Informative article, as always. Presumably this camera's analytics performance should improve over time. Does it retrain itself? Does the manufacturer set expectations on rate of improvement, time to steady state, etc.? This AI approach makes a lot of sense and is probably the way all camera analytics will go in time.
The learning on most "deep learning" cameras comes from a pre-trained neural network, which is done in development, and updated in the camera(s) through firmware updates.
I would agree with you that AI is likely to replace "standard" analytics over time. It's getting more and more common, and there are a number of options with decent performance at this price point. Is it on the level of top-performing analytics? Not yet, but compared to standard VMD or intrusion analytics, a huge improvement.
after looking at so many cameras and AI, analytics and the shortcomings of their algorithms. have the manufacturers ever reached out to you guys (IPVM) to get advise and even ask you to assist with testing updated firmware or refined algorithms on these and other cameras? or do they just not care there is so many false positives.
Faheem, usually manufacturers give us feedback and updated firmware. It's almost become a joke inside of IPVM that each time we tell manufacturers our preliminary results show problems, they respond with some variant of (I am only slightly being facetious) "Oh, no, no, we'll get the good firmware now."
LOL, ya well thats what happens when you give excellent information, advise and testing services for free to them. you need to go to them and bundle it with a heft price tag then they would have taken you serious.
i have the exact same issue with uniview. no matter how many examples and test i do on their alphaview ptz to show them the inefficiencies of the algorithm and try to get them to fix it, not only for my benefit but for theirs as well. we go no no where.