Can be a bit misleading comparing this CNN based facial recognition platform with the SNN based BrainChip Studio and Hardware Accelerator. The following article by Bob Beachler (it's been posted before) is an interesting read in regard to the problems of large datasets involved in CNN based deep neural networks.
https://www.gsnmagazine.com/node/48817
The Yitu numbers are impressive nevertheless. Presume these numbers (3secs per 1.8b faces) are results based on having a huge dataset of photographic licenses. Numbers quoted for SNvision (back in the days of Spikenet) was 100m images could be processed in 1 sec. I think that was the speed from processing a similar photographic database. That was some years ago. Note the Yitu system is good at putting a name to a face but remember it's no good if your searching for a suspect in a crowd and the only identifying feature is a particular shirt pattern or a tatoo.
Add to My Watchlist
What is My Watchlist?
