Originally posted by niklas What knowledge is there about the K3iii and it's Deep Learning Image Recognition Subject Tracking System?
"Deep learning" is probably not what you think it means. Nothing is being learned in the actual camera.
Deep Learning (in basic terms) happens on a powerful computer with an AI algorithm (software code).
They feed the system a huge number of images (say it's people's faces) and they "train" the algorithm to recognize the shapes, patterns, color tones of skin etc with all images of someone facing the camera.
Now they do the same with more people images but this time, the people are facing 45 degrees to the left. They then do it again with more people images facing 45 degrees to the right.
Then they start the process all over again but this time, the people are facing 90 degrees so only the side of their face is in the image. They keep doing this until that software code can start to "learn" and recognize that these are what the focus should be nailing and tracking.
Then they'll feed more images but this time, they'll have the software "learn" that it's eyes it should really focus on. Eventually the software will figure out that if the eye can't be recognized, it will fall back to the facial pattern.
Then they'll feed a bunch of animal images and "train" the software to recognize those as points of focus. Then they'll do a bunch of car images, etc etc.
That's really all deep learning is. But once that data is put into the camera via firmware, the camera itself won't learn anything, but it now has the "smarts" to be able to track more efficiently since the firmware has ALL the information that was taught to it regarding what to recognize and what to do when it recognizes the patterns it knows.