It uses machine learning and is trained in several scaling technologies.
If you’re interested in technical analysis of console games, you’ve probably already heard of pixel counting. While on PC you know exactly the resolution in games and have full control over it (that is, unless you use dynamic resolution), on consoles it is pre-set by the developers and not everyone is willing to reveal the values. in mode descriptions.
Editors at Digital Foundry and others use pixel counting to measure resolution, meaning they literally count pixels so they know the resolution at which the game is initially rendered, but not the output resolution.
This method has existed with us for a very long time and during this time it has not changed at all, quite the contrary. As anti-aliasing has improved and different scaling technologies have been introduced, counting pixels has become significantly more complex and getting the count right has become more difficult than ever. Labor-intensive manual work may soon end.
The Pixel-Count ML application was recently introduced. It promises complete automation of the entire process, and it is not just a simple algorithm, but a complex model using machine learning. He is trained in several upscaling technologies (currently FSR, TSR and TAAU) and is gradually improving.
Below you will find screenshots directly from the application, which were provided to me by the creators themselves. These are real screenshots, not just mockups. Pixel-Count ML is almost ready for release and should be coming to Android and iOS soon.
Source: Pixel-Count ML.
Source: Pixel-Count ML.
Source: Pixel-Count ML.
Source: Pixel-Count ML.
Source: Pixel-Count ML.
Source: Pixel-Count ML.
Source: Pixel-Count ML.
Source: Pixel-Count ML.
Source: Pixel-Count ML.
What does the application promise? It will offer players the opportunity to upload a screenshot from a given game (it must be in 4K resolution and must be losslessly compressed, i.e. ideally in PNG format), and then analyze it either using the PixelNative model (to measure native resolution), or PixelFusion. , if the game uses scaling, it will measure the input and output resolution.
AJ from the AJ Gaming YouTube channel developed the model that the application will use. He recorded videos on it comparing different console versions of games, and for a long time used labor-intensive manual pixel counting. He gradually partially automated and simplified the task, but this was not enough for him, so late last year he began creating an ML model that will soon be presented to the public.
In conclusion, I would like to thank the creators who provided me with preliminary access to the portal using the mentioned model to measure resolution not only from screenshots, but also from videos. You’ll soon see the first results in other technical analyzes of PS5 Pro games.
If you want to follow updates about the app, I will direct you to the official Twitter account.
Source :Indian TV
