University of Maryland students have created a sweater with a unique print that AI cameras can’t see. That is, the camera does not recognize the person who put on such a sweater.
How did the students achieve this result:
- We used the COCO data set, on which the YOLOv2 computer vision algorithm is trained.
- We drove COCO through the detector and revealed the computer’s pattern of recognizing a person.
- We applied the resulting template to the gradient descent algorithm. That is, we created the opposite template.
- Converted the template to an image and placed it on the sweater.
The cameras see not a person but some other object they have not been trained to recognize. True, if the neural network is “trained” to search for a person in such a sweater, he will be remembered.
The students want to create a line of clothes that won’t let the cameras see them with a unique print.
Do you believe in startups? Would you buy this jacket?