Face recognition Library for automatic cluster

Question

This question is the addition to this one, does those SDks have the automatic cluster function. i.e. which one has ability to group lots of face into several clusters based on recognition results?

Thanks

Answer 1

Most of those do. I have personal experience with the PittPatt one. It works reasonably well (60% or so good clusters, 15% or so big false clusters of several people, 25% or so duplicate clusters of same individual). YMMV based upon input data and parameters.

However, automatic clustering for lots of people is a very hard problem and you may end up spending a ton of time tuning parameters to get it to work.

What application are you trying to target?

Overall, I'd argue that the current state of technology is largely geared towards consumer photo apps. With a little bit of human intervention, you can quickly recognize and tag clusters of photos for a small set (<20 or so) of people (family and friends, etc.) very accurately.

If you're trying to do an application which does automatic clustering of a lot of strangers in a crowd, then software doesn't do all that well. That said, even humans perform rather poorly at such a task.