Games with pix2pix

The game consist of playing an autonomous car in a city ; you have to don’t hurt other car and go around the city. The difficulty remains on the graphical form that you see, that are transformed from reality

It follows two algortihmics logics :

Image segmentation that consists of the decomposition of the image with color.

It could be done via human or using algorithms. Colors need to be consistent on all the dataset.

Then you have the pix2pix algorithm using GAN technique for image2image translation. It goes from one type of image to another.

In This project it was trained to from segmentation images to “photorealistic” images

I was thinking about this concept of machine—computer vision; what do AI see ?

Thought, I also made primarly to see if it could work in a plausible virtual world, if it worked in a way the player could drive in a reasonable framerate, if the render could actually be playable.

Here is a bit of gameplay :