I’m almost done with the introductory AI course, with three classes left. Also almost done with Chapter 10 of The Seventh Law of Power, though the two are not related, other than serving as evidence I can still multitask. Marta has just had to deal with a suspiciously inept assassin and I’ve been deep in the weeds with Mediapipe and Tensorflow.
We did a series on facial recognition early in the course, with pictures of known politicians (posed) for training data and more candid shots for the recognition bit. The software was amazingly accurate. In the last few lessons we’ve moved more into facial and gesture recognition in the sense of spotting what is and isn’t a face on a more fundamental level than simple ID, and recognizing gestures.
In a recent lesson, the goal was to make the program recognize a face on camera, and draw key index points on that face, using our own as the target. While we were looking at the raw data, one of the first parameters to show up is how confident the program is that “this is a human face” with the probability given as some fraction of one. Our instructor was fussing that the routine never gave his own face a probability higher than .93, or 93% confidence.
Mine was never more confident than 75%.
When it was time to identify gestures, you can guess which one I chose.