Patents are always a pretty intriguing thing to track down in the tech industry; sometimes they represent ideas and products that will never see the light of day, but there are times when they provide clues for future versions. A newly released US Patent Application it could be the latter, as it addresses technology that could allow Nintendo to enhance visuals on its hardware.
Launched on March 25, 2020 and publicly released yesterday (September 30), the app is titled ‘Systems and Methods for Image Conversion by Machine Learning’, and while the initial language may be a scratch, it’s essentially a similar idea. to NVIDIA DLSS. It is short for Deep Learning Super Sampling in NVIDIA’s case, which works on some of its GPUs to improve image resolution. and quality in real time, while being remarkably efficient and ensuring that the graphics card is under less strain. It’s impressive technology, and it’s been front and center in many conversations about how Nintendo could produce a new Switch-style handheld that generates higher resolution images while still running at low power output. Digital Foundry has explored this in detail.
What makes this app intriguing is that Nintendo is clearly exploring this internally – one of the named parties in the app is Alexandre Delattre, who is a co-founder of Nintendo European Research and Development. It is also recognized in the ‘Introduction’ of the patent that this is an area that is being explored throughout the industry:
Machine learning can give computers the ability to “learn” a specific task without expressly programming the computer for that task. One type of machine learning system is called convolutional neural networks (CNN), a class of deep learning neural networks. These networks (and other forms of machine learning) can be used to, for example, help automatically recognize if there is a cat in a photo. Learning takes place by using thousands or millions of photos to “train” the model to recognize when there is a cat in a photo. While this can be a powerful tool, the processing resulting from using a trained model (and training the model) can still be computationally expensive when implemented in a real-time environment.
Image upconversion is a technique that allows the conversion of images produced in a first resolution (eg, 540p or 960 × 540 resolution with 0.5 megapixels) to a higher resolution (eg, 1080p resolution , 1920 × 1080, with 2.1 megapixels). This process can be used to display first resolution images on a higher resolution screen. Thus, for example, a 540p image can be displayed on a 1080p television and (depending on the nature of the upconversion process) it can be displayed with higher graphical fidelity compared to if the 540p image were displayed directly with the traditional one (for eg linear) on a 540 television. Different techniques for image upconversion can trade off speed (eg, how long the process takes to convert a given image) and the quality of the converted image. For example, if an upconverting process is performed in real time (for example, during a video game), the image quality of the resulting upconverted image may be affected.
Accordingly, it will be appreciated that new and improved techniques, systems and processes are continually sought after in these areas of technology.
Ultimately, it should come as no surprise that Nintendo is looking into scaling through machine learning, as it is likely to be a vital factor if the company chooses to retain a Switch-style form factor while offering higher graphical fidelity in the future. It is also interesting if Nintendo will continue to use NVIDIA technology in future devices; If you develop your own solution, you may not need NVIDIA DSSL tools. Of course, depending on what and who you believe in, there are reports that ‘4K’ development units are now available.
Let us know what you think in the comments!