Increasingly popular among gamers gets ESRGAN technology (Enhanced Super Resolution Generative Adversarial Networks). The latter is a neural network, which have implemented technologies for scaling images with a 2-8-fold increase in quality. How does it work? Algorithm “fed” the original image with low resolution, then it not only increases the native resolution of the latest, but also improves the quality of images, adding realistic details and textures, making images “more natural”. In several passes of the algorithm usually creates an image with even greater accuracy and realism.
Using ESRGAN users have significantly improved texture to games such as Monkey Island, Morrowind, Metroid, Myst, Crusader No Remorse, Final Fantasy 7, Doom, Resident Evil 2 and 3, while retaining their overall style. Just a few days ago came out with a modification with high resolution textures for the cult classic Max Payne. With ESRGAN network was improved texture, and apply manual editing. In total were replaced a few thousand textures, about 95 percent.
The result looks like this:
Read several examples of handling textures using neural networks ESRGAN below:
It
Became
It
Became
It
Became
For those wishing to try out the technology yourself, you can use this guide (it is highly desirable knowledge of English). ESRGAN source code is here.
In the opinion of the regulars of the forums, this technology is the future of game development. Among the developers is not a new practice of production of so-called HD versions of old games. In some cases we are talking about a complete redrawing of new textures, other developers are a bit “trash”, converting the resolution of the existing textures. But in any case, for processing the original textures and permissions takes a lot of time.
Technologies such ESRGAN able to significantly speed up these processes by performing the processing of relevant data and giving the result in just few seconds, which makes them very attractive from a practical point of view. This, apparently, agree Nvidia. Graphics solutions recently became interested in various artificial intelligence technologies. For example, one of the latest developments is the company’s GameWorks technology: Materials & Textures using machine learning and CUDA technology for their graphics cards. If you want to evaluate this technology (now there is a beta test), you first need to register on the website of the company — here it is. Earlier we also reported that it trained it has created the neural network to process 3D graphics, thereby shifting this task to the GPU.
For more details see description of the technology ESRGAN here. To discuss the potential of this technology, we invite you in our Telegram chat.