Intel is delivering surprisingly successful results with Photorealism Enhancement Technology under development.
Intel announced its new machine learning project called Photorealism Enhancement. The company revealed its technology with Grand Theft Auto V.
Passing the game through the processes created by researchers Stephan R. Richter, Hassan Abu Alhaija and Vladlen Kolten yields a surprising result. Intel researchers suggest that some of photorealism comes from the datasets that feed the neural networks. The group also offers a more in-depth and comprehensive explanation of how image enhancement actually works. Researchers at Intel Labs used the City View Dataset, which consists of images of a German city city center captured with a car-built camera for training.
When a different artistic style is applied to shots using machine learning techniques, the results are often temporarily unstable. While strange objects that jump around, appear and reappear frame by frame can be disturbing, it can also become controversial how real the results look. The rendered effects in this new approach do not show any of these objects. Because in addition to processing images generated by the game engine of Grand Theft Auto V, the neural network also uses other processed data such as the depth the game engine reaches.
With the new technique, the road surface is flattened, while the accents on the vehicles are more pronounced, and the vegetation on the hills looks thicker and more vibrant in several clips. What’s even more impressive is that the researchers think that with the right hardware and more optimization, the game footage could be improved further. In this way, it is suggested that software updates may be sufficient for the games to look much better.
On the other hand, we recommend that you take a look at our GTA Online skip-the-line guide.