Person Detection Update: A New Experiment for Premium Features

[sodcam]

This is an old post from back in May, but I thought it shed light on the whole edge vs cloud AI discussion

@WyzeTao

Since we first launched v2 and Pan about 2 years ago, we kept adding a lot of features on top of it, detection zone, Alexa integration, Google Home integration, Edge AI (removed later), CMC, etc. The new features added a lot of complexity on the code and put quite some pressure on camera CPU and memory. I remember the average CPU load was 90%+ when Xnor AI algorithm was running. Then CMC was added which pushed the firmware to its limit. All of above impacted our camera stability and give us very little room to improve.

Several months ago we decided for a re-write of our firmware code. More than 60% of the code was re-written. The whole process consisted multiple months with cleaner architectural design and better performance design. Our test showed the CPU usage dropped 20%-25% in running time which is substantial. All the change will give us better extensibility and stability in long run. The saved CPU and memory can be used for future Edge AI related work that we couldn’t add before. The change was for the good future of Wyze cameras.