Real-Time Vision — Edge Deployment

Edge inference is the discipline of getting a 90-accuracy model to run at 30 fps on a device with 2 GB of RAM. Every percentage point of accuracy is traded against milliseconds of latency.