Humans are still learning much in the fields of Artificial Intelligence (AI), and anything new about what a computer can do to mimic humans can be surprising.
Researchers have created an AI that can render fictional street scenes, only based on human requests. At a glance, it looks like a blurry amateurish dashcam photograph. But actually, the image is solely generated by an imaginative neural network, stitching together its memories of real streets it was trained on.
Qifeng Chen, co-author of the research paper, said that the scientists created the algorithm and trained it using images from more than 3,000 city streets.
Chen's system starts by processing a photo of a real street it has never seen before, but that has been labelled so the AI knows which bits are supposed to be cars, people, roads and so on. The AI then uses this layout as a guide to generate a completely new image.
The result, is images of realistic rendering of cars, streets, and buildings that don’t actually exist.
The advancement of AI can help humans in doing a lot of things. In this case, AI can help developers in creating virtual worlds, like in games. Instead of spending hours to create artificial cities, developers may become directors in the future, using voice commands to make computers to create all those things.
However, the way in which this algorithm works presents its fictional worlds to look like hand-painted and blurry. Not sharp and crisp like traditional CGI. But this makes it feel more 'real'. Although with the dreamlike quality, the softness of colors, instead of powerful contrasts, for example, makes the image feels like a real thing.
There are still challenges and limitations in algorithms as the network isn't yet able to fill in all the details expected in photos. But the results speak for themselves.
The ultimate goal, according to the report, is to create photo-realistic images. Chen says "Using deep learning to render video games could be the future." He said that he had already experimented with using the algorithm to replace the game world in Grand Theft Auto V.
The algorithm was also trained and tested on a smaller database of photos of domestic interiors.