The image (gif) you see above is from a self-driving-car, taking notes of the surroundings in Grand Theft Auto V first person mode to make itself an actually more proficient system. Yes, you just heard that correctly. Self-driving cars are playing GTA V to get better at not killing people in real life. It may sound silly, but the A.I is figured out (in roughly 49 hours of gameplay) what would have taken years of actual road time.
HAHAHAHAHA MOM AND DAD, I TOLD YOU VIDEO GAMES WOULD SAVE LIVES EVENTUALLY!
Sorry, just needed to get that out of my system. From TheDailyMail:
Using computer vision algorithms, the team labeled thousands of images in just a few hours, which will be used to teach autonomous cars to recognize different objects.
It is the kind of thing that blows your mind in many ways. Basically, what GTA V is doing for self driving cars is to help them with recognition of objects, people, and variables that could potentially cause accidents. It is nothing short of remarkable, and the physics and cityscape of GTA V are perfect so they are utilizing that code to help make the code for self driving cars work better. More on the science of it:
Using computer vision algorithms, the team can label thousands of objects in a few hours, which can be used to teach self-driving object recognition.
To begin their work, researchers extracted 25 thousand frames from Grand Theft Auto V that varied in weather conditions and took place during different times of day.
Then, using computer vision algorithms, the team annotated all of the images in just 49 hours.
For each image taken from the game, the team added a corresponding ID map that identifies the mesh, texture and shade found with a specific pixel.
Then the team broke down each image into patches for pixels that share common characteristics.
To give you an idea of just how far beyond your realm of thinking this goes, here is what you see:
and here is what the “car” sees in same exact situation:
To find out more about this fascinating tech, click here.