This week we're looking at a video displaying the utility, efficiency, and almost always accurate classifications of a "thing classifier". The first thing you noticed is how fast the classifier recognizes things, especially "persons" considering how drastic someone can look depending on their hair is done, their body shape, or how they are dressed. It's a remarkable feat of technology. The fact that they recognize people so fast and the visualization of their classifications in the video is starting to make me think that it's completely plausible when Facebook recognizes your friend completely accurately, who is out of focus and 20ft away, and suggests tagging them in a photo.
To the left we have an example of the "thing classifier" working great in a busy frame outlining them almost perfectly. We can see all the different angles a person can take, their skin colors, their clothing, etc. that the classifier has no problem recognizing. The two kids on the bottom right I didn't even notice myself until I scanned all the boxes. The algorithm recognizes people and shapes better than me sometimes! In the picture you can see too that it recognized a cup vs a bowl. Being able to distinguish that in this picture that is absolutely surrounded by bowls is impressive on its own as well.
Even with all the good things I have said, there's bound to be some flaws. Sometimes colors and lighting will more closely display a resemblance to something else, just as some natural camouflage animals might take advantage of, will mess with the recognizer and have them output false positives. To the left and right we see a hawk absolutely be falsely recognized as a dog and a person respectfully.
The applications of this software can easily go for a more ethical approach or a corrupt approach.