cloud Detection?

ML Kit is all about on device. So if you use ML Kit, none of that data goes to the server side. It's also perfectly free. The Firebase API offer 10,000 labels, whereas the on-device API offer 400 labels. And in those cases, I would do basically what Matage just suggests, which is to do maybe a bit of both so that you don't send every single frame of what the cameras see over to the cloud. Instead, have the object detection and tracking API doing the initial shifting and basically drawing rectangle on the screen, telling the user, hey, is this the image that you want to detect? And then you send it over to the