NVIDIA dedicates the This week’s edition from su series AI Decoded a the very direct relationship between the RTX platform and the quality of the videos we watch. A very interesting topic, which we will discuss below, but which I have also found impossible not to relate to a very interesting publication by my colleague Isidro, on the reaction of users to the proliferation of processors and SoCs that integrate an NPU. A reaction that, as you can read in said article, is not very positive at the moment.
I say that it is impossible for me not to establish a relationship because, of course, in that text, and it is made explicit in it, it is said that there is little interest, at the moment, in integrated NPUs, as a new paradigm of local artificial intelligence. However, the situation is very different if we talk about the interest generated by GPUs as great accelerators of local AI, something that is fully accredited with the weight in the market of NVIDIA RTX graphics adaptersand the large number of AI-based functions that we have been running locally for years thanks to them.
Typically, when talking about artificial intelligence and the NVIDIA RTX platform, the first thing that comes to mind is the set of technologies grouped under DLSS (intelligent upscaling, frame generation, and ray reconstruction), there are many other functions that also rely on the specialization of the Tensor cores, and Video scaling is one of the most importantas it substantially reduces the impact of peak bandwidth on video quality.
As we are reminded in this installment of AI Decoded, upscaling is a technique that has been used for a long time, but the problem is that, as a rule, the methods used for this increase in image size are somewhat, let’s say, rudimentary, so the final result tends to leave a lot to be desired. And it is at this point, as you may have guessed, that Artificial intelligence results in a huge improvement with respect to the previous methods.
The method used by the NVIDIA RTX platform is to Take each frame and perform two actions with itOn the one hand, it performs a bicubic rescaling up to the desired resolution (from 1,080p to 4K in the example image), and on the other, it will analyze the image and generate the necessary elements to enhance the contrast. And to this we must add that it also analyzes the motion vectors in order to be able to generate non-existent information that is consistent with the original image.
Additionally, NVIDIA technology It also analyses the received video signal in search of possible defects in it.. So if, for example, we receive a signal in which some type of artifact has crept in due to a problem in the transmission or in the transmitter, the software will be able to identify it and discard it, instead of treating it as if it were something correct, which would result in its rescaling.
Source: www.muycomputer.com
