Friday, February 14, 2020

How to fool Google's image-recognition algorithm

Google's image-recognition algorithm does a pretty good job of recognizing objects—as long as they're positioned in their normal orientation relative to the background. But these neural networks fail if the objects are positioned upside down or rotated even slightly.
"A fire truck, for example, seen from head on, could be correctly recognized. But once pointed up in the air, and turned around several times, the same fire truck is mis-classified by the neural net as a school bus or a fireboat or a bobsled. The upshot is that the state of the art in image recognition is 'naive,' and some greater understanding of three-dimensional structures seems needed to help them get better. State of the art neural networks such as Google's Inception are good at 'classifying' things in pictures, they conclude, but they are not really recognizing objects, in the true sense of that expression."
Read more: ZD Net: Google's image recognition AI fooled by new tricks

2 comments:

Tina said...

Just goes to show you that although computers are supposed to be smart, they will never replace the discerning eye. Remember that when you get into a self-driving car! Happy Valentine's Day to James and Family!

Eugene Arenhaus said...

These neural networks cannot distinguish the object from the background. They are responding to the images in the training set on the whole.